Well-posedness of incompressible flows

2022-08-23   Varnothing

We investigate into the local well-posedness of incompressible fluid equations. $\require{extpfeil}\Newextarrow{\xRightarrow}{5,5}{0x21D2}$ First we consider the incompressible Navier-Stokes system in $\Omega\subset \R^n:$
\begin{equation}
\label{NS}
\begin{split}
&u_t+u\cdot \nabla u+\nabla p = \nu \Delta u+f,\\
&\nabla \cdot u= 0,\\
\end{split}
\tag{NS}
\end{equation}Here $u $ is the velocity, $p$ is the pressure and $f$ is the exterior force. The constant $\nu>0$ denote the viscosity.
The incompressible Euler equations arise as the varnishing viscosity limit $\nu\to 0$:
\begin{equation}
\label{Eul}
\begin{split}
&u_t+u\cdot \nabla u+\nabla p = f,\\
&\nabla \cdot u= 0,\\
\end{split}
\tag{Eul}
\end{equation}The method we adopt in this lecture is different from mild solution, since the dissipation does not occur in the Euler flow.

Smooth Approximation

Now we will prove the local well-posedness of smooth solution of Navier-Stokes/Euler equation by the regular approximation and Picard contraction. Consider the mollified equations:
\begin{equation}
\label{mNS}
\begin{split}
& u_t^\epsilon + M_\epsilon (M_\epsilon u^\epsilon \cdot\nabla M_\epsilon u^\epsilon )=-\nabla p^\epsilon +\nu M_\epsilon (M_\epsilon \Delta u^\epsilon )\\
& \nabla \cdot u^\epsilon =0.
\end{split}
\tag{mNS}
\end{equation}Here $M_\epsilon$ is the standard mollification operator (see Appendix Mollification). Applying the Leray projection $P$ to the both side, we can get that
\begin{equation}
\label{trans}
u_t^\epsilon=\nu M_\epsilon (M_\epsilon \Delta u^\epsilon )-P M_\epsilon (M_\epsilon u^\epsilon \cdot\nabla M_\epsilon u^\epsilon )\eqqcolon F_\epsilon^1(u^\epsilon )-F_\epsilon^2(u^\epsilon )\eqqcolon F_\epsilon (u^\epsilon ).
\end{equation}The first claim is the global existence for regularized system:

Given an initial data $u_0\in H^k_{\sigma }, k\in \N$. For any $\epsilon>0$, there is a unique global solution $u^\epsilon \in C^1([0,\infty);H^k_\sigma )$ for the regularized \eqref{mNS}.
We can show the local existence via the Picard theorem (Lemma 10). By the transformation \eqref{trans}, it is enough to check that $F_\epsilon :H^{k}_\sigma\to H^k_{\sigma}$ and $F_\epsilon$ is locally Lipschitz.

  1. For $u\in H^k_\sigma$, it is clear that $\nabla \cdot F_\epsilon (u)=0$ since $P$ maps into divergence-free vector fields. Moreover, we can compute by Proposition 5 thatThe detailed calculation for the last term is that: $$\norm{M_\epsilon u\otimes M_\epsilon u }_{H^k}\le C\norm{M_\epsilon u }_{L^\infty} \norm{u }_{H^k}\le C\epsilon^{-d/2} \norm{u }_{L^2}\norm{u}_{H^k}\le C C\epsilon^{-d/2}\norm{u}_{H^k}^2. $$\begin{equation*}
    \begin{split}
    \norm{F_\epsilon (u ) }_{H^k}=& \nu\norm{M_\epsilon^2 \Delta u }_{H^k}+\norm{PM_\epsilon \br{\nabla \cdot (M_\epsilon u \otimes M_\epsilon u ) } }_{H^k}\\
    \le & \nu \norm{M_\epsilon^2 u }_{H^{k+2}}+C\epsilon^{-1}\norm{M_\epsilon u\otimes M_\epsilon u }_{H^k}\\
    \le & C\br{\nu \epsilon^{-2}\norm{u}_{H^k}+\epsilon^{-\frac d2-1}\norm{u}_{H^k}^2 }.
    \end{split}
    \end{equation*}Then it implies $F_\epsilon (u)\in H^k_\sigma$ immediately.
  2. We show that $F_\epsilon$ is locally continuous in any open ball
    \begin{equation*}
    B_R\coloneqq \set{u\in H^k_\sigma| \norm{u}_{H^k}\le R }.
    \end{equation*} Given $u_1,u_2\in B_R$, a direct result is $\norm{F^1_\epsilon (u_1)-F^1_\epsilon(u_2) }_{H^k}\le C\nu\epsilon^{-2}\norm{u_1-u_2}_{H^k} $. As for the second part, we have that
    \begin{equation*}
    \begin{split}
    \norm{F^2_\epsilon(u_1 )-F^2_\epsilon (u_2) }_{H^k}\le & C\norm{M_\epsilon (u_1-u_2)\cdot \nabla M_\epsilon u_1 }_{H^k}+C\norm{M_\epsilon u_2\cdot \nabla M_\epsilon (u_1-u_2) }_{H^k}\\
    \le & C (\norm{\nabla M_\epsilon u_1 }_{L^\infty} \norm{u_1-u_2}_{H^k}+\norm{\nabla M_\epsilon u_1 }_{H^k}\norm{M_\epsilon (u_1-u_2)}_{L^\infty}\\
    &+\norm{M_\epsilon u_2 }_{L^\infty} \norm{\nabla M_\epsilon (u_2-u_1) }_{H^k}+\norm{u_2}_{H^k}\norm{\nabla M_\epsilon (u_1-u_2 ) }_{L^\infty})\\
    \le & C \epsilon^{-(\frac d2 +1 )}\norm{u_1}_{L^2}\norm{u_1-u_2 }_{H^k }+C \epsilon^{-(1+k+\frac d2 ) }\norm{u_1}_{L^2}\norm{u_1-u_2 }_{L^2 }\\
    &+C\epsilon^{-(\frac d2+1 )}\norm{ u_2}_{L^2} \norm{u_1-u_2 }_{H^k }+ C \epsilon^{-(1+k+\frac d2 ) }\norm{u_2}_{L^2}\norm{u_1-u_2 }_{L^2 } \\
    \le & C\epsilon^{-(\frac d2+1+k )}\br{\norm{u_1}_{L^2}+\norm{u_2}_{L^2} }\norm{u_1-u_2 }_{H^k}.
    \end{split}
    \end{equation*} The final result is $\norm{F_\epsilon (u_1)-F_\epsilon (u_2)}_{H^k}\le C(\epsilon,d,k,\norm{u_i}_{L^2} )\norm{u_1-u_2}_{H^k}$, so $F_\epsilon$ is locally Lipschitz in $H^k_\sigma$.

In conclusion, the Picard theorem grants a unique solution $u^\epsilon\in C^1\br{[0,T_\epsilon);H^k_\sigma}$ for some $T_\epsilon>0$. Then we attempt to extend the existence interval to infinity. It is enough to show that for any solution $u^\epsilon $, $\norm{u^\epsilon }_{H^k}$ is bounded by $\norm{u_0 }_{H^k}$. Indeed, we can show
$$\sup_{0\le t\le T} \norm{u^\epsilon (t) }_{L^2}\le \norm{u_0 }_{L^2} $$ for any $T\in [0,T_\epsilon)$, by multiplying $u^\epsilon $ to the both side of \eqref{mNS}. Furthermore, we set $u_1=u^\epsilon, u_2=0$, then by the above estimates
\begin{equation*}
\frac{d}{dt} \norm{u^\epsilon (t) }_{H^k}=\norm{F_\epsilon^2 (u^\epsilon ) }_{H^k} \le C\br{\epsilon,d, k, \norm{u^\epsilon }_{L^2} } \norm{u^\epsilon }_{H^k}\le C\br{\epsilon,d, k,\norm{u_0 }_{L^2} } \norm{u^\epsilon }_{H^k}.
\end{equation*} Grönwall's inequality implies $\norm{u^\epsilon (t) } _{H^k}\le \norm{u_0 }_{H^k}e^{Ct} $ immediately.

Besides, we prove the following high-order energy estimate for the regularized solution $u^\epsilon$, which will be applied in further discussion.

Given an initial data $u_0\in H^k_\sigma, k\in \N$. Then the unique regularized solution $u^\epsilon\in C^1([0,\infty); H^k_\sigma )$ to equation \eqref{mNS} satisfies
\begin{equation*}
\frac 12 \frac d{dt}\norm{u^\epsilon }_{H^k}^2+\nu \norm{ \nabla M_\epsilon u^\epsilon }_{H^k}^2\le C_k \norm{\nabla M_\epsilon u^\epsilon }_{L^\infty }\norm{u^\epsilon }_{H^k }^2.
\end{equation*}
Consequently for $k> d/2+1$, we have that
\begin{equation*}
\frac d{dt} \norm{u^\epsilon }_{H^k }\le C_k \norm{ \nabla M_\epsilon u^\epsilon }_{L^\infty } \norm{u^\epsilon }_{H^k}\le C_k \norm{u^\epsilon }_{H^k}^2.
\end{equation*} And hence for any $\epsilon $,
\begin{equation}
\label{appr bound}
\sup_{0\le t\le T } \norm{u^\epsilon }_{H^k}\le \frac{\norm{u_0 }_{H^k} }{ 1-C_k T\norm{u_0 }_{H^k } }.
\end{equation} It is notable that the above estimates is independent of viscosity constant $\nu$.
We take the derivative $\partial^\alpha, |\alpha |\le k $ of equation \eqref{trans} and then the $L^2-$integration with $\partial^\alpha u^\epsilon $, it follows:1
\begin{equation*}
\begin{split}
\frac 12 \frac{d}{dt}\norm{\partial^\alpha u^\epsilon }_{L^2 }^2 =&\nu \br{ M_\epsilon^2 \Delta \partial^\alpha u^\epsilon, \partial^\alpha u^\epsilon }- \br{\partial^\alpha PM_\epsilon (M_\epsilon u^\epsilon \cdot \nabla M_\epsilon u^\epsilon ), \partial^\alpha u^\epsilon } \\
=&-\nu \norm{\nabla M_\epsilon \partial^\alpha u^\epsilon }_{L^2}^2-\br{ PM_\epsilon (M_\epsilon u^\epsilon \cdot \nabla \partial^\alpha M_\epsilon u^\epsilon ), \partial^\alpha u^\epsilon }\\
&+ \br{\partial^\alpha PM_\epsilon (M_\epsilon u^\epsilon \cdot \nabla M_\epsilon u^\epsilon )-PM_\epsilon (M_\epsilon u^\epsilon \cdot \nabla \partial^\alpha M_\epsilon u^\epsilon) , \partial^\alpha u^\epsilon }.
\end{split}
\end{equation*} Here the divergence-free condition $\nabla \cdot u^\epsilon=0 $ implies:
\begin{equation*}
\br{ PM_\epsilon (M_\epsilon u^\epsilon \cdot \nabla \partial^\alpha M_\epsilon u^\epsilon ), \partial^\alpha u^\epsilon }=\br{ M_\epsilon u^\epsilon , \nabla \br{\partial^\alpha M_\epsilon u^\epsilon }^2 }=0.
\end{equation*} Summing over $|\alpha |\le k$, by Lemma 9 we have
\begin{equation*}
\begin{split}
&\frac 12\frac {d}{dt}\norm{ u^\epsilon }_{H^k}^2 +\nu \norm{\nabla M_\epsilon u^\epsilon }_{H^k}^2\\
& \le \norm{u^\epsilon }_{H^k} \sum_{|\alpha |\le k}\norm{\partial^\alpha (M_\epsilon u^\epsilon \cdot \nabla M_\epsilon u^\epsilon )- (M_\epsilon u^\epsilon \cdot \nabla \partial^\alpha M_\epsilon u^\epsilon) }_{L^2} \\
& \le C_k\norm{u^\epsilon }_{H^k}\br{ \norm{\nabla M_\epsilon u^\epsilon }_{L^\infty }\norm{ \partial^{m-1} M_\epsilon u^\epsilon }_{L^2}+\norm{\nabla M_\epsilon u^\epsilon }_{L^\infty}\norm{\partial^m M_\epsilon u^\epsilon }_{L^2} }\\
& \le C_k\norm{ \nabla M_\epsilon u^\epsilon }_{L^\infty } \norm{u^\epsilon }_{H^k}^2.
\end{split}
\end{equation*} Finally we get the high-order energy estimate.

With the above regularized result, we now show the local existence of the Euler/Navier-Stokes equations \eqref{NS} provided the initial data with enough regularity.

Given initial data $u_0\in H_\sigma^k$ with $k\ge [d/2 ]+2$. Then there exists a time $T$ with bound $T\le C_k \norm{u_0 }_{H^k}^{-1}$, such that for any $\nu\in [0,\infty)$, there exists unique solution
$$u^\nu\in C\br{[0,T];C^2 (\R^d ) }\cap C^1([0,T];C(\R^d ) ) $$ to Euler/Naiver-Stokes equations \eqref{NS}. Moreover, we have uniform energy estimate
$$\sup_{0\le t\le T}\norm{u^\nu }_{H^k }\le \frac{\norm{u_0 }_{H^k} }{1-C_kT \norm{u_0 }_{H^k} } .$$

Indeed, the significant observation is that $u^\nu$ is the limit of $u^\epsilon$ of corresponding regularized system (in subsequence). The strategy for the proof of Theorem 3 is to establish the energy estimate for $u^\epsilon$, then we show a contraction in $L^2-$norm and apply an interpolation inequality to prove convergence as $\epsilon\to 0 $.

We show the approximated solution $u^\epsilon$ is precompact in $C_t (H^{k-2}_{loc} )$ by Aubin-Lions lemma 6. The bound \eqref{appr bound} in the Remark tells $\set{u^\epsilon}$ is bounded in $L^\infty_t (H^k )$. Moreover, we can prove $u^\epsilon$ is Lipschitz in $H^{k-2}$ by the estimate:
\begin{equation*}
\begin{split}
\norm{u_t^\epsilon }_{H^{k-2} }=& \norm{F_\epsilon (u^\epsilon ) }_{H^{k-2} }\\
\le & \nu \norm{\Delta M_\epsilon u^\epsilon }_{H^{k-2} }+\norm{\nabla \cdot (M_\epsilon u^\epsilon \otimes M_\epsilon u^\epsilon ) }_{H^{k-2}}\\
\le & \nu \norm{u^\epsilon }_{H^k}+\norm{M_\epsilon u^\epsilon }_{L^\infty }\norm{M_\epsilon u^\epsilon }_{H^{k-1} }\\
\le & \nu \norm{u^\epsilon }_{H^k }+C\norm{u^\epsilon }_{H^k }^2.
\end{split}
\end{equation*} Consequently, since $H^{k}\overset{K}{\hookrightarrow} H^{k-2}\hookrightarrow H^{k-2}$ in bounded domain, we see $\set{ u^\epsilon}$ is precompact in $C_t (H^{k-2}_{loc} )$, i.e., there exists $u^\epsilon \xrightarrow{C_t H^{k-2}_{loc} } u^\nu$ in subsequence sense. Now we show $u^{\nu}$ is indeed the solution of following equation:
\begin{equation*}
u^\nu_t = F_\nu (u^\nu ), F_\nu (u)=\nu \Delta u+P (u\cdot \nabla u ).
\end{equation*} Indeed, we can show by term splitting and mollification properties that2
\begin{equation*}
\begin{split}
\frac{d}{dt} \norm{u^\epsilon -u^\nu }_{L^2}\le C \br{\epsilon +\norm{u^\epsilon -u^\nu }_{L^2} },
\end{split}
\end{equation*} and thus
\begin{equation*}
\sup_{0\le t\le T} \norm{u^\epsilon -u^\nu }_{L^2 }\le e^{CT}\br{\epsilon +\norm{u_0^\epsilon-u_0 }_{L^2} }-\epsilon \le C(T )\epsilon .
\end{equation*} The final step is to recover the pressure.
Given a intial data $u_0\in H^k_\sigma, k\ge [d/2 ]+2$, then for any viscosity $\nu\ge 0 $, there exists a maximal time of existence $T_*$ and unique solution $$u^\nu\in C\br{[0,T_* ); H^k_\sigma }\cap C^1 \br{[0,T_*); H^{k-2}_\sigma } $$ to the Euler/Navier-Stokes equation \eqref{NS}. Moreover, if $T^*\sl\infty$ then
$$\lim_{t\to T^* } \norm{u^\nu (t) }_{H^k }=\infty. $$

Appendix

We introduce the standard bump function $\rho:\R^d\mapsto \R$ as
\begin{equation*}
\rho(x)=\begin{cases}
\frac 1c e^{-\frac {1}{1-|x|^2} } & |x|\le 1 \\
0, |x|>1,
\end{cases}
\end{equation*}where $c=\int_{B_1}e^{-\frac{1}{1-|x|^2} }dx $. It can be check that $\rho\in \Di(\R^n )$ and $\int \rho dx=1$. Let $\rho_\epsilon(x)\coloneqq \epsilon^{-d}\rho(x/\epsilon)$, we define the $\epsilon-$mollification of $u\in L^1_{loc}(\Omega )$ as
\begin{equation*}
M_\epsilon u=\rho_\epsilon * u=\epsilon^{-d} \int \rho\br{\frac{\cdot -y}{\epsilon} } u(y)dy.
\end{equation*}Some important properties about mollification operator is listed as following.

  1. $M_\epsilon u\in \Ei(\R^n)$ and $\supp M_\epsilon u\in \Omega_\epsilon, \partial^\alpha M_\epsilon u=u* \partial^\alpha \rho_\epsilon$;
  2. If $u\in L^p_{loc}(\Omega )$, then $M_\epsilon u\to u$ in $L^p_{loc}(\Omega )$;
  3. If $u\in C(\Omega)$, then $M_\epsilon u\to u$ in $C_{loc}(\Omega)$;
  4. If $u\in H^s(\R^d),s\in \R$, then $M_\epsilon u\to u$ in $H^s$. Particularly, $M_\epsilon u\to u$ in $H^{s-1}$ at rate $O(\epsilon )$, i.e.
    \begin{equation*}
    \norm{M_\epsilon u-u }_{H^{s-1}}\le C\epsilon \norm{u}_{H^s};
    \end{equation*}
  5. If $u\in H^{k}(\R^d), k\in \N$, then for $\epsilon>0$,
    \begin{equation*}
    \begin{split}
    \norm{M_\epsilon u }_{H^{k+m}}\le & \frac{C_{m,k}}{\epsilon^m}\norm{u }_{H^k};\\
    \norm{M_\epsilon \partial^m u }_{L^\infty}\le & \frac{C_k}{\epsilon^{d/2+k} }\norm{u}_{L^2}.
    \end{split}
    \end{equation*}
  6. $M_\epsilon, P$ and $\partial^\alpha$ are commutative with each other.
See proofs in Universal notes, Chapter Differential for 1., 2. and 3. We just prove 4 here: The convergence in $H^s$ can be reach immediately by Lebesgue dominated convergence theorem and density argument. As for $H^{s-1}$, notice that
\begin{equation*}
\begin{split}
\norm{M_\epsilon u-u }_{H^{s-1}}^2=&\int ( 1+|\xi |^2 )^{s-1}(\hat \rho_\epsilon (\xi )-1)^2 \hat u(\xi)^2 d\xi\\
=& \int \frac{(\hat \rho_\epsilon (\xi )-1)^2}{(1+ |\xi|^2 ) } ( 1+|\xi |^2 )^{s} \hat u(\xi)^2 d\xi \\
\le & C\norm{\frac{(\hat \rho_\epsilon (\xi )-1)^2}{(1+ |\xi|^2 ) } }_{L^\infty_\xi }\norm{u }_{H^s}^2.
\end{split}
\end{equation*} Then our work is to estimate the rate of $\norm{\frac{(\hat \rho_\epsilon (\xi )-1)^2}{(1+ |\xi|^2 ) } }_{L^\infty_\xi }$.

Let $X\overset{K}{\hookrightarrow } Y\hookrightarrow Z$ be Banach spaces. Let $T>0$ and $\set{u_n }$ be a bounded sequence in $L^\infty (0,T; X ) $. Suppose also it is equicontinuous in $C([0,T];Z )$. Then $\set{u_n }$ is precompact in $C([0,T ];Y )$.
In the practice, the equicontinuity in $C_t (Z)$ can be reached if we prove $u_n \in W^{1,\infty}_t (Z)=C^{0,1}_t(Z)=Lip_t (Z) $.
First we prove the following claim by contradiction: For every $\eta>0$, there exists $C_\eta>0$ such that
\begin{equation}
\label{Compactness}
\norm{u }_Y\le \eta \norm{v }_{X }+C_\eta \norm{v }_Z.
\end{equation} Indeed, if there is a $\eta_0>0$ and sequence $\set{u_n } $ such that
\begin{equation*}
\norm{u_n }_Y>\eta_0 \norm{u_n }_X+n\norm{u_n}_Z,\forall n\in \N.
\end{equation*} Let $w_n=u_n/\norm{u_n}_X$, then the compact embedding $X\overset{K}{\hookrightarrow} Y$ implies $\set{w_n }$ is precompact in $Y$, i.e., there is a subsequence $w_{n_k}\to w_0\not=0$ in $Y$. This implies also $\norm{w_{n_k}}_Y\le C$ and then $w_{n_k}\to 0$ in $Z$, contradiction to $w_0\not=0$.

Now we show the continuity in time: we assume $\norm{u_n (t) }_{X}\le C_1,\forall n\in \N, t\in [0,T] $. Then for any $\epsilon>0,t_0\in [0,T] $, there exists $\delta (\epsilon,t_0 )>0 $ such that for any $|t-t_0|\sl\delta $, we have
\begin{equation*}
\norm{u_n(t)-u_n(t_0) }_Z\le \epsilon/(2 C_\eta),\forall n\in \N.
\end{equation*} Let $\eta=\epsilon/ (4C_1)$, then \eqref{Compactness} implies that
\begin{equation*}
\begin{split}
\norm{u_n(t)-u_n(t_0) }_Y\le & \frac{\epsilon }{4C_1}\norm{u_n(t)-u_n(t_0) }_X+C_{\frac \epsilon {4C_1} } \norm{u_n (t)-u_n(t_0) }_Z\\
\le & \frac{\epsilon }{4C_1}\cdot 2C_1+ C_{\frac \epsilon{4C_1} }\cdot \frac{\epsilon }{2C_{\frac \epsilon{4C_1}} }=\epsilon.
\end{split}
\end{equation*} Consequently, we see $u_n\in C([0,T];Y )$ for any $n\in \N$. Moreover, since $X\hookrightarrow Z$, then $\set{u_n} $ is uniformly bounded in $Z$. Together with equi-continuity on $C([0,T];Z)$, Arzela-Ascoli lemma implies $\set{u_n}$ is precompact in $C([0,T];Z)$, i,e, $\epsilon>0,\exists N>0,$ s.t.
\begin{equation*}
\sup_{t\in [0,T]} \norm{u_{n_k}(t) -u_{n_l} (t) }_Z\le \frac{\epsilon}{C_{\epsilon/4C_1 } },\forall k,l>N.
\end{equation*} Apply the inequality \eqref{Compactness} with $\eta=\epsilon/(4C_1)$ again, we have
\begin{equation*}
\begin{split}
\norm{u_{n_k} (t)-u_{n_l} (t) }_{Y}\le \frac{\epsilon }{4C}\norm{u_{n_k}(t)-u_{n_l}(t) }_X+C_{\frac \epsilon {4C} } \norm{u_{n_k} (t)-u_{n_l}(t) }_Z\le \epsilon.
\end{split}
\end{equation*} Take supremum for $t$ at both hand, we derive the precompactness in $C([0,T];Y)$.

Suppose $\Omega= \R^d$ and $n,j,k$ are non-negative integers and that $1\le p,q,r\le \infty$ are real numbers such that
\begin{equation*}
\frac 1p=\frac jd+\br{\frac 1r-\frac kd }\alpha+\frac{1-\alpha}{q}, \frac jk\le \alpha \le 1.
\end{equation*} then there exists a constant $C$ depending on $k,d,j,q,r,\alpha$ such that
\begin{equation*}
\norm{\partial^j u}_{L^p}\le C\norm{\partial^k u }_{L^r}^\alpha \norm{u }_{L^q}^{1-\alpha}.
\end{equation*}
If $\Omega\subset \R^d$ is a bounded Lipschitz domain, the interpolation has the same hypotheses as above and reads
\begin{equation*}
\norm{\partial^j u }_{L^p}\le C_1\norm{\partial^k u }_{L^r}^\alpha\norm{u}_{L^q}^{1-\alpha}+C_2\norm{u}_{L^s}
\end{equation*} for arbitrary $s\ge 1$, where the constants $C_1,C_2$ depend on $\Omega$ and $s$ in addition to the other parameters.
$H^s(\R^d)$ is a algebra if $s>d/2$, i.e. if $u,v\in H^s$, then $uv\in H^s$ with
\begin{equation*}
\norm{uv}_{H^s}\le C\norm{u}_{H^s}\norm{v}_{H^s}.
\end{equation*}
We claim that $\norm{\hat u}_{L^1}\le C\norm{u}_{H^s}$ first:
\begin{equation*}
\norm{\hat u}_{L^1}=\int \ip{\xi}^{-s}\ip{\xi}^s|\hat u(\xi) |d\xi\le \int \ip{\xi}^{-2s}d\xi \cdot \norm{u}_{H^s}\le C\norm{u}_{H^s}
\end{equation*} provided $s>d/2$. Consequently, we have
\begin{equation*}
\begin{split}
\norm{uv}_{H^s}=&\norm{ \ip{\xi}^s|\widehat{uv}(\xi) | }_{L^2_\xi}\le \norm{\int \ip{\xi}^s\abs{\hat u(\xi-\eta)\hat v(\eta)}d\eta }_{L^2_\xi}\\
\le &C\norm{\int \ip{\xi-\eta}^s \abs{\hat u(\xi-\eta)\hat v(\eta) } d\eta}_{L^2_\xi}+C\norm{\int \ip{\eta}^s \abs{\hat u(\xi-\eta)\hat v(\eta) } d\eta}_{L^2_\xi}\\
\le & C\norm{ \ip{\cdot }^s |\hat u| *|\hat v |(\xi)}_{L^2_\xi}+C\norm{|\hat u |*\ip{\cdot }^s|\hat v|(\xi) }_{L^2_\xi}\\
\le & C\norm{\ip{\cdot }^s|\hat u | }_{L^s}\norm{\hat v }_{L^1}+C\norm{\hat u}_{L^1}\norm{\ip{\cdot }^s|\hat v | }_{L^s}\\
\le & C\norm{u}_{H^s}\norm{\hat v}_{L^1}+\norm{v }_{H^s}\norm{\hat u}_{L^1}\le C\norm{u}_{H^s}\norm{v}_{H^s.}
\end{split}
\end{equation*} Here in the third step, we use the fact: For $r>0,$
\begin{equation*}
\begin{split}
(1+|\xi |^2 )^r\le & (1+2|\xi-\eta |^2+2|\eta|^2 )^r\\
\le & 2^r\br{1+|\xi-\eta |^2+1+|\eta|^2 }^r\\
\le & C\br{1+|\xi-\eta |^2 }^r +C(1+|\eta |^2 )^r
\end{split}
\end{equation*} where $C=\max\{2^r,2^{2r-1}\}$, so $\ip{\xi}^ s\le C\ip{\xi-\eta }^s+C\ip{\eta}^s$.
The result can be generalized to $W^{k,p}$ case. See Tao's lecture, Week 4, Proposition 1.1 for details (proved via Littlewood-Paley decomposition).
For any $k\in \N$, there exists $C>0$ such that for any $u,v\in H^k\cap L^\infty (\R^d)$,
\begin{equation*}
\norm{uv}_{H^k}\le C\br{\norm{u}_{L^\infty}\norm{\partial^k v}_{L^2}+\norm{\partial^k u}_{L^2}\norm{v}_{L^\infty} },
\end{equation*} \begin{equation*}
\sum_{|\alpha |\le k } \norm{\partial^\alpha (uv )-u\partial^\alpha v }_{L^2}\le C\br{ \norm{ \nabla u}_{L^2 }\norm{\partial^{k-1}v }_{L^2}+\norm{\partial^k u}_{L^2}\norm{v}_{L^\infty} }.
\end{equation*}
The main idea is control the cross term by Gagaliardo-Nirenberg inequality (Lemma 7). See Majda-Bertozzi-Ogawa, Lemma 3.4 for details.
This can be generalized to $H^s$ for $s\ge 0$ case:
\begin{equation*}
\norm{uv}_{H^k}\le C\br{\norm{u}_{L^\infty}\norm{v}_{H^s}+\norm{ u}_{H^s}\norm{v}_{L^\infty} }.
\end{equation*} See Tao's blog 254A, Notes 1, Appendix for details (proved via Littlewood-Paley theory).
Let $X$ be a Banach space and $O\usubset X$. Suppose $F: O\to X$ is locally Lipschitz, i.e. $\forall u\in O$, there exists $L>0$ and $u\in U\subset O$ such that
\begin{equation*}
\norm{F(u_1)-F(u_2)}_X\le L\norm{u_1-u_2}_{X},\forall u_1,u_2\in U.
\end{equation*} Then for any $u_0\in O$, there exists a time $T$ such that the ODE
\begin{equation*}
\begin{cases}
\frac{d u}{dt}=F(u)\\
u|_{t=0}=u_0
\end{cases}
\end{equation*} has a unique (local) solution $u\in C^1\br{[0,T); O}$.
You can see a similar proof in Equation Notes Theorem 2.1.

  1. Notice that $\rho_\epsilon * u \cdot v=\int \rho_\epsilon (y-x)u(x)dy\ \cdot v(x)=\int \rho_\epsilon (y-x)v(x)dy\ \cdot u(x)=u \cdot \rho_\epsilon * v. $ 

  2. You can see similar proof in Majda-Bertozzi-Ogawa Lemma 3.7.