Advertisement

Non-renormalized solutions to the continuity equation

  • Stefano Modena
  • László SzékelyhidiJr.Email author
Open Access
Article
  • 25 Downloads

Abstract

We show that there are continuous, \(W^{1,p}\) (\(p<d-1\)), incompressible vector fields for which uniqueness of solutions to the continuity equation fails.

Mathematics Subject Classification

Primary 35A02 Secondary 35F10 

1 Introduction

In this paper we consider the continuity equation
$$\begin{aligned} \begin{aligned} \partial _t \rho + \text {div }(\rho u)&= 0, \\ \text {div }u&= 0, \end{aligned} \end{aligned}$$
(1)
in a d-dimensional periodic domain, \(d \ge 3\), for a time-dependent incompressible vector field \(u : [0,1] \times {\mathbb {T}}^d \rightarrow {\mathbb {R}}^d\) and an unknown density \(\rho : [0,1] \times {\mathbb {T}}^d \rightarrow {\mathbb {R}}\). Here and in the sequel \({\mathbb {T}}^d = {\mathbb {R}}^d/{\mathbb {Z}}^d\) is the d-dimensional flat torus. We will also always assume, without loss of generality, that the time interval is [0, 1]. We prove in these notes the following theorem.

Theorem 1.1

(Non-uniqueness for Sobolev and continuous vector fields). Let \(\varepsilon >0\), \({{\bar{\rho }}} \in C^\infty ({\mathbb {T}}^d)\) with \(\int _{{\mathbb {T}}^d}{{\bar{\rho }}}\,dx=0\). Then there exist
$$\begin{aligned} \rho \in C\big ([0,1]; L^1({\mathbb {T}}^d)\big ), \quad u \in C([0,1] \times {\mathbb {T}}^d) \cap \bigcap _{1 \le p < d-1} C\big ([0,1]; W^{1,p} ({\mathbb {T}}^d)\big ) \end{aligned}$$
such that \((\rho ,u)\) is a weak solution to (1) and \(\rho (0) \equiv 0\) at \(t=0\), \(\rho (1) \equiv {{\bar{\rho }}}\) at \(t=1\) and
$$\begin{aligned} \max _{(t,x) \in [0,1] \times {\mathbb {T}}^d} |u(t,x)| \le \varepsilon . \end{aligned}$$
(2)

By weak solution we mean solution in the sense of distributions.

We have chosen the periodic setting for simplicity and to emphasize that the key phenomenon we are studying is the effect of (local) low regularity. As with other applications of convex integration (e.g. [9, 14]), we expect that analogous statements hold also in the full space \({\mathbb {R}}^d\).

It is well known that the theory of classical solutions to (1) is closely connected to the ordinary differential equation
$$\begin{aligned} \begin{aligned} \partial _t X(t,x)&= u(t,X(t,x)), \\ X(0,x)&= x, \end{aligned} \end{aligned}$$
(3)
via the formula \(\rho (t) {\mathcal {L}}^d = X(t)_\sharp (\rho (0){\mathcal {L}}^d)\) or, equivalently, due to the incompressibility,
$$\begin{aligned} \rho (t, X(t,x))=\rho (0,x). \end{aligned}$$
(4)
In particular, for Lipschitz vector fields u the well-posedness theory for (1) follows from the Cauchy–Lipschitz theory for ordinary differential equations applied to (3).
It is in general of great interest to investigate the existence and uniqueness of weak solutions to the Cauchy problem for (1) in the case of non-smooth vector fields, and the connection to the Lagrangian problem (3), (4). The general question can be formulated as follows. Fix an exponent \(r \in [1,\infty ]\), denote by \(r'\) its dual Hölder, \(1/r + 1/r' = 1\), and assume that a vector field
$$\begin{aligned} u \in L^1 \big (0,1; L^r({\mathbb {T}}^d) \big ) \end{aligned}$$
(5)
is given. What can be said about existence and uniqueness of weak solutions in the class of densities
$$\begin{aligned} \rho \in L^\infty \big (0,1; L^{r'}({\mathbb {T}}^d) \big ) \ ? \end{aligned}$$
(6)
The choice of class (6) for \(\rho \) is motivated by the fact that for classical solutions to (1) every spatial \(L^{r'}\) norm is preserved in time. Once (6) is fixed, the choice of the class (5) for u is natural, since in this way \(\rho u \in L^1((0,1) \times {\mathbb {T}}^d)\) and thus the notion of distributional solution to (1) is well defined.
While existence of weak solutions can be easily shown under the assumptions (5), (6), the uniqueness question is much harder. In 1989 DiPerna and Lions [11] proved that uniqueness holds in the class (6) if
$$\begin{aligned} Du \in L^1 \big ( 0,1; L^{r}({\mathbb {T}}^d) \big ) \end{aligned}$$
(7)
i.e. if u enjoys Sobolev regularity with exponent r. Moreover, in this case, the incompressibility assumption can be relaxed to \(\text {div }u \in L^\infty \). In the class of bounded densities the uniqueness result was later extended by Ambrosio [1] in 2004, for fields \(u \in L^1(0,1; BV)\) with \(\text {div }u \in L^\infty \) and very recently by Bianchini and Bonicatto [3] in 2017 in the case of BV nearly incompressible vector fields.

In all of these results an important additional feature is the connection to a suitable extension to (3), i.e. the link between the Eulerian and the Lagrangian picture. More precisely, under assumption (7), there exists a unique distributional solution to (3) for a.e. x, such that \(x \mapsto X(t,x)\) is measure preserving for all t (assuming \(\text {div }u = 0\)): such flow map is called regular Lagrangian flow (see [2] for a general discussion). Then the unique solution to the continuity or the transport equation is given by (4), as in the smooth case.

On the other side, several non-uniqueness counterexamples are known, but they mainly concern the case when the field is “very far” from being incompressible (e.g. \(\text {div }u \notin L^\infty \), see [11]) or the case when no bounds on one full derivative of u are available (see, for instance, the counterexample in [11] for a field \(u \in L^1 (0,1; W^{s,1})\) for every \(s<1\), but \(u \notin L^1(0,1; W^{1,1})\) or the counterexample in [10] for a field \(u \in L^1(\varepsilon , 1; BV)\) for every \(\varepsilon >0\), but \(u \notin L^1(0,1; BV)\)). In all these counterexamples, however, non-uniqueness for the PDE (1) is a consequence of a Lagrangian non-uniqueness for the associated ODE (3). We refer to [16] and to [2] for a more detailed discussion.

Very recently, we proved in [16] the analog of Theorem 1.1, for fields and densities in the class
$$\begin{aligned} \rho \in C\big ([0,1]; L^{r'}({\mathbb {T}}^d)\big ), \quad u \in C \big ( [0,1]; L^r({\mathbb {T}}^d) \big ) \cap C\big ([0,1]; W^{1,p} ({\mathbb {T}}^d)\big ), \end{aligned}$$
with
$$\begin{aligned} r \in (1,\infty ), \quad p \in [1,\infty ), \end{aligned}$$
(8)
and
$$\begin{aligned} \frac{1}{r'} + \frac{1}{p} > 1 + \frac{1}{d-1}. \end{aligned}$$
(9)
The result in [16] shows that uniqueness can fail even for incompressible, Sobolev vector fields (i.e. fields for which the Lagrangian problem (3) is well posed, in the sense of the regular Lagrangian flow), if the integrability exponent p of Du is much lower than the one provided in (7) by DiPerna and Lions’ theory, as specified in (9).

The end-point \(r=\infty \), corresponding in (9) to \(p < d-1\), is excluded in [16]. The main result of this notes, namely Theorem 1.1, shows that such end-point case can indeed be reached and, in addition, quite surprisingly, the vector field produced by Theorem 1.1 is continuous in time and space, not only bounded.

We postpone to Sect. 2 a technical discussion about why the case \(r=\infty \) was out of reach in [16] and which new ideas are introduced in these notes to deal with such problem.

We would like now to briefly comment about the continuity of the vector field u produced by Theorem 1.1. It was observed by Caravenna and Crippa [7] that the boundedness or the continuity of the vector field (in addition to some Sobolev regularity) could play a key role in the uniqueness problem in the class of integrable densities \(\rho \in L^1((0,1) \times {\mathbb {T}}^d)\). It thus turns out to be a very interesting question to ask if, in fact, boundedness or continuity plus Sobolev regularity are enough to guarantee uniqueness. Theorem 1.1 shows that this is not the case, if the integrability of Du is lower than a dimensional threshold (precisely, \(d-1\)).1

The idea that the boundedness or the continuity of u can play a crucial role in the uniqueness problem is confirmed by the fact that the majority of the result concerning existence and uniqueness of the regular Lagrangian flow associated to a Sobolev or BV vector field u assume that \(u \in L^\infty \) (see, for instance, the recent survey [2]).

On a different point of view, it is a classical result (see, for instance, [12]) that the boundedness of u, even without any further Sobolev regularity, is enough to have uniqueness, if a small viscosity is added to the continuity equation:
$$\begin{aligned} \partial _t \rho + \text {div }(\rho u) = \nu \Delta \rho , \quad \nu > 0, \end{aligned}$$
(10)
while in [16] we showed that uniqueness for (10) can drastically fail is u is Sobolev, but not bounded.

The result in Theorem 1.1 is quite surprising, even in comparison with our previous result in [16]. Indeed, for vector fields produced by Theorem 1.1 the Lagrangian picture is very well behaved: first, the Sobolev regularity implies the existence and uniqueness of the regular Lagrangian flow. Second, the continuity of the field implies that the trajectories provided by the regular Lagrangian flow are \(C^1\) in time (and this was not the case for the fields produced in [16]). Third, the bound (2) means that the length of each trajectory is at most \(\varepsilon >0\), i.e. particles almost don’t move (and, again, this was not the case for the fields produced in [16]). Observe also that \(\varepsilon \) in (2) depends neither on the length of the time interval [0, 1] nor on the \(L^1\) distance between the initial and the final datum \(\Vert \rho (1) - \rho (0)\Vert _{L^1({\mathbb {T}}^d)} = \Vert \bar{\rho }\Vert _{L^1({\mathbb {T}}^d)}\). Nevertheless uniqueness in the Eulerian world is completely lost.

We conclude this introduction observing that Theorem 1.1 is an immediate application of the following theorem, whose proof is the topic of all next sections.

Theorem 1.2

Let \(\varepsilon >0\). Let \( \rho _0: [0,1] \times {\mathbb {T}}^d \rightarrow {\mathbb {R}}\), \(u_0 : [0,1] \times {\mathbb {T}}^d \rightarrow {\mathbb {R}}^d\) be smooth with
$$\begin{aligned} \begin{aligned} \int _{{\mathbb {T}}^d} \rho _0(0,x) dx&= \int _{{\mathbb {T}}^d} \rho _0(t,x) dx, \\ \mathrm{div}\, u_0&= 0, \end{aligned} \end{aligned}$$
(11)
for every \(t \in [0,1]\). Set
$$\begin{aligned} E := \big \{t \in [0,1] \ : \ \partial _t \rho _0(t) + \mathrm{div} \,(\rho _0(t) u_0(t)) = 0 \big \}. \end{aligned}$$
(12)
Then there exist \(\rho : [0,1] \times {\mathbb {T}}^d \rightarrow {\mathbb {R}}\), \(u:[0,1] \times {\mathbb {T}}^d \rightarrow {\mathbb {R}}^d\) such that
  1. (a)
    \(\rho \), u have the following regularity:
    $$\begin{aligned} \rho \in C\Big ([0,1]; L^1 ({\mathbb {T}}^d)\Big ), \quad u \in C\Big ([0,1] \times {\mathbb {T}}^d \Big ) \cap \bigcap _{1\le p < d-1} C\Big ([0,1]; W^{1, p}({\mathbb {T}}^d) \Big ); \end{aligned}$$
     
  2. (b)

    \((\rho , u)\) is a weak solution to (1);

     
  3. (c)

    for every \(t \in E\), \(\rho (t) = \rho _0(t)\), \(u(t) = u_0(t)\);

     
  4. (d)
    \(\rho \) is \(\varepsilon \)-close to \(\rho _0\) i.e.
    $$\begin{aligned} \begin{aligned} \max _{t \in [0,1]} \Vert \rho (t) - \rho _0(t)\Vert _{L^1({\mathbb {T}}^d)}&\le \varepsilon . \end{aligned} \end{aligned}$$
     
Condition (d) can be substituted by the following:
  1. (d’)
    u is \(\varepsilon \)-close to \(u_0\) i.e.
    $$\begin{aligned} \begin{aligned} \Vert u - u_0\Vert _{C^0([0,1] \times {\mathbb {T}}^d)} \le \varepsilon . \end{aligned} \end{aligned}$$
     

We emphasize that conditions (d) or (d’) amount to approximability in a strong norm. This is at variance with the h-principle type approximability in a weak norm, as for instance in [5, 8]. In particular, it is easy to see that conditions (d) and (d’) cannot hold simultaneously; Indeed, such a statement would imply that one can construct a sequence \((\rho _k,u_k)\) of weak solutions to (1) such that \((\rho _k,u_k,\rho _ku_k)\rightarrow (\rho _0,u_0,\rho _0u_0)\) in \(L^1\), so that necessarily \((\rho _0,u_0)\) is also a weak solution to (1). See also Remark 4.2.

Proof of Theorem 1.1 assuming Theorem 1.2

Let \(\chi : [0,1] \rightarrow {\mathbb {R}}\) be such that \(\chi \equiv 0\) on [0, 1 / 4], \(\chi \equiv 1\) on [3 / 4, 1]. Apply Theorem 1.2 with \( \rho _0(t,x) := \chi (t) {{\bar{\rho }}}(x)\), \( u_0 = 0\). By (c), \(\rho (0) \equiv 0\) at \(t = 0\) and \(\rho (1) \equiv {{\bar{\rho }}}\) at \(t=1\). Moreover, by (d’), \(\Vert u\Vert _{C^0} \le \varepsilon \). \(\square \)

2 Comments on the proof

We describe in this section what problems arise when one tries to extend the proof provided in [16] to Theorem 1.1, i.e. to the end-point case \(r=\infty \) and which new ideas are introduced to solve such problems.

2.1 Sketch of the paper [16]

We first briefly sketch the proof provided in [16] for the analog of Theorem 1.1 under the conditions (8), (9). The proof is based on a convex integration scheme, with both oscillations and concentration playing a key role. More precisely, the density \(\rho \) and the field u are defined as limit of a sequence \((\rho _q)_q\), \((u_q)_q\) of smooth approximate solutions to the continuity equation
$$\begin{aligned} \partial _t \rho _q + \text {div }(\rho _q u_q) = - \text {div }R_q, \end{aligned}$$
(13)
where \(R_q\) is a smooth vector field converging strongly to zero
$$\begin{aligned} \Vert R_q\Vert _{C_t L^1_x} \lesssim \delta _q \end{aligned}$$
(14)
with \(\delta _q = 2^{-q}\) and \((\rho _q)_q\), \((u_q)_q\) satisfy
$$\begin{aligned} \sum _q \Vert \rho _q - \rho _{q-1}\Vert _{C_t L^{r'}_x}&< \infty , \end{aligned}$$
(15a)
$$\begin{aligned} \sum _q \Vert u_q - u_{q-1}\Vert _{C_t L^r_x}&< \infty , \end{aligned}$$
(15b)
and
$$\begin{aligned} \sum _q \Vert D u_q - Du_{q-1}\Vert _{C_t L^p_x} < \infty . \end{aligned}$$
(16)
In this way \(\rho , u\) are a weak solution to (1) and, moreover, they have the desired regularity.
The sequence \((\rho _q, u_q, R_q)\) is constructed recursively. Assuming \((\rho _{q-1}, u_{q-1}, R_{q-1})\) are given, one defines
$$\begin{aligned} \rho _{q} = \rho _{q-1} + \vartheta _{q}, \quad u_{q} = u_{q-1} + w_{q}, \end{aligned}$$
(17)
where
$$\begin{aligned} \vartheta _{q}(t,x) := F \big (R_{q-1}(t,x)\big ) \Theta _{\mu _q}(\lambda _q x), \quad w_q(t,x) := G\big (R_{q-1}(t,x)\big ) W_{\mu _q} (\lambda _q x), \end{aligned}$$
(18)
where \(\lambda _q\) is an oscillation parameter and \(\mu _q\) is a concentration parameter, suitably chosen at each step of iteration, FG are nonlinear functions and \(\{\Theta _\mu \}_{\mu > 0}\) (resp. \(\{W_\mu \}_{\mu >0}\)) is a family of Mikado densities (resp. Mikado fields) (see Proposition 5.1 below and in particular estimates (47)).
It is proven in [16] that \(\vartheta _q, w_q\) satisfy the following estimates:
$$\begin{aligned} \Vert \vartheta _q\Vert _{C_t L^{r'}_x}&\lesssim \Vert R_{q-1}\Vert _{C_t L^1_x}^{1/r'}, \end{aligned}$$
(19a)
$$\begin{aligned} \Vert w_q\Vert _{C_t L^r_x}&\lesssim \Vert R_{q-1}\Vert _{C_t L^1_x}^{1/r}, \end{aligned}$$
(19b)
$$\begin{aligned} \Vert \vartheta _q\Vert _{C_t L^{1}_x}&\lesssim \mu _q^{-\gamma _1}, \end{aligned}$$
(19c)
$$\begin{aligned} \Vert w_q\Vert _{C_t L^1_x}&\lesssim \mu _q^{-\gamma _2}, \end{aligned}$$
(19d)
$$\begin{aligned} \Vert D w_q\Vert _{C_t L^p_x}&\lesssim \lambda _q \mu _q^{-\gamma _3}, \end{aligned}$$
(19e)
where
$$\begin{aligned}&\gamma _1 = (d-1) \bigg (1 - \frac{1}{r'} \bigg ), \quad \gamma _2 = (d-1) \bigg (1 - \frac{1}{r} \bigg ), \\&\gamma _3 = (d-1) \Bigg [ \frac{1}{r'} + \frac{1}{p} - \bigg ( 1 + \frac{1}{d-1} \bigg ) \Bigg ]. \end{aligned}$$
Notice that \(\gamma _1 > 0\) because \(r < \infty \) (and thus \(r' > 1\)), \(\gamma _2 >0\) because \(r>1\) and \(\gamma _3 > 0\) because of (9). Estimates (19a), (19b) together with the inductive assumption (14) applied to \(R_{q-1}\) guarantee the convergences in (15). Estimate (19e) guarantees the convergence in (16), provided at each step \(\mu _q \gg \lambda _q\), or, more precisely, \(\mu _q = \lambda _q^c\), with c any constant satisfying \(c>1/\gamma _3\).
A computation then shows that, in order for (13) to be satisfied, \(R_q\) must be defined as
$$\begin{aligned} -R_q = \text {div }^{-1} \bigg [ \underbrace{\text {div }(\vartheta _q w_q - R_{q-1})}_{\text {quadratic term}} + \underbrace{\partial _t \vartheta _q + \text {div }(\vartheta _q u_{q-1}) + \text {div }(\rho _{q-1} w_q)}_{\text {linear term}} \bigg ]. \end{aligned}$$
(20)
In order to prove (14), one first uses the oscillation parameter \(\lambda _q\) to make the (antidivergence of the) quadratic term small. Then, in order to estimate the linear term, one can use concentration. For instance, for the term \(\vartheta _q u_{q-1}\), we can use (19c)
$$\begin{aligned} \big \Vert \text {div }^{-1} \big ( \text {div }(\vartheta _q u_{q-1} ) \big ) \Vert _{C_t L^1_x} = \Vert \vartheta _q u_{q-1}\big \Vert _{C_t L^1_x} \lesssim \Vert \vartheta _q\Vert _{C_t L^1_x} \le \mu _q^{-\gamma _1} \le \delta _q, \end{aligned}$$
(21)
provided \(\mu _q\) is chosen large enough. A similar estimate holds for \(\partial _t \vartheta _q\), again using (19c), while for \(\rho _{q-1} w_q\) one must use (19d).

This shows that \(R_q\) can be suitably defined in order to satisfy (20), thus concluding the proof in [16] for the analog of Theorem 1.1 under the assumptions (8), (9). Let us now discuss why the above proof does not apply to Theorem 1.1, i.e. to the case \(r=\infty \), \(r' = 1\).

2.2 First issue

If \(r=\infty \), then estimate (19b) becomes \(\Vert w_q\Vert _{C_{tx}} \lesssim 1\) and this is not enough to prove the convergence in (15b). This issue is solved, modifying the definition of \(\rho _q, u_q\) in (17) as
$$\begin{aligned} \rho _q := \rho _{q-1} + \eta _q \vartheta _q, \quad u_q := u_{q-1} + \frac{1}{\eta _q} w_q, \end{aligned}$$
and choosing \(\eta _q := \Vert R_{q-1}\Vert ^{-1/2}_{C_t L^1_x}\). In this way, using (19a), we get
$$\begin{aligned} \Vert \rho _q - \rho _{q-1}\Vert _{C_t L^1_x} \lesssim \eta _q \Vert R_{q-1}\Vert _{C_t L^1_x} \le \Vert R_{q-1}\Vert _{C_t L^1_x}^{1/2} \le \delta _{q-1}^{1/2} \end{aligned}$$
and
$$\begin{aligned} \Vert u_q - u_{q-1}\Vert _{C_{tx}} \lesssim \frac{1}{\eta _q} \le \Vert R_{q-1}\Vert _{C_t L^1_x}^{1/2} \le \delta _{q-1}^{1/2}, \end{aligned}$$
so that the convergences in (15) still holds, and, moreover, the limit vector field \(u = \lim u_q\) is continuous, being the uniform limit of smooth fields. See Sect. 4 and, in particular, estimates (43), (44).

2.3 Second issue

The second issue concerns the analysis of the linear term in (20) and in particular estimate (21) and the companion estimate for \(\partial _t \vartheta _q\). Indeed, if \(r=\infty \) and \(r'=1\), then \(\gamma _1 = 0\) and thus the concentration paramter \(\mu _q\) can not be used in (21) to make the linear term smaller than \(\delta _q\).

This issue is solved using the inverse flow map associated to \(u_{q-1}\), an idea used in [4] in the framework of the Euler equation, see also [5, 13]. Precisely, one separately considers
$$\begin{aligned} \text {Linear term in}~ (20) = \underbrace{\partial _t \vartheta _q + \text {div }(\vartheta _q u_{q-1})}_{\text {transport term}} + \underbrace{\text {div }(\rho _{q-1}u_q)}_{\text {Nash term}}. \end{aligned}$$
(22)
While for the Nash term an estimate similar to (21) still holds, since \(\gamma _2 = d-1 >0\), in order to treat the transport term, one modifies the definition of \(\vartheta _q\) and \(w_q\) as follows. The time interval [0, 1] is split into N small intervals \(\{I_i\}_i\) of size 1 / N. Denoting by \(t_i\) the middle point of each \(I_i\), one considers the inverse flow map \(\Phi _i\) associated to \(u_{q-1}\)
$$\begin{aligned} {\left\{ \begin{array}{ll} \partial _t \Phi _i + (u_{q-1} \cdot \nabla ) \Phi _i = 0, \\ \Phi _i(t_i,x) = x, \end{array}\right. } \end{aligned}$$
and a partition of unity \(\{\zeta _i\}\) subordinated to the partition \(\{I_i\}_i\) of [0, 1]. The definition in (18) is then modified as follows:
$$\begin{aligned} \begin{aligned} \vartheta _{q}(t,x)&:= F \big (R_{q-1}(t,x)\big ) \sum _i \zeta _i(t) \Theta _{\mu _q} \big ( \lambda _q \Phi _i(t,x) \big ), \\ w_q(t,x)&:= G\big (R_{q-1}(t,x)\big ) \sum _i \zeta _i(t) W_{\mu _q} \big ( \lambda _q \Phi _i(t,x) \big ). \end{aligned} \end{aligned}$$
(23)
With this new definition, the transport term in (22) assumes the form
$$\begin{aligned} \text {Transport term in} \, (22) = \sum _i H_i(t,x) \Theta _{\mu _q} \big ( \lambda _q \Phi _i(t,x) \big ). \end{aligned}$$
The oscillation parameter \(\lambda _q\) can now be used to show that
$$\begin{aligned} \text {div }^{-1} \bigg [ \text {Transport term in} \, (22) \bigg ] \approx \frac{1}{\lambda _q} \lesssim \delta _q. \end{aligned}$$
See Sect. 6.3.

2.4 Third issue

The third issue appears because of the new definition (23) of \(\vartheta _q, w_q\). Indeed if at some time \(t \in [0,1]\) two cutoffs \(\zeta _i(t) \ne 0\), \(\zeta _{i+1}(t) \ne 0\) are active, then in the quadratic term in (20) a term of the form
$$\begin{aligned} \text {div }\bigg [ F(R_{q-1}) G(R_{q-1}) \Theta _{\mu _q} \big ( \lambda _q \Phi _i(t,x) \big ) W_{\mu _q} \big ( \lambda _q \Phi _{i+1}(t,x) \big ) \bigg ] \end{aligned}$$
(24)
appears, i.e. a non-trivial interaction between a Mikado density and a Mikado field. In general there is no reason why one should be able to find a small antidivergence of such term. The problem can be solved, using, at each step q of the construction, two different oscillation parameters \(\lambda _q', \lambda _q''\) and two different concentration parameters \(\mu _q', \mu _q''\) with
$$\begin{aligned} \lambda _q' \ll \lambda _q'', \quad \mu _q' \ll \mu _q'' \end{aligned}$$
and modifying one more time the definition of \(\vartheta _q, w_q\) as follows:
$$\begin{aligned} \begin{aligned} \vartheta _q(t,x)&= F(R_{q-1}(t,x)) \left[ \sum _{i \text { odd}} \zeta _i(t) \Theta _{\mu _q'} \big (\lambda _q' \Phi _i(t,x) \big ) + \sum _{i \text { even}} \zeta _i(t) \Theta _{\mu _q''} \big (\lambda _q'' \Phi _i(t,x) \big ) \right] , \\ w_q(t,x)&= G(R_{q-1}(t,x)) \left[ \sum _{i \text { odd}} \zeta _i(t) W_{\mu _q'} \big (\lambda _q' \Phi _i(t,x) \big ) + \sum _{i \text { even}} \zeta _i(t) W_{\mu _q''} \big (\lambda _q'' \Phi _i(t,x) \big ) \right] . \end{aligned} \end{aligned}$$
See (55) for the slightly different, precise definition of the perturbations. With this new definition, the main term in the non-trivial interaction in (24) becomes of the form
$$\begin{aligned} \Theta _{\mu '_q} \big ( \lambda _q' \Phi _i(t,x) \big ) W_{\mu _q''} \big ( \lambda _q'' \Phi _{i+1}(t,x) \big ) \ \text { or } \ \Theta _{\mu ''_q} \big ( \lambda _q'' \Phi _i(t,x) \big ) W_{\mu _q'} \big ( \lambda _q' \Phi _{i+1}(t,x) \big ), \end{aligned}$$
(25)
i.e. the product of a fast oscillating function (with frequency \(\lambda _q'\)) with a very fast oscillating function (with frequency \(\lambda _q''\)), where one of the two factors (namely \(W_{\mu _q'}\) or \(W_{\mu _q''}\)) is small in \(L^1({\mathbb {T}}^d)\) because of the concentration mechanism (compare with estimate (19d)). One can then use an improved Hölder inequality (see Lemma 3.4) to show that the terms in (25) are small in \(L^1\) and thus conclude the proof of Theorem 1.1. See Sect. 6.2 and in particular Lemma 6.1.

3 Technical tools

In this section we provide some technical tools which will be frequently used in the following. We start by fixing some notation:
  • \({\mathbb {T}}^d = {\mathbb {R}}^d / {\mathbb {Z}}^d\) is the d-dimensional flat torus, \(d \ge 3\).

  • If g(x) is a smooth function of \(x \in {\mathbb {T}}^d\), we denote by \(\Vert g\Vert _{L^p({\mathbb {T}}^d)}\), or simply by \(\Vert g\Vert _{L^p}\), its \(L^p\)-norm, for \(p \in [1,\infty ]\).

  • If f(tx) is a smooth function of \(t \in [0,1]\) and \(x \in {\mathbb {T}}^d\), we denote by
    • \(\Vert f\Vert _{C^k}\) the sup norm of f together with the sup norm of all its derivatives in time and space up to order k;

    • \(\Vert f(t)\Vert _{C^k({\mathbb {T}}^d)}\), or simply \(\Vert f(t)\Vert _{C^k}\), the sup norm of \(x \mapsto f(t,x)\) together with the sup norm of all its spatial derivatives up to order k at fixed time t;

    • \(\Vert f(t)\Vert _{L^p({\mathbb {T}}^d)}\), or simply \(\Vert f(t)\Vert _{L^p}\), the \(L^p\) norm of f in the spatial derivatives, at fixed time t.

  • If \(f: [0,1] \rightarrow {\mathbb {R}}\) is a function of time only, we will denote by \(\dot{f} = df/dt\) its derivative.

  • \(C^\infty _0({\mathbb {T}}^d)\) is the set of smooth functions on the torus with zero mean value.

  • \({\mathbb {N}}= \{0,1,2, \ldots \}\), \({\mathbb {N}}^* = {\mathbb {N}}\setminus \{0\}\).

  • We will use the notation \(C(A_1, \ldots , A_n)\) to denote a constant which depends only on the numbers \(A_1, \ldots , A_n\).

3.1 Diffeomorphisms of the flat torus

We discuss in this section standard properties of diffeomorphisms of the flat torus. Let \(\Phi : {\mathbb {R}}^d \rightarrow {\mathbb {R}}^d\) be a smooth diffeomorphism. We say that \(\Phi \) is a diffeomorphism of \({\mathbb {T}}^d\), and we write \(\Phi : {\mathbb {T}}^d \rightarrow {\mathbb {T}}^d\), if
$$\begin{aligned} \Phi (x + k) = \Phi (x) + k, \quad \text { for every } k \in {\mathbb {Z}}^d. \end{aligned}$$
We say that a diffeomorphism \(\Phi : {\mathbb {T}}^d \rightarrow {\mathbb {T}}^d\) is measure-preserving if \(|\det D \Phi (x)| = 1\) for every \(x \in {\mathbb {T}}^d\). Given a diffeomorphism \(\Phi \), we will often consider
  1. (1)

    the derivative \(D \Phi : {\mathbb {T}}^d \rightarrow {\mathbb {R}}^{d \times d}\);

     
  2. (2)

    the inverse-matrix of the derivative \((D \Phi )^{-1} : {\mathbb {T}}^d \rightarrow {\mathbb {R}}^{d \times d}\);

     
  3. (3)

    higher order derivatives of the inverse-matrix of the derivative \(D^k ((D \Phi )^{-1}) : {\mathbb {T}}^d \rightarrow {\mathbb {R}}^{d(k+2)}\).

     
Observe that, given a matrix \(A \in {\mathbb {R}}^{d \times d}\), with \(|\det A| = 1\), it holds \(|A| \ge 1\), where \(|A| := \max _{|u| = 1} |A u|\) is the norm of matrix A. Therefore if \(\Phi \) is a measure-preserving diffeomorphism, then \(|D \Phi (x)| \ge 1\) for every \(x \in {\mathbb {T}}^d\) and thus \(1 \le \Vert D \Phi \Vert _{C^k}^\alpha \le \Vert D \Phi \Vert _{C^k}^\beta \) for every \(0< \alpha < \beta \). Recall also that for a given invertible matrix A,
$$\begin{aligned} A^{-1} = \frac{1}{\det A} (\mathrm{cof}\, A)^T, \end{aligned}$$
where \((\mathrm{cof} \, A)^T\) is transpose of the cofactor matrix of A.

Lemma 3.1

Let \(\Phi : {\mathbb {T}}^d \rightarrow {\mathbb {T}}^d\) be a measure-preserving smooth diffeomorphism. Then, for every \(k \in {\mathbb {N}}\),
$$\begin{aligned} \Vert D^k ((D \Phi )^{-1})\Vert _{C^0({\mathbb {T}}^d)} \le C_k \Vert D \Phi \Vert _{C^k({\mathbb {T}}^d)}^{d-1}, \end{aligned}$$
where \(C_k\) is a constant depending only on k (and on the dimension d).

Proof

For any fixed \(x \in {\mathbb {T}}^d\) it holds
$$\begin{aligned} \Big |\big [D \Phi (x)\big ]^{-1}\Big | = \bigg | \frac{1}{\det D \Phi (x)} (\mathrm{cof} \, D \Phi (x))^T \bigg | = \big | \mathrm{cof} \, D \Phi (x)\big |. \end{aligned}$$
(26)
The conclusion now follows from the definition of cofactor matrix.

Lemma 3.2

Let \(G: {\mathbb {T}}^d \rightarrow {\mathbb {R}}^d\), \(g: {\mathbb {T}}^d \rightarrow {\mathbb {R}}\) be smooth and assume \(\text {div }G = g\). Let \(\Phi : {\mathbb {T}}^d \rightarrow {\mathbb {T}}^d\) be a measure-preserving diffeomorphism of the torus. Then
$$\begin{aligned} \mathrm{div}\, \Big [ (D \Phi )^{-1} G(\Phi ) \Big ] = g (\Phi ). \end{aligned}$$

Proof

We show that for every \(\varphi \in C^\infty ({\mathbb {T}}^d)\) it holds
$$\begin{aligned} \int _{{\mathbb {T}}^d} \varphi \, \text {div }\Big [ (D\Phi )^{-1} G(\Phi ) \Big ] dx = \int _{{\mathbb {T}}^d} \varphi \, g(\Phi ) dx. \end{aligned}$$
(27)
Set \({{\tilde{\varphi }}} := \varphi \circ \Phi ^{-1}\). It holds
$$\begin{aligned} \begin{aligned} \int _{{\mathbb {T}}^d} \varphi \, \text {div }\Big [ (D\Phi )^{-1} G(\Phi ) \Big ] dx&= \int _{{\mathbb {T}}^d} {{\tilde{\varphi }}} (\Phi ) \, \text {div }\Big [ (D\Phi )^{-1} G(\Phi ) \Big ] dx \\&= - \int _{{\mathbb {T}}^d} \big [ (D \Phi )^T \nabla {{\tilde{\varphi }}}(\Phi ) \big ] \cdot \big [ (D\Phi )^{-1} G(\Phi ) \big ] dx \\&= - \int _{{\mathbb {T}}^d} \nabla {{\tilde{\varphi }}}(\Phi ) \cdot G( \Phi ) dx \\ \text {(changing variable} \, y = \Phi (x)\text {)}&= - \int _{{\mathbb {T}}^d} \nabla {{\tilde{\varphi }}} \cdot G \, dy \\&= \int _{{\mathbb {T}}^d} {{\tilde{\varphi }}} \, \text {div }G \, dy \\&= \int _{{\mathbb {T}}^d} {{\tilde{\varphi }}} \, g \, dy \\&= \int _{{\mathbb {T}}^d} \varphi (\Phi ^{-1})\, g \, dy \\ \text {(changing variable} \, x = \Phi ^{-1}(y)\text {)}&= \int _{{\mathbb {T}}^d} \varphi \, g(\Phi ) dx, \end{aligned} \end{aligned}$$
thus concluding the proof of the lemma.

Lemma 3.3

Let \(g: {\mathbb {T}}^d \rightarrow {\mathbb {R}}\) be a smooth function. Let \(\Phi : {\mathbb {T}}^d \rightarrow {\mathbb {T}}^d\) be a measure-preserving diffeomorphism. Then for every \(p \in [1, \infty ]\) and \(k \in {\mathbb {N}}\), \(k \ge 1\),
$$\begin{aligned} \begin{aligned} \Vert g \circ \Phi \Vert _{L^p({\mathbb {T}}^d)}&= \Vert g\Vert _{L^p({\mathbb {T}}^d)}, \end{aligned} \end{aligned}$$
and
$$\begin{aligned} \begin{aligned} \Vert g \circ \Phi \Vert _{W^{k,p}({\mathbb {T}}^d)}&\le C_k \Vert D \Phi \Vert _{C^{k-1}({\mathbb {T}}^d)}^k \Vert g\Vert _{W^{k,p}({\mathbb {T}}^d)}. \end{aligned} \end{aligned}$$

The proof is an easy application of the chain rule and thus it is omitted.

3.2 Properties of fast oscillations

We discuss now some properties of fast oscillating periodic functions. For a given \(g : {\mathbb {T}}^d \rightarrow {\mathbb {R}}\) and \(\lambda \in {\mathbb {N}}^*\), we set
$$\begin{aligned} g_\lambda (x) := g(\lambda x). \end{aligned}$$
Observe that for every \(p \in [1, \infty ]\) and \(k \in {\mathbb {N}}\),
$$\begin{aligned} \Vert D^k g_\lambda \Vert _{L^{p}({\mathbb {T}}^d)} = \lambda ^k \Vert D^k g\Vert _{L^{p}({\mathbb {T}}^d)}. \end{aligned}$$
(28)

3.2.1 Improved Hölder inequality

In the same spirit as in [16] and [6], we now prove an improved Hölder inequality for the product of a slow oscillating function with a fast oscillating functions composed with a diffeomorphism.

Lemma 3.4

(Improved Hölder inequality). Let \(f,g: {\mathbb {T}}^d \rightarrow {\mathbb {R}}\) be smooth functions, \(\lambda \in {\mathbb {N}}^*\) and \(\Phi : {\mathbb {T}}^d \rightarrow {\mathbb {T}}^d\) be a measure-preserving diffeomorphism. Then for every \(p \in [1,\infty ]\),
$$\begin{aligned} \Vert f g_\lambda \Vert _{L^p({\mathbb {T}}^d)} \le \Vert f\Vert _{L^p({\mathbb {T}}^d)} \Vert g\Vert _{L^p({\mathbb {T}}^d)} + \frac{C_p}{\lambda ^{1/p}} \Vert f\Vert _{C^1({\mathbb {T}}^d)} \Vert g\Vert _{L^p({\mathbb {T}}^d)} \end{aligned}$$
(29)
and
$$\begin{aligned} \Vert f \cdot (g_\lambda \circ \Phi )\Vert _{L^p({\mathbb {T}}^d)}\le & {} \Vert f\Vert _{L^p({\mathbb {T}}^d)} \Vert g\Vert _{L^p({\mathbb {T}}^d)} \nonumber \\&+ \frac{C_p}{\lambda ^{1/p}}\Vert f\Vert _{C^1({\mathbb {T}}^d)} \Vert D \Phi \Vert ^{d-1}_{C^0({\mathbb {T}}^d)} \Vert g\Vert _{L^p({\mathbb {T}}^d)}. \end{aligned}$$
(30)

Here \(f \cdot (g_\lambda \circ \Phi )\) is the function \(x \mapsto f(x) g(\lambda \Phi (x))\).

Proof

For a proof of (29), see [16, Lemma 2.1]. Concerning (30), we argue as follows. Since \(\Phi \) is a measure-preserving diffeomorphism, it holds
$$\begin{aligned} \Vert f \cdot (g_\lambda \circ \Phi )\Vert _{L^p} = \Vert (f \circ \Phi ^{-1}) \cdot g_{\lambda }\Vert _{L^p}. \end{aligned}$$
Therefore we can apply (29) to get
$$\begin{aligned} \begin{aligned} \Vert f \cdot (g_\lambda \circ \Phi )\Vert _{L^p}&\le \Vert f \circ \Phi ^{-1}\Vert _{L^p} \Vert g\Vert _{L^p} + \frac{C_p}{\lambda ^{1/p}} \Vert f \circ \Phi ^{-1}\Vert _{C^1} \Vert g\Vert _{L^p} \\ \text {(by Lemma}~3.3~ \text {and} \, (28)\text {)}&\le \Vert f\Vert _{L^p} \Vert g\Vert _{L^p} + \frac{C_p}{\lambda ^{1/p}} \Vert f\Vert _{C^1} \Vert D(\Phi ^{-1})\Vert _{C^0} \Vert g\Vert _{L^p} \\&\le \Vert f\Vert _{L^p} \Vert g\Vert _{L^p} + \frac{C_p}{\lambda ^{1/p}} \Vert f\Vert _{C^1} \Vert (D\Phi )^{-1}\Vert _{C^0} \Vert g\Vert _{L^p} \\ \text {(by Lemma}~3.1\text {)}&\le \Vert f\Vert _{L^p} \Vert g\Vert _{L^p} + \frac{C_p}{\lambda ^{1/p}} \Vert f\Vert _{C^1} \Vert D\Phi \Vert ^{d-1}_{C^0} \Vert g\Vert _{L^p}. \end{aligned} \end{aligned}$$
\(\square \)

3.2.2 Antidivergence operators

In this section we introduce two antidivergence operators, a standard and an improved one, in the same spirit as in [16].

For \(f \in C^\infty _0({\mathbb {T}}^d)\) there exists a unique \(u \in C^\infty _0({\mathbb {T}}^d)\) such that \(\Delta u = f\). The operator \(\Delta ^{-1}: C^\infty _0({\mathbb {T}}^d) \rightarrow C^\infty _0({\mathbb {T}}^d)\) is thus well defined. We define the standard antidivergence operator as \(\nabla \Delta ^{-1}: C^\infty _0({\mathbb {T}}^d) \rightarrow C^\infty ({\mathbb {T}}^d; {\mathbb {R}}^d)\). It clearly satisfies \(\text {div }(\nabla \Delta ^{-1} f) = f\).

Lemma 3.5

For every \(k \in {\mathbb {N}}\) and \(p \in [1, \infty ]\), the standard antidivergence operator satisfies the bounds
$$\begin{aligned} \big \Vert D^k (\nabla \Delta ^{-1} g) \big \Vert _{L^p({\mathbb {T}}^d)} \le C_{k,p} \Vert D^k g\Vert _{L^p({\mathbb {T}}^d)}. \end{aligned}$$
(31)

For the proof, see [16, Lemma 2.2] .

We now introduce an improved antidivergence operator, which allows us to get better (w.r.t \(\nabla \Delta ^{-1}\)) estimates, if applied to the product of a slow oscillating function with a fast oscillating one composed with a diffeomorphism.

Lemma 3.6

Let \(f, g: {\mathbb {T}}^d \rightarrow {\mathbb {R}}\) be smooth function withLet \(\lambda \in {\mathbb {N}}^*\) and \(\Phi : {\mathbb {T}}^d \rightarrow {\mathbb {T}}^d\) be a smooth, measure-preserving diffeomorphism. Then there exists a smooth vector field \(u : {\mathbb {T}}^d \rightarrow {\mathbb {R}}^d\) so thatand for every \(k \in {\mathbb {N}}\), \(p \in [1, \infty ]\),
$$\begin{aligned} \Vert u\Vert _{W^{k,p}({\mathbb {T}}^d)} \le C_{k,p} \lambda ^{k-1} \Vert f\Vert _{C^{k+1}({\mathbb {T}}^d)} \Vert D\Phi \Vert _{C^{k}({\mathbb {T}}^d)}^{d-1+k} \Vert g\Vert _{W^{k,p}({\mathbb {T}}^d)}. \end{aligned}$$
(33)
We will use the notation

Remark 3.7

The same result holds if fg are vector fields and \(\cdot \) in (32) denotes the scalar product.

Proof

Since g has zero mean value, we can define
$$\begin{aligned} G := \nabla \Delta ^{-1} g. \end{aligned}$$
(34)
Let us denote by \(H: {\mathbb {T}}^d \rightarrow {\mathbb {R}}^d\) the vector field
$$\begin{aligned} H := \frac{1}{\lambda } (D\Phi )^{-1} G_\lambda (\Phi ). \end{aligned}$$
(35)
By Lemma 3.2 it holds
$$\begin{aligned} \text {div }H = \frac{1}{\lambda } (\text {div }G_\lambda ) (\Phi ) = g_\lambda (\Phi ). \end{aligned}$$
We now setLet us first check that u satisfies (32). It holdsWe prove now that (33) holds. We first estimate H as follows:
$$\begin{aligned} \Vert H\Vert _{W^{k,p}}&\le \frac{1}{\lambda } \left\| (D \Phi )^{-1} \right\| _{C^k} \Vert G_\lambda \circ \Phi \Vert _{W^{k,p}} \\ \text {(by Lemma}~3.1\text {)}&\le \frac{C_k}{\lambda }\Vert D \Phi \Vert _{C^k}^{d-1} \Vert G_\lambda \circ \Phi \Vert _{W^{k,p}} \\ \text {(by Lemma}~3.3\text {)}&\le \frac{C_k}{\lambda } \Vert D \Phi \Vert _{C^k}^{d-1+k} \Vert G_\lambda \Vert _{W^{k,p}} \\ \text {(by}~(28)\text {)}&\le C_k \lambda ^{k-1} \Vert D \Phi \Vert _{C^k}^{d-1+k} \Vert G\Vert _{W^{k,p}} \\ \text {(by Lemma}~3.5\text {)}&\le C_{k,p} \lambda ^{k-1} \Vert D \Phi \Vert _{C^k}^{d-1+k} \Vert g\Vert _{W^{k,p}}. \end{aligned}$$
Using now again Lemma 3.5, we can write
$$\begin{aligned} \Vert u\Vert _{W^{k,p}}&\le \Vert fH\Vert _{W^{k,p}} + C_{k,p} \Vert \nabla f \cdot H\Vert _{W^{k,p}} \\&\le \Vert f\Vert _{C^k} \Vert H\Vert _{W^{k,p}} + C_{k,p} \Vert \nabla f\Vert _{C^k} \Vert H\Vert _{W^{k,p}} \\&\le C_{k,p} \Vert f\Vert _{C^{k+1}} \Vert H\Vert _{W^{k,p}} \\&\le C_{k,p} \lambda ^{k-1} \Vert f\Vert _{C^{k+1}} \Vert D \Phi \Vert _{C^k}^{d-1+k} \Vert g\Vert _{W^{k,p}}, \end{aligned}$$
which is what we wanted to prove.

Remark 3.8

In Lemma 3.6, if \(f,g,\Phi \) are smooth functions of (tx), \(t \in [0,1]\), \(x \in {\mathbb {T}}^d\) and at each time \(t \in [0,1]\), they satisfy the assumptions of Lemma 3.6, then we can apply \({\mathcal {R}}\) at each time and define a time-dependent vector field \(u(t, \cdot )\) satisfying (32) and (33). Moreover u turns out to be a smooth function of (tx).

3.2.3 Mean value and fast oscillations

In this section we prodide an estimate on the mean value of the product of a slow oscillating function with a fast oscillating function composed with a diffeomorphism.

Lemma 3.9

Let \(f, g : {\mathbb {T}}^d \rightarrow {\mathbb {R}}\), with Open image in new window. Let \(\lambda \in {\mathbb {N}}^*\) and \(\Phi : {\mathbb {T}}^d \rightarrow {\mathbb {T}}^d\) be a measure-preserving diffeomorphism. Thenand

Proof of Theorem 1.1 assuming Theorem 1.2

For a proof of (36), see [16, Lemma 2.6]. The proof of (37) follows from (36), observing that\(\square \)

4 Statement of the main proposition and proof of Theorem 1.2

We assume without loss of generality \({\mathbb {T}}^d\) is the periodic extension of the unit cube \([0,1]^d\). The following proposition contains the key facts used to prove Theorem 1.2. Let us first introduce the continuity-defect equation:
$$\begin{aligned} \left\{ \begin{aligned}&\partial _t \rho + \text {div }(\rho u) = - \text {div }R, \\&\text {div }u = 0. \end{aligned} \right. \end{aligned}$$
(38)
We will call R the defect field.

Proposition 4.1

There exists a constant \(M>0\) such that the following holds. Let \(p \in [1, d-1)\), \(\eta , \delta > 0\) and let \((\rho _0, u_0, R_0)\) be a smooth solution of the continuity-defect equation (38). Then there exists another smooth solution \((\rho _1, u_1, R_1)\) of (38) such that for every \(t \in [0,1]\),
$$\begin{aligned} \Vert \rho _1(t) - \rho _0(t)\Vert _{L^1({\mathbb {T}}^d)}&\le M \eta \Vert R_0(t)\Vert _{L^1({\mathbb {T}}^d)}, \end{aligned}$$
(39a)
$$\begin{aligned} \Vert u_1(t) - u_0(t)\Vert _{C^0({\mathbb {T}}^d)}&\le M \eta ^{-1}, \end{aligned}$$
(39b)
$$\begin{aligned} \Vert u_1(t) - u_0(t)\Vert _{W^{1,p}({\mathbb {T}}^d)}&\le \delta , \end{aligned}$$
(39c)
$$\begin{aligned} \Vert R_1(t)\Vert _{L^1({\mathbb {T}}^d)}&\le \delta , \end{aligned}$$
(39d)
and, moreover, if at some time \(t \in [0,1]\), \(R_0(t) =0\), then
$$\begin{aligned} \rho _1(t) - \rho _0(t) = u_1(t) - u_0(t) = R_1(t) = 0. \end{aligned}$$

Proof of Theorem 1.2 assuming Proposition 4.1

For \(\rho _0, u_0\) in the statement of Theorem 1.2, define
$$\begin{aligned} R_0(t) := - \nabla \Delta ^{-1} \Big (\partial _t \rho _0(t) + \text {div }(\rho _0(t) u_0(t)) \Big ). \end{aligned}$$
By (11), \(R_0\) is well defined, it is smooth and \((\rho _0, u_0, R_0)\) solve the continuity-defect equation.
Let \((p_q)_{q \in {\mathbb {N}}}\) be a fixed increasing sequence of real numbers such that \(p_q \rightarrow d-1\) as \(q \rightarrow \infty \). Let also \((\eta _q)_{q \in {\mathbb {N}}}\), \((\delta _q)_{q \in {\mathbb {N}}}\) be two sequence of positive real numbers, which will be fixed later. Starting from \((\rho _0, u_0, R_0)\), we can recursively apply Proposition 4.1 to obtain a sequence \((\rho _q, u_q, R_q)_{q \in {\mathbb {N}}}\) of smooth solutions to the continuity-defect equation such that
$$\begin{aligned} \Vert \rho _{q+1}(t) - \rho _q(t)\Vert _{L^1({\mathbb {T}}^d)}&\le M \eta _q \Vert R_q(t)\Vert _{L^1({\mathbb {T}}^d)}, \end{aligned}$$
(40a)
$$\begin{aligned} \Vert u_{q+1}(t) - u_q(t)\Vert _{C^0({\mathbb {T}}^d)}&\le M \eta _q^{-1}, \end{aligned}$$
(40b)
$$\begin{aligned} \Vert u_{q+1}(t) - u_q(t)\Vert _{W^{1,p_q}({\mathbb {T}}^d)}&\le \delta _q, \end{aligned}$$
(40c)
$$\begin{aligned} \Vert R_{q+1}(t)\Vert _{L^1({\mathbb {T}}^d)}&\le \delta _q, \end{aligned}$$
(40d)
for all times \(t \in [0,1]\) and
$$\begin{aligned} \rho _{q+1}(t) = \rho _q(t), \quad u_{q+1}(t) = u_q(t), \quad R_{q+1}(t) = 0, \end{aligned}$$
for all times t such that \(R_q(t) = 0\). Therefore, by induction, we get from (40a) and (40d) that for all \(t \in [0,1]\) and all \(q \in {\mathbb {N}}\),
$$\begin{aligned} \Vert \rho _{q+1}(t) - \rho _q(t)\Vert _{L^1({\mathbb {T}}^d)} \le M \eta _q \delta _{q-1}, \end{aligned}$$
(41)
where we set \(\delta _{-1} := \max _{t \in [0,1]} \Vert R_0(t)\Vert _{L^1}\) and, moreover,
$$\begin{aligned} \rho _{q+1}(t) = \rho _q(t), \quad u_{q+1}(t) = u_q(t) \quad \text { for all } t \in E, \end{aligned}$$
(42)
where E was defined in (12). We now choose \((\delta _q)_{q \in {\mathbb {N}}}\) so that
$$\begin{aligned} \sum _{q=0}^{+\infty } \delta _q< \sum _{q=0}^{+\infty } \delta _q^{1/2} < \infty \end{aligned}$$
and
$$\begin{aligned} \eta _q := \sigma \delta _{q-1}^{-1/2} \end{aligned}$$
for \(q \in {\mathbb {N}}\), where \(\sigma > 0\) is a positive number, to be defined later. From (41) we get, for all \(t \in [0,1]\),
$$\begin{aligned} \sum _{q = 0}^{+\infty } \Vert \rho _{q+1}(t) - \rho _q(t)\Vert _{L^1({\mathbb {T}}^d)} \le M \sum _{q=0}^{+\infty } \eta _q \delta _{q-1} = M \sigma \sum _{q=0}^{+\infty } \delta _{q-1}^{1/2} < \infty , \end{aligned}$$
(43)
and thus there exists \(\rho \in C([0,1]; L^1({\mathbb {T}}^d))\) so that \(\rho _q \rightarrow \rho \) in \(C([0,1]; L^1({\mathbb {T}}^d))\). Similarly, using (40b), for all \(t \in [0,1]\),
$$\begin{aligned} \sum _{q=0}^{+\infty } \Vert u_{q+1}(t) - u_q(t)\Vert _{C^0({\mathbb {T}}^d)} \le M \sum _{q=0}^{+\infty } \eta _q^{-1} = M \sigma ^{-1} \sum _{q=0}^{+\infty } \delta _{q-1}^{1/2} < \infty , \end{aligned}$$
(44)
and thus there exists \(u \in C([0,1] \times {\mathbb {T}}^d; {\mathbb {R}}^d)\) so that \(u_q \rightarrow u\) uniformly. It follows now from (40d) that \(\rho , u\) solve (1).
To prove that \(u \in \bigcap _{1 \le p < d-1} C_t W_x^{1, p}\), fix \(p \in [1, d-1)\). There is \(q^*\) so that \(p_{q} > p\) for every \(q > q^*\). We now have, for all \(t \in [0,1]\),
$$\begin{aligned} \begin{aligned} \sum _{q=0}^{+\infty } \Vert u_{q+1}(t) - u_q(t)\Vert _{W^{1, p}({\mathbb {T}}^d)}&= \sum _{q=0}^{q^*} \Vert u_{q+1}(t) - u_q(t)\Vert _{W^{1, p}({\mathbb {T}}^d)} \\&\quad + \sum _{q=q^*+1}^{+\infty } \Vert u_{q+1}(t) - u_q(t)\Vert _{W^{1, p}({\mathbb {T}}^d)} \\ \text {(since} \, p< p_q \, \text {for} \, q > q^*\text {)}&\le \sum _{q=0}^{q^*} \Vert u_{q+1}(t) - u_q(t)\Vert _{W^{1, p}({\mathbb {T}}^d)} \\&\quad + \sum _{q=q^*+1}^{+\infty } \Vert u_{q+1}(t) - u_q(t)\Vert _{W^{1, p_q}({\mathbb {T}}^d)} \\ \text {(by}~ (\mathrm{40c})\text {)}&\le \sum _{q=0}^{q^*} \Vert u_{q+1}(t) - u_q(t)\Vert _{W^{1, p}({\mathbb {T}}^d)} + \sum _{q=q^*+1}^{+\infty } \delta _q < \infty , \end{aligned} \end{aligned}$$
thus proving that \(u \in C([0,1]; W^{1, p}({\mathbb {T}}^d))\). This concludes the proof of parts (a), (b) in the statement of Theorem 1.2.
It follows from (42) that \(\rho (t) = \rho _0(t)\) and \(u(t) = u_0(t)\), whenever \(t \in E\), and thus part (c) is also proven. To prove (d), we observe that, from (43), for all \(t \in [0,1]\),
$$\begin{aligned} \Vert \rho (t) - \rho _0(t)\Vert _{L^1({\mathbb {T}}^d)} \le \sum _{q = 0}^{+\infty } \Vert \rho _{q+1}(t) - \rho _q(t)\Vert _{L^1({\mathbb {T}}^d)} \le M \sigma \sum _{q=0}^\infty \delta _{q-1}^{1/2} \end{aligned}$$
and thus (d) follows choosing
$$\begin{aligned} \sigma := \frac{\varepsilon }{M \sum _{q=0}^{+\infty } \delta _{q-1}^{1/2}}. \end{aligned}$$
Alternatively, to achieve (d’), we observe that, from (44), for all \(t \in [0,1]\),
$$\begin{aligned} \Vert u - u_0\Vert _{C^0} \le M \sigma ^{-1} \sum _{q=0}^{+\infty } \delta _{q-1}^{1/2} \end{aligned}$$
and thus (d)’ follows choosing
$$\begin{aligned} \sigma := \frac{M \sum _{q=0}^{+\infty } \delta _{q-1}^{1/2}}{\varepsilon }. \end{aligned}$$
\(\square \)

Remark 4.2

Estimate (39a), with \(\eta \) in the r.h.s., and estimate (39b), with \(\eta ^{-1}\) in the r.h.s. show that conditions (d)–(d’) in the statement of Theorem 1.2 can not be simultaneously achieved.

5 The perturbations

In this and the next two sections we prove Proposition 4.1. In particular in this section we fix the constant M in the statement of the proposition, we define the functions \(\rho _1\) and \(u_1\) and we estimate them. In Sect. 6 we define \(R_1\) and we estimate it. In Sect. 7 we conclude the proof of Proposition 4.1.

5.1 Mikado fields and Mikado densities

We recall the following proposition from [16].

Proposition 5.1

Let \(a, b \in {\mathbb {R}}\) with
$$\begin{aligned} a+b = d-1. \end{aligned}$$
(45)
For every \(\mu >2d\) and \(j=1, \ldots , d\) there exist a Mikado density \(\Theta _{\mu }^{j} : {\mathbb {T}}^d \rightarrow {\mathbb {R}}\) and a Mikado field \(W_\mu ^j :{\mathbb {T}}^d \rightarrow {\mathbb {R}}^d\) with the following properties.
  1. (a)
    It holds where \(\{e_j\}_{j=1,\ldots ,d}\) is the standard basis in \({\mathbb {R}}^d\).
     
  2. (b)
    For every \(k \in {\mathbb {N}}\) and \(r \in [1,\infty ]\)
    $$\begin{aligned} \begin{aligned} \Vert D^k \Theta _\mu ^j\Vert _{L^r({\mathbb {T}}^d)}&\le M_{k}\, \mu ^{a + k - (d-1) /r}, \\ \Vert D^k W_\mu ^j\Vert _{L^{r}({\mathbb {T}}^d)}&\le M_{k}\, \mu ^{b + k - (d-1)/r}, \end{aligned} \end{aligned}$$
    (47)
    where \(M_{k}\) is a constant which depends only on k, but not on r and \(\mu \).
     
  3. (c)

    For \(j \ne k\), \(\mathrm {supp} \ \Theta _\mu ^j = \mathrm {supp} \ W_\mu ^j\) and \(\mathrm {supp} \ \Theta _\mu ^j \cap \mathrm {supp} \ W_{\mu }^k = \emptyset \).

     
We now define the constant M in the statement of Proposition 4.1 as
$$\begin{aligned} M := 4d \max \Big \{ M_{0}, \, M_{0}^2, \, M_0 + M_1 \Big \}, \end{aligned}$$
(48)
and we choose
$$\begin{aligned} a := d-1, \quad b := 0, \end{aligned}$$
(49)
in Proposition 5.1. In this way for each direction \(j =1,\ldots , d\), we obtain a family of Mikado densities \(\{\Theta _\mu ^j\}_{\mu > 2d}\) and fields \(\{W_\mu ^j\}_{\mu >2d}\), obeying the following estimates:
$$\begin{aligned} \begin{aligned} \sum _{j=1}^d \Vert \Theta _\mu ^j\Vert _{L^1({\mathbb {T}}^d)}, \ \sum _{j=1}^d \Vert W_\mu ^j\Vert _{L^{\infty }({\mathbb {T}}^d)}, \ \sum _{j=1}^d \Vert \Theta _\mu ^j W_\mu ^j\Vert _{L^1({\mathbb {T}}^d)}&\le \frac{M}{4}, \end{aligned} \end{aligned}$$
(50)
and
$$\begin{aligned} \begin{aligned} \Vert W_\mu ^j\Vert _{L^{1}({\mathbb {T}}^d)} \le M \mu ^{-(d-1)}, \quad \Vert W_\mu ^j\Vert _{W^{1, p}({\mathbb {T}}^d)} \le M \mu ^{1-(d-1)/p}, \end{aligned} \end{aligned}$$
(51)
and
$$\begin{aligned} \Vert \Theta _\mu ^j\Vert _{C^1({\mathbb {T}}^d)} \le M \mu ^d, \quad \Vert W_\mu ^j\Vert _{C^1({\mathbb {T}}^d)} \le M \mu . \end{aligned}$$
(52)

5.2 Definition of the perturbations

We are now in a position to define \(\rho _1\), \(u_1\). The constant M has already been fixed in (48). Let thus \(p \in [1, d-1)\), \(\eta , \delta >0\) and \((\rho _0, u_0, R_0)\) be a smooth solution to the continuity-defect equation (38).

Let
$$\begin{aligned} \begin{aligned}&\tau \in 1/{\mathbb {N}}^*&\text { ``time scale''} \\&\lambda ', \lambda '' \in {\mathbb {N}}&\text { ``oscillation'' } \\&\mu ', \mu '' > 2d&\text { ``concentration'' } \end{aligned} \end{aligned}$$
be parameters, which will be fixed later. Set
$$\begin{aligned} N := 1/\tau \in {\mathbb {N}}^*. \end{aligned}$$
For every \(i=1,2,\ldots , N\), let \(I_i := [i\tau , (i+1)\tau ]\) and let \(t_i := (i+1/2)\tau \) be the midpoint of \(I_i\). Consider a partition of unity \(\{\zeta _i\}_{i=1,\ldots , N}\) subordinate to the family of intervals \(\{I_i\}_{i=1,\ldots , N}\). More precisely, for every \(i=1,\ldots , N\), \(\zeta _i \in C^\infty ([0,1])\) and
  • \(\mathrm {supp} \ \zeta _i \in [(i-1/3)\tau , (i+1+1/3)\tau ]\);

  • \(\zeta _i (t) \in [0,1]\) for every \(t \in [0,1]\);

  • \(\sum _{i=1}^N \zeta _i^2(t) = 1\) for every \(t \in [0,1]\).

Notice that for every time \(t \in [0,1]\) there is at most one odd index \(i_1\) and one even index \(i_2\) so that \(\zeta _i(t) = 0\) for every \(i \ne i_1, i_2\). For every \(i=1,\ldots , N\), let \(\Phi _i : [0,1] \times {\mathbb {T}}^d \rightarrow {\mathbb {T}}^d\) be the solution to
$$\begin{aligned} {\left\{ \begin{array}{ll} \partial _t \Phi _i + (u_0 \cdot \nabla ) \Phi _i = 0, \\ \Phi _i(t_i,x) = x, \end{array}\right. } \end{aligned}$$
(53)
i.e. the inverse flow map associated to the vector field \(u_0\), starting at time \(t_i\). Notice that, for fixed t, \(\Phi _i(t) : {\mathbb {T}}^d \rightarrow {\mathbb {T}}^d\) is a measure-preserving diffeomorphism.
We denote by \(R_{0,j}\) the components of \(R_0\), i.e.
$$\begin{aligned} R_0(t,x) := \sum _{j=1}^d R_{0,j}(t,x) e_j. \end{aligned}$$
Let also \(\psi : [0,1] \rightarrow {\mathbb {R}}\) be a smooth function such that \(\psi (t) \in [0,1]\) for every \(t \in [0,1]\) and
$$\begin{aligned} \psi (t) = {\left\{ \begin{array}{ll} 0, &{} \text {if } \Vert R_0(t)\Vert _{L^1({\mathbb {T}}^d)} \le \delta /8, \\ 1, &{} \text {if } \Vert R_0(t)\Vert _{L^1({\mathbb {T}}^d)} \ge \delta /4. \end{array}\right. } \end{aligned}$$
(54)
We set
$$\begin{aligned} \rho _1 := \rho _0 + \vartheta + \vartheta _c , \quad u_1 := u_0 + w, \end{aligned}$$
where \(\vartheta , \vartheta _c, w\) are defined as follows. First of all, let \(\Theta _\mu ^j\), \(W_\mu ^j\), \(j=1,\ldots , d\), be the family (depending on \(\mu \)) of Mikado densities and fields provided by Proposition 5.1, with ab chosen as in (49). We setThe factor \((D \Phi _i(t,x))^{-1}\) is the inverse matrix of \(D \Phi _i(t,x)\). Observe that for fixed \(t_0 \in [0,1]\), there are at most one odd index \(i_1\) and one even index \(i_2\) so that \(\zeta _i(t) =0\) if \(i \ne i_1, i_2\) and t is close enough to \(t_0\) (say, \(|t - t_0| \le 2\tau /3\)). Therefore for such times t we can write
$$\begin{aligned} \begin{aligned} \vartheta (t)&= \eta \, \psi (t) \left\{ \zeta _{i_1}(t) \sum _{j=1}^d R_{0,j}(t) \Theta _{\mu '}^j\big (\lambda ' \Phi _{i_1}(t) \big )\right. \\&\left. \quad + \zeta _{i_2}(t) \sum _{j=1}^d R_{0,j}(t) \Theta _{\mu ''}^j\big (\lambda '' \Phi _{i_2}(t) \big ) \right\} , \\ w(t)&= \frac{\psi (t)}{\eta } \left\{ \zeta _{i_1}(t) \sum _{j=1}^d (D \Phi _{i_1}(t))^{-1} W_{\mu '}^j \big ( \lambda ' \Phi _{i_1}(t) \big )\right. \\&\left. \quad + \zeta _{i_2}(t) \sum _{j=1}^d (D \Phi _{i_2}(t))^{-1} W_{\mu ''}^j \big ( \lambda '' \Phi _{i_2}(t) \big ) \right\} , \end{aligned} \end{aligned}$$
(56)
which we will call the fixed-time form of the perturbations. Notice that \(\vartheta \) and w are smooth functions. Notice also that \(\vartheta + \vartheta _c\) has zero mean value in \({\mathbb {T}}^d\) at each time t. Finally observe that w is a sum of terms of the form \((D \Phi )^{-1} G( \Phi )\), with
$$\begin{aligned} \Phi = \Phi _i(t), \quad G = (W_{\mu '}^j)_{\lambda '} \text { or } G = (W_{\mu ''}^j)_{\lambda ''}. \end{aligned}$$
Since \(\text {div }(W_{\mu })_\lambda = 0\) for every \(\mu , \lambda \) (see Proposition 5.1), we get from Lemma 3.2 that each one of these terms is divergence free and thus \(\text {div }w = 0\). Therefore
$$\begin{aligned} \text {div }u_1 = \text {div }u_ 0 + \text {div }w = 0. \end{aligned}$$

Remark 5.2

Observe that, thanks to the cutoff in time \(\psi \), if \(R_0(t) \equiv 0\), then
$$\begin{aligned} \vartheta (t) = \vartheta _c(t) = w(t) \equiv 0. \end{aligned}$$

5.3 Estimates on the perturbation

In this section we estimate \(\vartheta \), \(\vartheta _c\), w.

Lemma 5.3

(\(L^1\)-norm of \(\vartheta \)). For every time \(t \in [0,1]\),
$$\begin{aligned} \begin{aligned} \Vert \vartheta (t)\Vert _{L^1({\mathbb {T}}^d)}&\le \frac{M\eta }{2} \Vert R_0(t)\Vert _{L^1({\mathbb {T}}^d)} + C\Big (M, \eta , \Vert R_0\Vert _{C^1}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} \Big ) \bigg (\frac{1}{\lambda '} + \frac{1}{\lambda ''} \bigg ). \end{aligned} \end{aligned}$$

Proof

Since we have to estimate \(\Vert \vartheta (t)\Vert _{L^1({\mathbb {T}}^d)}\) for every fixed time t, we can assume that \(\vartheta (t)\) has the form (56). In (56) each term in the summation over j has the form \(f \cdot (g_\lambda \circ \Phi )\), with
$$\begin{aligned} \begin{aligned} \begin{aligned} f&= R_{0,j}(t, \cdot ) \\ \Phi&= \Phi _{i_1}(t, \cdot ) \\ g&= \Theta _{\mu '}^j \\ \lambda&= \lambda ' \\ \end{aligned}&\quad \text {or} \quad&\begin{aligned} f&= R_{0,j}(t, \cdot ), \\ \Phi&= \Phi _{i_2}(t, \cdot ), \\ g&= \Theta _{\mu ''}^j, \\ \lambda&= \lambda ''. \end{aligned} \end{aligned} \end{aligned}$$
(57)
Therefore we can apply the improved Hölder inequality, Lemma 3.4, to get
$$\begin{aligned} \begin{aligned} \Vert \vartheta (t)\Vert _{L^1}&\le \eta \Vert R_0(t)\Vert _{L^1} \sum _{j=1}^d \Vert \Theta _{\mu '}^j\Vert _{L^1} \\&\quad + \frac{C(\eta , \Vert R_0\Vert _{C^1}, \max _{i=1, \ldots , n} \Vert D \Phi _i\Vert _{C^0})}{\lambda '} \sum _{j=1}^d \Vert \Theta _{\mu '}^j\Vert _{L^1} \\&\quad + \eta \Vert R_0(t)\Vert _{L^1} \sum _{j=1}^d \Vert \Theta _{\mu ''}^j\Vert _{L^1} \\&\quad + \frac{C(\eta , \Vert R_0\Vert _{C^1}, \max _{i=1, \ldots , n} \Vert D \Phi _i\Vert _{C^0})}{\lambda ''} \sum _{j=1}^d \Vert \Theta _{\mu ''}^j\Vert _{L^1} \\ \text {(by}~50\text {)}&\le \frac{M}{2} \eta \Vert R_0(t)\Vert _{L^1} + C\Big (M, \eta , \Vert R_0\Vert _{C^1}, \max _{i=1, \ldots , n} \Vert D \Phi _i\Vert _{C^0}\Big ) \bigg ( \frac{1}{\lambda '} + \frac{1}{\lambda ''} \bigg ). \end{aligned} \end{aligned}$$
\(\square \)

Lemma 5.4

(Estimate on \(\vartheta _c\)). For every time \(t \in [0,1]\),
$$\begin{aligned} |\vartheta _c(t)| \le C\Big (M, \eta ,\Vert R_0\Vert _{C^1}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} \Big ) \bigg ( \frac{1}{\lambda '} + \frac{1}{\lambda ''} \bigg ). \end{aligned}$$

Proof

As in the proof of Lemma 5.3, we can use for \(\vartheta (t)\) the form (56) and we observe that each term in the summation over j has the form \(f \cdot (g_\lambda \circ \Phi )\), with \(f,\Phi ,g, \lambda \) as in (57). We can thus apply Lemma 3.9 to get:
$$\begin{aligned} \begin{aligned} \quad \quad \quad |\vartheta _c(t)|&\le C\Big (\eta , \Vert R_0\Vert _{C^1}, \max _{i=1, \ldots , n} \Vert D \Phi _i\Vert _{C^0}\Big ) \left[ \frac{1}{\lambda '}\sum _{j=1}^d \Vert \Theta _{\mu '}^j\Vert _{L^1} + \frac{1}{\lambda ''} \sum _{j=1}^d \Vert \Theta _{\mu ''}^j\Vert _{L^1}\right] \\ \quad \quad \quad \text {(by}~(50)\text {)}&\le C\Big (M, \eta , \Vert R_0\Vert _{C^1}, \max _{i=1, \ldots , n} \Vert D \Phi _i\Vert _{C^0}\Big ) \Bigg ( \frac{1}{\lambda '} + \frac{1}{\lambda ''} \Bigg ). \end{aligned} \end{aligned}$$
\(\square \)

Lemma 5.5

(\(C^0\) norm of w). For every time \(t \in [0,1]\),
$$\begin{aligned} \begin{aligned} \Vert w(t)\Vert _{C^0({\mathbb {T}}^d)}&\le \frac{M}{2\eta } \max _{i=1,\ldots , N} \Vert (D \Phi _i)^{-1}\Vert _{C^0 (\mathrm {supp} \ \zeta _i \times {\mathbb {T}}^d)}. \end{aligned} \end{aligned}$$

Proof

As in the proof of Lemma 5.3 we can use for w(t) the form (56). Therefore
$$\begin{aligned} \begin{aligned} \Vert w(t)\Vert _{C^0({\mathbb {T}}^d)}&\le \frac{1}{\eta } \max _{i=1,\ldots , N} \Vert (D \Phi _i)^{-1}\Vert _{C^0(\mathrm {supp} \ \zeta _i \times {\mathbb {T}}^d)} \left( \sum _{j=1}^d \Vert W_{\mu '}^j\Vert _{L^\infty ({\mathbb {T}}^d)} + \Vert W_{\mu ''}^j\Vert _{L^\infty ({\mathbb {T}}^d)} \right) \\ \text {(by}~(50)\text {)}&\le \frac{M}{2\eta } \max _{i=1,\ldots , N} \Vert (D \Phi _i)^{-1}\Vert _{C^0(\mathrm {supp} \ \zeta _i \times {\mathbb {T}}^d)}, \end{aligned} \end{aligned}$$
which is what we wanted to prove.

Lemma 5.6

(\(W^{1,p}\) norm of w). For every time \(t \in [0,1]\),
$$\begin{aligned} \Vert w(t)\Vert _{W^{1, p}({\mathbb {T}}^d)} \le C \Big (M, \eta , \max _{i=1, \ldots , N} \Vert D \Phi _i \Vert _{C^1} \Big ) \bigg (\lambda ' (\mu ')^{1 -(d-1)/p} + \lambda '' (\mu '')^{1 -(d-1)/p} \bigg ). \end{aligned}$$

Proof

As in the proof Lemma 5.5 we can use for w(t) the form (56). Taking one partial derivative \(\partial _k\), we get
$$\begin{aligned} \begin{aligned} \partial _k w(t)&= \frac{\psi (t)}{\eta } \Bigg \{ \zeta _{i_1}(t) \sum _{j=1}^d \bigg [ \partial _k (D \Phi _{i_1}(t))^{-1} W_{\mu '}^j\big (\lambda ' \Phi _{i_1}(t) \big ) \\&\quad + \lambda ' (D \Phi _{i_1}(t))^{-1} D W_{\mu '}^j \big (\lambda ' \Phi _{i_1}(t) \big ) D \Phi _{i_1}(t) e_k \bigg ] \\&\quad + \zeta _{i_2}(t) \sum _{j=1}^d \bigg [ \partial _k (D \Phi _{i_2}(t))^{-1} W_{\mu ''}^j\big (\lambda '' \Phi _{i_2}(t) \big ) \\&\quad + \lambda '' (D \Phi _{i_2}(t))^{-1} D W_{\mu ''}^j \big (\lambda '' \Phi _{i_2}(t) \big ) D \Phi _{i_2}(t) e_k \bigg ] \Bigg \}. \end{aligned} \end{aligned}$$
We now apply the classical Hölder inequality to estimate \(\Vert \partial _k w(t)\Vert _{L^{p}}\):
$$\begin{aligned} \Vert \partial _k w (t)\Vert _{L^{p}}\le & {} \frac{\psi (t)}{\eta } \Bigg \{ \max _{i=1, \ldots , N} \big \Vert D (D \Phi _i)^{-1} \big \Vert _{C^0} \bigg ( \sum \limits _{j=1}^d \Vert W_{\mu '}^j\Vert _{L^{p}} + \Vert W_{\mu ''}^j\Vert _{L^{p}} \bigg ) \\&\quad + \max _{i=1, \ldots , N} \big \Vert (D \Phi _i)^{-1}\big \Vert _{C^0} \Vert D\Phi _i\Vert _{C^0} \cdot \cdot \bigg (\lambda ' \sum \limits _{j=1}^d \Vert DW_{\mu '}^j\Vert _{L^{p}} + \lambda '' \sum \limits _{j=1}^d \Vert DW_{\mu ''}^j\Vert _{L^{p}} \bigg ) \Bigg \} \\ \text {(by Lemma}~3.1\text {)}\le & {} C \Big (\eta , \max _{i=1, \ldots , N} \Vert D \Phi _i\Vert _{C^1} \Big ) \bigg (\lambda ' \sum \limits _{j=1}^d \Vert W_{\mu '}^j\Vert _{W^{1,p}} + \lambda '' \sum \limits _{j=1}^d \Vert W_{\mu ''}^j\Vert _{W^{1,p}} \bigg ) \\ \text {(by}~ (51)\text {)}\le & {} C \Big (M, \eta , \max _{i=1, \ldots , N} \Vert D \Phi _i \Vert _{C^1} \Big ) \bigg (\lambda ' (\mu ')^{1 -(d-1)/p} + \lambda '' (\mu '')^{1 -(d-1)/p} \bigg ). \end{aligned}$$
A similar (and even easier) computation holds for \(\Vert w(t)\Vert _{L^{p}}\), thus concluding the proof of the lemma.

6 The new defect field

In this section we continue the proof of Proposition 4.1, defining the new defect field \(R_1\) and estimating it.

6.1 Definition of the new defect field

We want to define \(R_1\) so that
$$\begin{aligned} -\text {div }R_1 = \partial _t \rho _1 + \text {div }(\rho _1 u_1). \end{aligned}$$
(58)
Let us compute
$$\begin{aligned}&\partial _t \rho _1 + \text {div }(\rho _1 u_1) = \text {div }(\vartheta w - R_0) \nonumber \\&\quad + \Big [ \partial _t (\vartheta + \vartheta _c) + \text {div }\big ( (\vartheta + \vartheta _c) u_0 \big ) \Big ] \nonumber \\&\quad + \text {div }(\rho _0 w ) + \text {div }(\vartheta _c w) \nonumber \\&= \text {div }( R^{\mathrm{interaction}} + R^{\mathrm{flow}} + R^{\psi }+ R^\mathrm{quadr}) \nonumber \\&\quad + \text {div }R^{\mathrm{transport}} \nonumber \\&\quad + \text {div }R^{\mathrm{Nash}} + \text {div }R^{\mathrm{corr}} \end{aligned}$$
(59)
where we put
$$\begin{aligned} \begin{aligned} R^{\mathrm{Nash}} := \rho _0 w, \quad R^{\mathrm{corr}} := \vartheta _c w, \end{aligned} \end{aligned}$$
(60)
and \(R^{\mathrm{interaction}}\), \(R^{\mathrm{flow}}\), \(R^{\psi }\), \(R^\mathrm{quadr}\), \(R^{\mathrm{transport}}\) will be defined respectively in (63), (64), (65), (66), (70) in such a way that
$$\begin{aligned}&\text {div }(\vartheta w - R_0) = \text {div }(R^{\mathrm{interaction}} + R^\mathrm{flow} + R^{\psi } + R^{\mathrm{quadr}}), \end{aligned}$$
(61a)
$$\begin{aligned}&\partial _t (\vartheta + \vartheta _c) + \text {div }\big ((\vartheta + \vartheta _c) u_0 \big ) = \text {div }R^{\mathrm{transport}}. \end{aligned}$$
(61b)
We thus define
$$\begin{aligned} - R_1 := R^{\mathrm{interaction}} + R^{\mathrm{flow}} + R^{\psi } + R^\mathrm{quadr} + R^{\mathrm{transport}} + R^{\mathrm{Nash}} + R^{\mathrm{corr}}, \end{aligned}$$
(62)
so that (58) holds.

6.2 Definition and estimates for \(R^{\mathrm{interaction}}, R^{\mathrm{flow}}, R^{\psi }, R^{\mathrm{quadr}}\)

In this section we define and estimate the vector fields \(R^\mathrm{quadr}\), \(R^{\mathrm{interaction}}\), \(R^\psi \) and \(R^{\mathrm{flow}}\) so that (61a) holds. First of all, we want to compute more explicitly \(\text {div }(\vartheta (t)w(t) - R_0(t))\), for every fixed time t. We can use the form (56) for \(\vartheta (t)\) and w(t). Exploiting the fact that for \(j \ne k\), \(\Theta _{\mu }^j\) and \(W_{\mu }^k\) have disjoint support (see Proposition 5.1), we have
$$\begin{aligned} \begin{aligned} \vartheta (t) w(t)&= \psi ^2(t) \left\{ \zeta _{i_1}^2(t) \sum \limits _{j=1}^d R_{0,j}(t) (D \Phi _{i_1}(t))^{-1} \Theta _{\mu '}^j\big (\lambda ' \Phi _{i_1}(t) \big ) W_{\mu '}^j \big ( \lambda ' \Phi _{i_1}(t) \big )\right. \\&\quad \left. + \zeta _{i_2}^2(t) \sum \limits _{j=1}^d R_{0,j}(t) (D \Phi _{i_2}(t))^{-1} \Theta _{\mu ''}^j\big (\lambda '' \Phi _{i_2}(t) \big ) W_{\mu ''}^j \big ( \lambda '' \Phi _{i_2}(t) \big ) \right\} \\&\quad + R^{\mathrm{interaction}}(t), \end{aligned} \end{aligned}$$
where we set
$$\begin{aligned}&R^{\mathrm{interaction}}(t) := \psi ^2(t) \zeta _{i_1}(t) \zeta _{i_2}(t) \nonumber \\&\quad \sum \limits _{j,k=1}^d\bigg [ R_{0,j}(t) (D\Phi _{i_2}(t))^{-1} \Theta _{\mu '}^j \big (\lambda ' \Phi _{i_1}(t) \big ) W_{\mu ''}^k \big (\lambda '' \Phi _{i_2}(t) \big ) \nonumber \\&\qquad + R_{0,j}(t) (D \Phi _{i_1}(t))^{-1} \Theta _{\mu ''}^j\big (\lambda '' \Phi _{i_2}(t) \big ) W_{\mu '}^k\big (\lambda ' \Phi _{i_1}(t) \big ) \bigg ]. \end{aligned}$$
(63)
On the other side, using the fact that \(\sum _{i=1}^N \zeta _i^2 \equiv 1\), we can write
$$\begin{aligned} \begin{aligned} R_0(t)&= \psi ^2(t) R_0(t) + R_0(t) \big [ 1 - \psi ^2(t) \big ] \\&= \psi ^2(t) R_0(t) - R^{\psi }(t) \\&\quad \text {(using that} ~ {\textstyle \sum \nolimits _{i=1}^N \zeta _i^2 \equiv 1}\text {)} \\&= \psi ^2(t) \Big \{ \zeta _{i_1}^2(t) R_0(t) + \zeta _{i_2}^2(t) R_0(t) \Big \} - R^{\psi }(t) \\&= \psi ^2(t) \Big \{ \zeta _{i_1}^2(t) \big [D\Phi _{i_1}(t)\big ]^{-1} R_0(t) + \zeta _{i_2}^2(t) \big [D \Phi _{i_2}(t)\big ]^{-1} R_0(t) \Big \} - R^{\mathrm{flow}}(t) - R^{\psi }(t) \\&= \psi ^2(t) \left\{ \zeta _{i_1}^2(t) \sum _{j=1}^d R_{0,j}(t) \big [D\Phi _{i_1}(t)\big ]^{-1} e_j + \zeta _{i_2}^2(t) \sum _{j=1}^d R_{0,j}(t) \big [D \Phi _{i_2}(t)\big ]^{-1} e_j \right\} \\&\qquad - R^{\mathrm{flow}}(t) - R^{\psi }(t), \end{aligned} \end{aligned}$$
where we set
$$\begin{aligned} \begin{aligned} - R ^{\mathrm{flow}}(t)&:= \psi ^2(t) \bigg \{ \zeta _{i_1}^2(t) \Big [ \mathrm{Id} - (D\Phi _{i_1}(t))^{-1} \Big ] R_0(t) \\&\qquad + \zeta _{i_2}^2(t) \Big [ \mathrm{Id} - (D\Phi _{i_2}(t))^{-1} \Big ] R_0(t) \bigg \} \end{aligned} \end{aligned}$$
(64)
with \(\mathrm{Id}\) being the identity matrix, and
$$\begin{aligned} - R^{\psi }(t) := R_0(t) \big [ 1 - \psi ^2(t) \big ]. \end{aligned}$$
(65)
Summarizing, we have
$$\begin{aligned} \text {div }&\big ( \vartheta (t) w(t) - R_0(t) \big ) \\&= \text {div }R^{\mathrm{interaction}}(t) + \text {div }R^{\mathrm{flow}}(t) + \text {div }R^{\psi }(t) \\&\quad + \psi ^2(t) \left\{ \zeta _{i_1}^2(t) \sum _{j=1}^d \text {div }\bigg [ R_{0,j} \big [ D \Phi _{i_1}(t) \big ]^{-1} \Big ( \Theta _{\mu '}^j \big (\lambda ' \Phi _{i_1}(t) \big ) W_{\mu '}^j \big ( \lambda ' \Phi _{i_1}(t) \big ) - e_j \Big ) \bigg ] \right. \\&\quad \left. + \zeta _{i_2}^2(t) \sum _{j=1}^d \text {div }\bigg [ R_{0,j} \big [ D \Phi _{i_2}(t) \big ]^{-1} \Big ( \Theta _{\mu ''}^j \big (\lambda '' \Phi _{i_2}(t) \big ) W_{\mu ''}^j \big ( \lambda '' \Phi _{i_2}(t) \big ) - e_j \Big ) \bigg ] \right\} \\&= \text {div }R^{\mathrm{interaction}}(t) + \text {div }R^{\mathrm{flow}}(t) + \text {div }R^{\psi }(t) \\&\quad + \psi ^2(t) \left\{ \zeta _{i_1}^2(t) \sum _{j=1}^d \nabla R_{0,j} \cdot \big [ D \Phi _{i_1}(t) \big ]^{-1} \Big ( \Theta _{\mu '}^j \big (\lambda ' \Phi _{i_1}(t) \big ) W_{\mu '}^j \big ( \lambda ' \Phi _{i_1}(t) \big ) - e_j \Big ) \right. \\&\quad \left. + \zeta _{i_2}^2(t) \sum _{j=1}^d \nabla R_{0,j} \cdot \big [ D \Phi _{i_2}(t) \big ]^{-1} \Big ( \Theta _{\mu ''}^j \big (\lambda '' \Phi _{i_2}(t) \big ) W_{\mu ''}^j \big ( \lambda '' \Phi _{i_2}(t) \big ) - e_j \Big ) \right\} , \end{aligned}$$
where in the last equality we used the fact that \(\text {div }((\Theta _\mu ^j)_\lambda (W_\mu ^j)_\lambda - e_j) = 0\) for every \(\mu , \lambda , j\) (see Proposition 5.1) and Lemma 3.2. We now observe that each term in the two summations over j has zero mean value (being a divergence) and it has the form \(f (D \Phi )^{-1} (g_\lambda \circ \Phi )\), for
$$\begin{aligned} \begin{aligned} \begin{aligned} f&= \nabla R_{0,j}(t, \cdot ) \\ \Phi&= \Phi _{i_1}(t, \cdot ) \\ g&= \Theta _{\mu '}^j W_{\mu '}^j - e_j \\ \lambda&= \lambda ' \end{aligned}&\quad \text { or } \quad&\begin{aligned} f&= \nabla R_{0,j}(t, \cdot ), \\ \Phi&= \Phi _{i_2}(t, \cdot ), \\ g&= \Theta _{\mu ''}^j W_{\mu ''}^j - e_j, \\ \lambda&= \lambda ''. \end{aligned} \end{aligned} \end{aligned}$$
We can therefore apply Lemma 3.6 and define
$$\begin{aligned} \begin{aligned}&R ^{\mathrm{quadr}}(t) \\&\quad := \psi ^2(t) \left\{ \zeta _{i_1}^2(t) \sum _{j=1}^d {\mathcal {R}} \bigg (\nabla R_{0,j} \cdot \big [ D \Phi _{i_1}(t) \big ]^{-1} \Big ( \Theta _{\mu '}^j \big (\lambda ' \Phi _{i_1}(t) \big ) W_{\mu '}^j \big ( \lambda ' \Phi _{i_1}(t) \big ) - e_j \Big ) \bigg ) \right. \\&\qquad \quad \left. + \zeta _{i_2}^2(t) \sum _{j=1}^d {\mathcal {R}} \bigg ( \nabla R_{0,j} \cdot \big [ D \Phi _{i_2}(t) \big ]^{-1} \Big ( \Theta _{\mu ''}^j \big (\lambda '' \Phi _{i_2}(t) \big ) W_{\mu ''}^j \big ( \lambda '' \Phi _{i_2}(t) \big ) - e_j \Big ) \bigg ) \right\} , \end{aligned}\nonumber \\ \end{aligned}$$
(66)
so that (61a) holds. We now separately estimate \(R^\mathrm{interaction}\), \(R^{\mathrm{flow}}\), \(R^{\psi }\), \(R^{\mathrm{quadr}}\). We start with \(R^{\mathrm{interaction}}\).

Lemma 6.1

For every time t it holds
$$\begin{aligned} \begin{aligned} \Vert R^{\mathrm{interaction}} (t)\Vert _{L^1({\mathbb {T}}^d)}&\le C \Big (M, \Vert R_0\Vert _{C^0}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} \Big )\\&\quad \bigg ( \frac{1}{(\mu ')^{d-1}} + \frac{1}{(\mu '')^{d-1}} + \frac{\lambda ' \mu '}{\lambda ''} + \frac{\lambda ' (\mu ')^d}{\lambda '' (\mu '')^{d-1}} \bigg ). \end{aligned} \end{aligned}$$

Proof

Consider the definition (63) of \(R^\mathrm{interaction}\). We start by estimating \(\Vert \Theta _{\mu '}^j (\lambda ' \Phi _{i_1}(t))\Vert _{C^1}\) and \(\Vert W_{\mu '}^k(\lambda ' \Phi _{i_1}(t))\Vert _{C^1}\), using (52) and the chain rule
$$\begin{aligned} \begin{aligned} \Vert \Theta _{\mu '}^j (\lambda ' \Phi _{i_1}(t))\Vert _{C^1}&\le C(M, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} ) \lambda ' (\mu ')^d, \\ \Vert W_{\mu '}^k (\lambda ' \Phi _{i_1}(t))\Vert _{C^1}&\le C(M, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} ) \lambda ' \mu '. \end{aligned} \end{aligned}$$
(67)
We now estimate \(\Theta _{\mu '}^j (\lambda ' \Phi _{i_1}(t) ) W_{\mu ''}^k (\lambda '' \Phi _{i_2}(t) )\), using the improved Hölder inequality, Lemma 3.4 and considering \(\lambda ''\) as the fast oscillation. We have
$$\begin{aligned} \begin{aligned}&\big \Vert \Theta _{\mu '}^j \big (\lambda ' \Phi _{i_1}(t) \big ) W_{\mu ''}^k \big (\lambda '' \Phi _{i_2}(t) \big ) \big \Vert _{L^1} \\&\quad \le \Vert \Theta _{\mu '}^j \Vert _{L^1} \Vert W_{\mu ''}^k \Vert _{L^1} + \frac{1}{\lambda ''} \big \Vert \Theta _{\mu '}^j \big (\lambda ' \Phi _{i_1}(t) \big )\big \Vert _{C^1} \big \Vert D \Phi _{i_2}(t)\big \Vert ^{d-1}_{C^0} \big \Vert W_{\mu ''}^k \big \Vert _{L^1} \\&\quad \le C\Big (M, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} \Big ) \bigg (\frac{1}{(\mu '')^{d-1}} + \frac{\lambda ' (\mu ')^d}{\lambda '' (\mu '')^{d-1}} \bigg ), \end{aligned} \end{aligned}$$
where in the last inequality we used (50), (51) and (67). A similar estimate holds for \(\Theta _{\mu ''}^j (\lambda '' \Phi _{i_2}(t)) W_{\mu '}^k (\lambda ' \Phi _{i_1}(t))\):
$$\begin{aligned} \begin{aligned}&\big \Vert \Theta _{\mu ''}^j (\lambda '' \Phi _{i_2}(t)) W_{\mu '}^k (\lambda ' \Phi _{i_1}(t)) \big \Vert _{L^1} \\&\quad \le \Vert \Theta _{\mu ''}^j \Vert _{L^1} \Vert W_{\mu '}^k \Vert _{L^1} + \frac{1}{\lambda ''} \Vert W_{\mu '}^k (\lambda ' \Phi _{i_1}(t)) \Vert _{C^1} \Vert D \Phi _{i_2}(t)\Vert ^{d-1}_{C^0} \Vert \Theta _{\mu ''}^j \Vert _{L^1} \\&\quad \le C\Big (M, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} \Big ) \bigg ( \frac{1}{(\mu ')^{d-1}} + \frac{\lambda ' \mu '}{\lambda ''} \bigg ). \end{aligned} \end{aligned}$$
Therefore
$$\begin{aligned} \begin{aligned}&\Vert R^{\mathrm{interaction}}(t)\Vert _{L^1} \\&\quad \le C \Big ( M, \Vert R_0\Vert _{C^0}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} \Big ) \cdot \\&\qquad \cdot \sum _{j,k=1}^d \bigg [ \big \Vert \Theta _{\mu '}^j \big (\lambda ' \Phi _{i_1}(t) \big ) W_{\mu ''}^k \big (\lambda '' \Phi _{i_2}(t) \big ) \big \Vert _{L^1} + \big \Vert \Theta _{\mu ''}^j (\lambda '' \Phi _{i_2}(t)) W_{\mu '}^k (\lambda ' \Phi _{i_1}(t)) \big \Vert _{L^1} \bigg ] \\&\quad \le C \big (M, \Vert R_0\Vert _{C^0}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} \big ) \bigg ( \frac{1}{(\mu ')^{d-1}} + \frac{1}{(\mu '')^{d-1}} + \frac{\lambda ' \mu '}{\lambda ''} + \frac{\lambda ' (\mu ')^d}{\lambda '' (\mu '')^{d-1}} \bigg ). \end{aligned} \end{aligned}$$
\(\square \)

Lemma 6.2

For every \(t \in [0,1]\),
$$\begin{aligned} \Vert R^{\mathrm{flow}}(t)\Vert _{L^1({\mathbb {T}}^d)} \le \Vert R_0\Vert _{C^0} \max _{i=1,\ldots , N} \big \Vert \mathrm{Id} - D\Phi _i(t)^{-1} \big \Vert _{C^0(\mathrm {supp} \ \zeta _i \times {\mathbb {T}}^d)}. \end{aligned}$$

Proof

The proof follows immediately from the definition of \(R^{\mathrm{flow}}\).

Lemma 6.3

For every \(t \in [0,1]\),
$$\begin{aligned} \Vert R^\psi (t) \Vert _{L^1({\mathbb {T}}^d)} \le \delta /4. \end{aligned}$$

Proof

If \(\psi ^2(t) \ne 1\), then, by (54), \(\Vert R_0(t)\Vert _{L^1} \le \delta /4\) and thus the conclusion follows.

Lemma 6.4

For every \(t \in [0,1]\),
$$\begin{aligned} \Vert R^{\mathrm{quadr}}(t)\Vert _{L^1({\mathbb {T}}^d)} \le C \Big (M, \Vert R_0\Vert _{C^2}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^1} \Big ) \bigg ( \frac{1}{\lambda '} + \frac{1}{\lambda ''} \bigg ). \end{aligned}$$

Proof

\(R^{\mathrm{quadr}}(t)\) is defined in (66) using Lemma 3.6. Applying the bounds provided by such proposition, with \(k=0\) and \(p=1\), we get
$$\begin{aligned} \begin{aligned} \Vert R^{\mathrm{quadr}}(t)\Vert _{L^1}&\le C \Big ( \Vert R_0\Vert _{C^2}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^1} \Big )\\&\quad \left( \frac{1}{\lambda '} \sum _{j=1}^d \Vert \Theta _{\mu '}^j W_{\mu '}^j - e_j\Vert _{L^1} + \frac{1}{\lambda ''} \sum _{j=1}^d \Vert \Theta _{\mu ''}^j W_{\mu ''}^j - e_j\Vert _{L^1}\right) \text {(by}~ (50)\text {)} \\&\le C \Big (M, \Vert R_0\Vert _{C^2}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^1} \Big ) \bigg ( \frac{1}{\lambda '} + \frac{1}{\lambda ''} \bigg ), \end{aligned} \end{aligned}$$
thus concluding the proof of the lemma.

6.3 Definition and estimates for \(R^{\mathrm{transport}}\)

In this section we define and estimate the vector fields \(R^\mathrm{transport}\) so that (61b) holds. First of all, we want to compute more explicitly \(\partial _t (\vartheta (t) + \vartheta _c(t)) + \text {div }((\vartheta (t)+\vartheta _c(t)) u_0(t))\), for every fixed time t. We can use the fixed-time form (56) for \(\vartheta (t)\) and w(t). We have
$$\begin{aligned} \partial _t&(\vartheta + \vartheta _c) + \text {div }((\vartheta + \vartheta _c)u_0) \nonumber \\&= {{\dot{\vartheta }}}_c + \sum _{j=1}^d \bigg \{ A^j_1 \Theta _{\mu '}^j \big (\lambda ' \Phi _{i_1} \big ) + A^j_2 \Theta _{\mu ''}^j \big (\lambda '' \Phi _{i_2} \big ) \nonumber \\&\quad + \lambda ' B^j_1 \cdot \Big [ \partial _t \Phi _{i_1} + (u_0 \cdot \nabla ) \Phi _{i_1} \Big ] + \lambda '' B^j_2 \cdot \Big [ \partial _t \Phi _{i_2} + (u_0 \cdot \nabla ) \Phi _{i_2} \Big ] \bigg \} \nonumber \\&= {{\dot{\vartheta }}}_c + \sum _{j=1}^d \bigg \{ A^j_1 \Theta _{\mu '}^j \big (\lambda ' \Phi _{i_1} \big ) + A^j_2 \Theta _{\mu ''}^j \big (\lambda '' \Phi _{i_2} \big ) \bigg \}, \end{aligned}$$
(68)
where
$$\begin{aligned} A_1^j&:= \eta \, \bigg [{{\dot{\psi }}} \zeta _{i_1} R_{0,j} + \psi \dot{\zeta }_{i_1} R_{0,j} + \psi \zeta _{i_1} \Big ( \partial _t R_{0,j} + \nabla R_{0,j} \cdot u_0 \Big ) \bigg ], \\ A_2^j&:= \eta \, \bigg [{{\dot{\psi }}} \zeta _{i_2} R_{0,j} + \psi \dot{\zeta }_{i_2} R_{0,j} + \psi \zeta _{i_2} \Big ( \partial _t R_{0,j} + \nabla R_{0,j} \cdot u_0 \Big ) \bigg ], \\ B_1^j&:= \eta \, \psi \zeta _{i_1} R_{0,j} \, \nabla \Theta _{\mu '}^j(\lambda ' \Phi _{i_1}), \\ B_2^j&:= \eta \, \psi \zeta _{i_2} R_{0,j} \, \nabla \Theta _{\mu ''}^j(\lambda '' \Phi _{i_2}), \end{aligned}$$
and we used (53). Here we used the notation \(\dot{f} = df/dt\), if \(f=f(t)\) is a function depending only on time. We now continue the chain of equalities in (68), by adding and subtracting the mean value of each term in the summations over j, as follows:The last equality is a consequence of the fact thatWe now observe that, at the fixed time t, each term in the last line in (69) has the form Open image in new window for
$$\begin{aligned} \begin{aligned} f&= A_1^j(t,\cdot )&\text { or }&f&= A_2^j(t,\cdot ), \\ \Phi&= \Phi _{i_1}(t, \cdot )&\text { or }&\Phi&= \Phi _{i_2}(t, \cdot ), \\ g&= \Theta _{\mu '}^j&\text { or }&g&= \Theta _{\mu ''}^j, \\ \lambda&= \lambda '&\text { or }&\lambda&= \lambda ''. \end{aligned} \end{aligned}$$
Since \(\Theta _\mu ^j\) has zero mean value (see Proposition 5.1), we can apply Lemma 3.6 and define

Lemma 6.5

For every \(t \in [0,1]\), it holds
$$\begin{aligned} \Vert R^{\mathrm{transport}}(t)\Vert _{L^1({\mathbb {T}}^d)} \le C \big (M, \eta , \delta , \tau , \Vert R_0\Vert _{C^2}, \Vert u_0\Vert _{C^1}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^1} \big ) \bigg ( \frac{1}{\lambda '} + \frac{1}{\lambda ''} \bigg ). \end{aligned}$$

Proof

First of all, we observe that
$$\begin{aligned} |\psi '(t)| \le C\big (\delta , \Vert R_0\Vert _{C^1} \big ), \qquad |\zeta _i'(t)| \le C(\tau ) \text { for all } i=1,\ldots , N. \end{aligned}$$
Therefore
$$\begin{aligned} \Vert A_1^j(t)\Vert _{C^1}, \Vert A_2^j(t)\Vert _{C^1} \le C \big ( \eta , \delta , \tau , \Vert R_0\Vert _{C^2}, \Vert u_0\Vert _{C^1} \big ). \end{aligned}$$
We defined \(R^{\mathrm{transport}}\) in (70) using the antidivergence operator provided by Lemma 3.6. We can thus apply the bounds provided by such proposition, with \(k=0\) and \(p=1\), to get
$$\begin{aligned} \begin{aligned} \Vert&R^{\mathrm{transport}}(t)\Vert _{L^1} \\&\le \sum _{j=1}^d \left\{ \frac{\Vert A_1^j(t)\Vert _{C^1} \Vert D\Phi _{i_1}\Vert _{C^1}^{d-1} \Vert \Theta _{\mu '}^j\Vert _{L^1}}{\lambda '} + \frac{\Vert A_2^j(t)\Vert _{C^1} \Vert D\Phi _{i_2}\Vert _{C^1}^{d-1} \Vert \Theta _{\mu ''}^j\Vert _{L^1}}{\lambda ''} \right\} \\&\le C \big (M, \eta , \delta , \tau , \Vert R_0\Vert _{C^2}, \Vert u_0\Vert _{C^1}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^1} \big ) \bigg ( \frac{1}{\lambda '} + \frac{1}{\lambda ''} \bigg ), \end{aligned} \end{aligned}$$
where in the last line we used (50).

6.4 Estimates for \(R^{\mathrm{Nash}}\) and \(R^{\mathrm{corr}}\)

In this section we estimate \(R^{\mathrm{Nash}}\) and \(R^{\mathrm{corr}}\).

Lemma 6.6

For every \(t \in [0,1]\),
$$\begin{aligned} \Vert R^{\mathrm{Nash}}(t)\Vert _{L^1({\mathbb {T}}^d)} \le C\big (\Vert \rho _0(t)\Vert _{C^0}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} \big ) \bigg ( \frac{1}{(\mu ')^{d-1}} + \frac{1}{(\mu '')^{d-1}} \bigg ). \end{aligned}$$

Proof

We have
$$\begin{aligned} \begin{aligned} \Vert R^{\mathrm{Nash}}(t)\Vert _{L^1}&= \Vert \rho _0(t) w(t)\Vert _{L^1} \\&\le \Vert \rho _0(t)\Vert _{C^0} \Vert w(t)\Vert _{L^1} \\ \text {(by Lemma}~3.1\text {)}&\le C\big (\eta , \Vert \rho _0(t)\Vert _{C^0}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} \big ) \sum _{j=1}^d \bigg ( \Vert W_{\mu '}^j\Vert _{L^1} + \Vert W_{\mu ''}^j\Vert _{L^1} \bigg ) \\ \text {(by}~(51)\text {)}&\le C\big (\eta , \Vert \rho _0(t)\Vert _{C^0}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} \big ) \bigg ( \frac{1}{(\mu ')^{d-1}} + \frac{1}{(\mu '')^{d-1}} \bigg ). \end{aligned} \end{aligned}$$
\(\square \)

Lemma 6.7

For every \(t \in [0,1]\),
$$\begin{aligned} \Vert R^{\mathrm{corr}}(t)\Vert _{L^1({\mathbb {T}}^d)} \le C(M, \eta , \Vert R_0\Vert _{C^1}, \max _{i=1, \ldots , N} \Vert D \Phi _i\Vert _{C^0} ) \bigg (\frac{1}{\lambda '} + \frac{1}{\lambda ''} \bigg ). \end{aligned}$$

Proof

We use Lemma 5.4 and Lemma 5.5:
$$\begin{aligned} \begin{aligned} \Vert \vartheta _c(t) w(t)\Vert _{L^1}&= |\vartheta _c(t)| \Vert w(t)\Vert _{L^1} \\&\le |\vartheta _c(t)| \Vert w(t)\Vert _{C^0} \\&\le C(M, \eta , \Vert R_0\Vert _{C^1}, \max _{i=1, \ldots , N} \Vert D \Phi _i\Vert _{C^0} ) \bigg (\frac{1}{\lambda '} + \frac{1}{\lambda ''} \bigg ). \end{aligned} \end{aligned}$$
\(\square \)

7 Proof of Proposition 4.1

In this section we conclude the proof of Proposition 4.1, and thus also the proof of Theorem 1.2 and, consequently, the proof of Theorem 1.1. We first prove that if \(R_0(t) = 0\) at some time \(t \in [0,1]\), then \(R_1(t) = 0\). Observe that if \(R_0(t) = 0\), then by Remark 5.2,
$$\begin{aligned} \rho _1(t) - \rho _0(t) = \vartheta (t) + \vartheta _c(t) = 0, \qquad u_1(t) - u_0(t) = w(t) = 0. \end{aligned}$$
Moreover, by (54), \(\psi \equiv 0\) on a neighborhood of t and thus \(\psi (t) = \psi '(t) = 0\). Therefore
$$\begin{aligned}&\psi (t) = 0 \quad \Longrightarrow R^{\mathrm{interaction}}(t) = R^{\mathrm{flow}}(t) = R^{\mathrm{quadr}}(t) = 0, \\&\psi (t) = \psi '(t) = 0 \quad \Longrightarrow \quad R^\mathrm{transport}(t) = 0, \\&R_0(t) = 0 \quad \Longrightarrow \quad R^\psi (t) = 0, \\&w(t) = w_c(t) = 0 \quad \Longrightarrow \quad R^{\mathrm{Nash}}(t) = R^\mathrm{corr}(t) = 0, \end{aligned}$$
and thus \(R_1(t) = 0\).
We now prove estimates (39a)–(39d). First of all, in view of Lemma 5.5 and Lemma 6.2, we choose \(\tau \) so small that
$$\begin{aligned}&\max _{i=1,\ldots , N} \Vert D \Phi _i^{-1}\Vert _{C^0 (\mathrm {supp} \ \zeta _i \times {\mathbb {T}}^d)} \le 2, \end{aligned}$$
(71a)
$$\begin{aligned} \Vert R_0\Vert _{C^0}&\max _{i=1,\ldots , N} \Vert \mathrm{Id} - D \Phi _i^{-1}\Vert _{C^0 (\mathrm {supp} \ \zeta _i \times {\mathbb {T}}^d)} \le \frac{\delta }{4}. \end{aligned}$$
(71b)
This is always possible since, by (53), \(\Phi _i(t_i, x) = x\) and thus \(D \Phi _i(t_i, x) = \mathrm{Id}\) for every \(i=1,\ldots , N\). We choose also \(\lambda ', \mu ', \lambda '', \mu ''\) such that \(1 \ll \lambda ' \ll \mu ' \ll \lambda '' \ll \mu ''\). More precisely, we set
$$\begin{aligned} \lambda ' = \lambda , \quad \mu ' := \lambda ^\alpha , \quad \lambda '' := \lambda ^\beta , \quad \mu '' := \lambda ^\gamma , \end{aligned}$$
for some
$$\begin{aligned} 1< \alpha< \beta < \gamma \end{aligned}$$
and \(\lambda \gg 1\) to be fixed later.
1 Estimate (39a). If \(R_0(t) = 0\), we have already seen that \(\rho _1(t) = \rho _0(t)\). We can thus assume \(R_0(t) \ne 0\). We have
$$\begin{aligned} \begin{aligned} \Vert \rho _1(t) - \rho _0(t)\Vert _{L^1}&\le \Vert \vartheta _0(t)\Vert _{L^1} + |\vartheta _c(t)| \\&\quad \text {(by Lemmas}~5.3 \, \text {and}~5.4\text {)} \\&\le \frac{M\eta }{2} \Vert R_0(t)\Vert _{L^1} + C \Big (M, \eta , \Vert R_0\Vert _{C^1}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} \Big ) \bigg ( \frac{1}{\lambda '} + \frac{1}{\lambda ''} \bigg ) \\&\le \frac{M\eta }{2} \Vert R_0(t)\Vert _{L^1} + C \Big (M, \eta , \Vert R_0\Vert _{C^1}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^0} \Big ) \bigg ( \frac{1}{\lambda } + \frac{1}{\lambda ^\beta } \bigg ) \\&\le M \eta \Vert R_0(t)\Vert _{L^1}, \end{aligned} \end{aligned}$$
if the constant \(\lambda \) is chosen large enough.
2 Estimate (39b). We have
$$\begin{aligned} \begin{aligned} \Vert u_1(t) - u_0(t)\Vert _{C^0}&\le \Vert w(t)\Vert _{C^0} \\ \text {(by Lemma}~5.5\text {)}&\le \frac{M}{2\eta } \max _{i=1,\ldots , N} \Vert (D \Phi _i)^{-1}\Vert _{C^0 (\mathrm {supp} \ \zeta _i \times {\mathbb {T}}^d)} \\ \text {(by}\, (\mathrm{71a})\text {)}&\le \frac{M}{\eta }. \end{aligned} \end{aligned}$$
3 Estimate (39c). We have
$$\begin{aligned} \begin{aligned} \Vert u_1(t) - u_0(t)\Vert _{W^{1, p}}&\le \Vert w(t)\Vert _{W^{1, p}}\\&\quad \text {(by Lemma}~5.6\text {)} \\&\le C \Big (M, \eta , \max _{i=1, \ldots , N} \Vert D \Phi _i \Vert _{C^1} \Big ) \bigg (\lambda ' (\mu ')^{1 -(d-1)/p} + \lambda '' (\mu '')^{1 -(d-1)/p} \bigg ) \\&\le C \Big (M, \eta , \max _{i=1, \ldots , N} \Vert D \Phi _i \Vert _{C^1} \Big ) \bigg ( \lambda ^{1 + \alpha (1 - (d-1)/p)} + \lambda ^{\beta + \gamma (1 - (d-1)/p)} \bigg ) \\&\le \delta , \end{aligned} \end{aligned}$$
if \(\alpha \), \(\beta , \gamma \) are chosen so that
$$\begin{aligned} 1 + \alpha \bigg ( 1 - \frac{d-1}{p} \bigg )&< 0, \end{aligned}$$
(72a)
$$\begin{aligned} \beta + \gamma \bigg ( 1 - \frac{d-1}{p} \bigg )&< 0, \end{aligned}$$
(72b)
and \(\lambda \) is large enough.
4 Estimate (39d). Recall the definition of \(R_1\) in (62). Using Lemmas 6.16.2, 6.3, 6.4, 6.5, 6.6, 6.7 and (71b), we get
$$\begin{aligned} \begin{aligned} \Vert R_1(t)\Vert _{L^1}&\le \Vert R^{\mathrm{interaction}}(t)\Vert _{L^1} + \Vert R^\mathrm{flow}(t)\Vert _{L^1} + \Vert R^\psi (t)\Vert _{L^1} + \Vert R^{\mathrm{quadr}}(t)\Vert _{L^1}\\&\quad + \Vert R^{\mathrm{transport}}(t)\Vert _{L^1} + \Vert R^{\mathrm{Nash}}(t)\Vert _{L^1} + \Vert R^{\mathrm{corr}}(t)\Vert _{L^1} \\&\le \frac{\delta }{2} + C \Big (M, \eta , \delta , \tau , \Vert \rho _0\Vert _{C^0}, \Vert u_0\Vert _{C^1}, \Vert R_0\Vert _{C^2}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^1} \Big ) \cdot \\&\quad \cdot \Bigg [ \frac{1}{\lambda '} + \frac{1}{\lambda ''} + \frac{1}{(\mu ')^{d-1}} + \frac{1}{(\mu '')^{d-1}} + \frac{\lambda ' \mu '}{\lambda ''} + \frac{\lambda ' (\mu ')^d}{\lambda '' (\mu '')^{d-1}} \Bigg ] \\&\le \frac{\delta }{2} + C \Big (M, \eta , \delta , \tau , \Vert \rho _0\Vert _{C^0}, \Vert u_0\Vert _{C^1}, \Vert R_0\Vert _{C^2}, \max _{i=1,\ldots , N} \Vert D \Phi _i\Vert _{C^1} \Big ) \cdot \\&\quad \cdot \Bigg [ \frac{1}{\lambda } + \frac{1}{\lambda ^\beta } + \frac{1}{\lambda ^{\alpha (d-1)}} + \frac{1}{\lambda ^{\gamma (d-1)}} + \frac{\lambda ^{1 + \alpha }}{\lambda ^\beta } + \frac{\lambda ^{1 + \alpha d}}{\lambda ^{\beta + \gamma (d-1)}} \Bigg ] \\&\le \frac{\delta }{2} + \frac{\delta }{2} \le \delta , \end{aligned} \end{aligned}$$
if
$$\begin{aligned} 1 + \alpha - \beta&< 0, \end{aligned}$$
(73a)
$$\begin{aligned} 1 + \alpha d - \beta - \gamma (d-1)&< 0, \end{aligned}$$
(73b)
and \(\lambda \) is large enough.
We still have to choose \(\alpha , \beta , \gamma \) so that (72), (73) are satisfied. This can be easily done as follows, recalling that \(p < d-1\). First we fix \(\alpha > 1\) so that
$$\begin{aligned} \alpha > \frac{1}{\frac{d-1}{p} - 1}, \end{aligned}$$
so that (72a) is satisfied. Then we choose \(\beta \) so that
$$\begin{aligned} \beta > 1 + \alpha , \end{aligned}$$
so that (73a) is satisfied. Finally we choose \(\gamma >1\) so that
$$\begin{aligned} \gamma> \frac{\beta }{\frac{d-1}{p} - 1} \text { and } \gamma > \frac{1 + \alpha d - \beta }{d-1}, \end{aligned}$$
so that (72b) and (73b) are satisfied. This concludes the proof of Proposition 4.1 and thus also the proof of Theorem 1.2 and, consequently, the proof of Theorem 1.1.

Footnotes

  1. 1.

    While the present paper was under review, G. Sattig and the first author, relying on the ideas introduced in the present paper, showed in [15] that this dimensional threshold can be raised up to d.

Notes

Acknowledgements

This research was supported by the ERC Grant Agreement No. 724298.

References

  1. 1.
    Ambrosio, L.: Transport equation and Cauchy problem for BV vector fields. Invent. Math. 158(2), 227–260 (2004)MathSciNetCrossRefGoogle Scholar
  2. 2.
    Ambrosio, L.: Well posedness of ODE’s and continuity equations with nonsmooth vector fields, and applications. Rev. Mat. Complut. 30(3), 427–450 (2017)MathSciNetCrossRefGoogle Scholar
  3. 3.
    Bianchini, S., Bonicatto, P.: A uniqueness result for the decomposition of vector fields in Rd. SISSA (2017)Google Scholar
  4. 4.
    Buckmaster, T., De Lellis, C., Isett, P., Székelyhidi Jr., L.: Anomalous dissipation for 1/5-Hölder Euler flows. Ann. Math. (2) 182(1), 127–172 (2015)MathSciNetCrossRefGoogle Scholar
  5. 5.
    Buckmaster, T., De Lellis, C., Székelyhidi Jr., L., Vicol, V.: Onsager’s conjecture for admissible weak solutions. Commun. Pure Appl. Math. 72, 229–274 (2019)MathSciNetCrossRefGoogle Scholar
  6. 6.
    Buckmaster, T., Vicol, V.: Nonuniqueness of weak solutions to the Navier–Stokes equation. Ann. Math. (2) 189(1), 101–144 (2019)MathSciNetCrossRefGoogle Scholar
  7. 7.
    Caravenna, L., Crippa, G.: Uniqueness and Lagrangianity for solutions with lack of integrability of the continuity equation. C. R. Math. Acad. Sci. Paris 354(12), 1168–1173 (2016)MathSciNetCrossRefGoogle Scholar
  8. 8.
    Daneri, S., Székelyhidi Jr., L.: Non-uniqueness and h-principle for Hölder-continuous weak solutions of the Euler equations. Arch. Ration. Mech. Anal. 224(2), 471–514 (2017)MathSciNetCrossRefGoogle Scholar
  9. 9.
    De Lellis, C., Székelyhidi Jr., L.: On admissibility criteria for weak solutions of the Euler equations. Arch. Ration. Mech. Anal. 195(1), 225–260 (2008)MathSciNetCrossRefGoogle Scholar
  10. 10.
    Depauw, N.: Non unicité des solutions bornées pour un champ de vecteurs BV en dehors d’un hyperplan. C. R. Math. Acad. Sci. Paris 337(4), 249–252 (2003)MathSciNetCrossRefGoogle Scholar
  11. 11.
    DiPerna, R.J., Lions, P.-L.: Ordinary differential equations, transport theory and Sobolev spaces. Invent. Math. 98(3), 511–547 (1989)MathSciNetCrossRefGoogle Scholar
  12. 12.
    Evans, L.C.: Partial Differential Equations. American Mathematical Society, Providence (2010)zbMATHGoogle Scholar
  13. 13.
    Isett, P.: A proof of Onsager’s conjecture. Ann. Math. (2) 188(3), 871–963 (2018)MathSciNetCrossRefGoogle Scholar
  14. 14.
    Isett, P., Oh, S.-J.: On nonperiodic Euler flows with Hölder regularity. Arch. Ration. Mech. Anal. 221(2), 725–804 (2016)MathSciNetCrossRefGoogle Scholar
  15. 15.
    Modena, S., Sattig, G.: Convex integration solutions to the transport equation with full dimensional concentration. arXiv (2019)Google Scholar
  16. 16.
    Modena, S., Székelyhidi, L.: Non-uniqueness for the transport equation with Sobolev vector fields. Ann. PDE 4(2), 18 (2018)MathSciNetCrossRefGoogle Scholar

Copyright information

© The Author(s) 2019

Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

Authors and Affiliations

  1. 1.Institut für MathematikUniversität LeipzigLeipzigGermany

Personalised recommendations