Nonlinear transport equations and quasiconformal maps

Let us consider a real number R>0. For each such R and given any two arbitrary points x and y in the plane, we can always divide the plane into two regions, |x-y|\leq R and |x-y|>R. Therefore,

If we minimize the term on the right hand of the inequality as a function of R, the best possible value attainable is R=\|f\|_1^\frac12\,\|f\|_\infty^{-\frac12}. This gives the bound (3). The bound (3) is an immediate consequence of (3). Concerning (c), let us assume that diam(\operatorname{supp}(f))=2R. It is not restrictive to assume 0\in\operatorname{supp}(f), so that \operatorname{supp}(f)\subset \overline{D(0,R)}. Then, at points x such that |x|>2R one has

as claimed.

The structure of the Cauchy Kernel makes it clear that

Moreover, from Lemma 3 (c) we know that \operatorname{\textbf{v}}^{\epsilon}(t,\cdot) vanishes at \infty like \frac{C}{|z|}. Therefore,

Thus, all the requirements in Reimann’s Theorem 2 are fulfilled, and quasiconformality follows, with quasiconformality constant

and by definition, \mathcal{K}_t\geq1. Therefore, part (a) is clear. Part (b) says that quasiconformal maps are quantitatively quasisymmetric. The interested reader should check [AIM, Corollary 3.10.4] for a detailed proof. Part (c) follows from [AIM, Theorem 13.1.5] and the classical area distortion estimates for \mathcal{K}_t-quasiconformal maps.

To prove the claim (a), let us remind from Lemma 4 that X^\epsilon(t,\cdot) is quasisymmetric. That is, given any three points z_0,z,w\in\mathbb C we have

As a consequence

In particular, by leaving w fixed one can easily get that X^\epsilon(t,\cdot) is equicontinuous at z_0.
The family of maps, \{X^\epsilon(t,\cdot)\}_\epsilon is pointwise equicontinuous and locally uniformly bounded. Therefore, Arzela-Ascoli theorem ensures the existence of a locally uniform accumulation point X(t,\cdot). It is worth mentioning that by classical tools in Geometric Function Theory [AIS, Theorem 3.1.3] it can only be either \mathcal{K}_t-quasiconformal or constant. To see that it cannot be constant, one must observe that the quasisymmetry bounds are preserved by uniform limits. Being two sided, these quasisymmetry bounds guarantee bijectivity. Therefore, the accumulation point X(t,\cdot) is \mathcal{K}_t-quasiconformal.

The \mathcal{K}_t-quasiconformality of X^\epsilon(t,\cdot) ensures the existence of a well-defined, uniformly elliptic Beltrami coefficient \mu^\epsilon(t,\cdot)=\frac{\overline\partial X^\epsilon(t,\cdot)}{\partial X^\epsilon(t,\cdot)}, and moreover we know that

The smoothness in time of \partial X^\epsilon(t,z) and \overline\partial X^\epsilon(t,z) guarantees that t\mapsto \mu^\epsilon(t,z) is also smooth. From the equation for the flow \dot{X^\epsilon}(t,z)=\operatorname{\textbf{v}}^\epsilon(t,X^\epsilon(t,z)) and the chain rule we get that

On the other hand, from the kernel structure we have

Thus

Now, given any time t>0, we can integrate on (0,t) the above inequality to obtain that

since X^\epsilon(0,z)=z implies \mu^\epsilon(0,z)=0. Now, if \omega^\epsilon_0(z)=0 then necessarily \mu^\epsilon(t,z)=0 and hence \overline\partial X^\epsilon(t,z)=0. The claim follows.

By Lemma 6 we know that X^\epsilon(t,\cdot) is conformal on a neighborhood of \infty. Therefore, it has around \infty a Laurent series development whose higher order term is linear,

Also, from Lemma 3 (b) and the integral representation of X^\epsilon(t,\cdot), we know that

Above, the maximum term on the right hand side (even depending on t and \epsilon) is finite and stays bounded as |z|\to\infty, due to the smoothness in t and z of X^\epsilon(t,z). Thus, for every fixed t and \epsilon>0 one has

As a consequence, (Proof) tells us that necessarily a^\epsilon(t)=1, and so (a) follows. To see (b), we observe that X^\epsilon(t,\cdot)-b^\epsilon(t) is a principal \mathcal{K}_t-quasiconformal map, because

as |z|\to\infty. Moreover, it is conformal outside of \operatorname{supp}\omega_0^\epsilon by Lemma 6. Hence, by [AIS, Theorem 13.1.2], we have the following area distortion estimates,

\forall E\supset\,\operatorname{supp}(\omega^{\epsilon}_{0}), as claimed.

Claim (a) can be proved by recalling that \omega^{\epsilon}(t,\cdot)\circ X^{\epsilon}(t,\cdot) =\omega_0^{\epsilon}(\cdot) and the facts that X^\epsilon(t,\cdot) preserves Lebesgue-null sets and \|\omega_0^\epsilon\|_\infty\leq\|\omega_0\|_\infty.

For (b), we use Lemmas 4 (a), 6 and 8 (b) to obtain

as desired. Estimate (c) follows from Lemma 3 (b).

One has

At the first term, we consider a disk D such that \operatorname{supp}\omega^\epsilon_0,\operatorname{supp}\omega_0\subset D, and use the higher integrability of quasiconformal jacobians. If 1<p<\frac{\mathcal{K}_t}{\mathcal{K}_t-1},

Above, \|\omega_0^\epsilon-\omega_0\|_{L^{1}(D)} converges to 0, while \|J^\epsilon(t,\cdot)\|_{L^{p}(D)} is bounded in terms of \mathcal{K}_t and |D|,

by Lemma 8 (b) and the reverse Hölder property of quasiconformal jacobians [AIS], and provided that \epsilon>0 is small enough. Concerning the second term, let us choose \omega_0^n\in C_0 such that \|\omega_0^n-\omega_0\|_1\leq 1/n and \|\omega_0^n\|_\infty\leq\|\omega_0\|_\infty. Then

Above, again because of the higher integrability of quasiconformal jacobians,

and similarly for \|\omega_0^n(X (-t,\cdot))-\omega_0(X (-t,\cdot))\|_1. Thus each of these two terms can be made smaller than \delta/3 if n is chosen large enough. The control of the second term comes by continuity. Precisely, as X^\epsilon(-t,\cdot)\to X(-t,\cdot) and \omega_0^n is continuous, there is \epsilon>0 such that \|\omega_0^n(X^\epsilon(-t,\cdot)-\omega_0^n(X(-t,\cdot))\|_\infty<\delta/3. Thus (a) follows. For the proof of (b), use (a) and Lemma 3 (b).

We first go from (11) to (11). We identify \mathbb R^2\equiv \mathbb C, and write the system (11) in complex notation,

Now, taking \overline\partial on the first equation, and obtain

or equivalently,

We now multiply by e^{-i\theta}, and use the \mathbb C-linearity of the transport operator \frac{d}{dt}+\operatorname{\textbf{v}}\cdot\nabla to get

After taking real and imaginary parts,

The above equations may be seen as scalar conservation laws for \text{Re}(e^{-i\theta}\,\overline\partial\operatorname{\textbf{v}}) and \text{Im}(e^{-i\theta}\,\overline\partial\operatorname{\textbf{v}}). The second one is homogeneous, and so from the initial condition

we deduce that at any time t>0

To see this, simply call \rho=2\,\text{Im}(e^{-i\theta}\overline\partial\operatorname{\textbf{v}}) and note it satisfies the following initial value problem,

which has \rho=0 as its unique solution, due to the smoothness of \operatorname{\textbf{v}}. As a consequence, e^{-i\theta}\overline\partial\operatorname{\textbf{v}}\in\mathbb R and so if we now denote \omega=2\text{Re}(e^{-i\theta}\overline\partial\operatorname{\textbf{v}}), then

Thus the first equation at (Proof) implies that

Now, since the second equation at (11) tells us that \omega\,\operatorname{div}\operatorname{\textbf{v}}= -\Delta q, we necessarily have for \omega a homogeneous transport equation

together with the initial condition \omega|_{t=0}=\operatorname{div}(e^{i\theta}\bar\operatorname{\textbf{v}})|_{t=0}=\omega_0 as claimed.

For the converse implication, we start by noting that our choice of the kernel K and the second equation in (11) tell us that 2e^{-i\theta}\overline\partial\mathbf{v}=\omega, which by assumption is real valued. We now use the first equation in (11), together with the \mathbb C-linearity of the complex operator, to get

or equivalently

We now complex conjugate at both sides of the equality, multiply by e^{i\theta}, and use \mathbb C-linearity of the transport operator, and obtain

By assumption, the right hand side above is real, whence e^{i\theta}(\overline{\operatorname{\textbf{v}}_t + \operatorname{\textbf{v}} \cdot\partial \operatorname{\textbf{v}}+\overline{\operatorname{\textbf{v}}}\cdot\overline\partial\operatorname{\textbf{v}}}) is a conservative vector field. Thus there exists a scalar valued potential q such that

This automatically gives the first equation at (11). Moreover, if we take real parts at (Proof),

or equivalently

as claimed. The third and fourth equations in (11) are automatic from the second and fourth in (11).