Processing math: 0%
Your browser does not support the canvas element.

Jincheng Yang

Institute for Advanced Study

%--Paired Delimiters-- \newcommand{\abs}[1]{\left\lvert #1 \right\rvert} \newcommand{\ang}[1]{\left\langle #1 \right\rangle} \newcommand{\bkt}[1]{\left\lbrack #1 \right\rbrack} \newcommand{\nor}[1]{\left\lVert #1 \right\rVert} \newcommand{\pth}[1]{\left( #1 \right)} \newcommand{\set}[1]{\left\lbrace #1 \right\rbrace} \newcommand{\abset}[1]{\abs{\set{#1}}} \newcommand{\ptset}[1]{\pth{\set{#1}}} %--Operators-- \newcommand{\grad}{\nabla} \newcommand{\La}{\Delta} \renewcommand{\div}{\operatorname{div}} \newcommand{\curl}{\operatorname{curl}} \newcommand{\tr}{\operatorname{tr}} \newcommand{\Hess}{\operatorname{Hess}} \newcommand{\mm}{\mathcal{M}} \newcommand{\inv}{^{-1}} \newcommand{\tensor}{\otimes} \newcommand{\cross}{\times} \newcommand{\weak}[1]{\xrightharpoonup{#1}} %--Notations-- \newcommand{\Vol}{\mathrm{Vol}} \newcommand{\loc}{\mathrm{loc}} \renewcommand{\dim}{\mathrm{dim}} \newcommand{\Span}{\mathrm{Span}} \newcommand{\diam}{\mathrm{diam}} \newcommand{\Id}{\mathrm{Id}} \newcommand{\Lip}{\mathrm{Lip}} \newcommand{\dist}{\mathrm{dist}} \newcommand{\inv}{^{-1}} %--Algebra-- \newcommand{\hfsq}[1]{\frac{\abs{#1} ^2}{2}} %--Sets-- \newcommand{\R}{\mathbb{R}} \newcommand{\RR}[1]{\R ^{#1}} \newcommand{\Rd}{\RR d} \newcommand{\Rn}{\RR n} \renewcommand{\S}{\mathbb{S}} \renewcommand{\Z}{\mathbb{Z}} \newcommand{\T}{\mathbb{T}} \newcommand{\ssubset}{\subset \subset} \newcommand{\spt}{\operatorname{spt}} \newcommand{\supp}{\operatorname{supp}} \newcommand{\diam}{\mathrm{diam}} \newcommand{\diag}{\operatorname{diag}} \renewcommand{\dim}{\mathrm{dim}} \newcommand{\dimH}{\mathrm{dim} _\mathcal{H}} \newcommand{\Sing}{\mathrm{Sing}} \newcommand{\Vol}{\mathrm{Vol}} \newcommand{\inds}[1]{\mathbf1_{\set{#1}}} \newcommand{\ind}[1]{\mathbf1_{#1}} %--Summations-- \newcommand{\SUM}[3]{\sum \cnt{#1}{#2}{#3}} \newcommand{\PROD}[3]{\prod \cnt{#1}{#2}{#3}} \newcommand{\cnt}[3]{ _{#1 = #2} ^{#3} } \newcommand{\seq}[2]{\set{#1 _{#2}} _{#2}} \newcommand{\seqi}[2]{\set{#1 _{#2}} \cnt{#2}1i} %--Superscript and Subscript \newcommand{\pp}[1]{^{(#1)}} \newcommand{\bp}[1]{_{(#1)}} \newcommand{\pps}[2][1]{^{#2 + #1}} \newcommand{\bps}[2][1]{_{#2 + #1}} \newcommand{\pms}[2][1]{^{#2 - #1}} \newcommand{\bms}[2][1]{_{#2 - #1}} %--Notations-- \newcommand{\loc}{\mathrm{loc}} \newcommand{\Span}{\operatorname{Span}} \newcommand{\argmin}{\operatorname{argmin}} \newcommand{\argmax}{\operatorname{argmax}} \newcommand{\osc}{\operatorname{osc}} \newcommand{\Id}{\mathrm{Id}} \newcommand{\Lip}{\operatorname{Lip}} \newcommand{\Leb}{\operatorname{Leb}} \newcommand{\PV}{\mathrm{P.V.}} \newcommand{\dist}{\operatorname{dist}} \newcommand{\at}[1]{\bigr\rvert _{#1}} \newcommand{\At}[1]{\biggr\rvert _{#1}} \newcommand{\half}{\frac12} %--Text-- \newcommand{\inn}{\text{ in }} \newcommand{\onn}{\text{ on }} \renewcommand{\ae}{\text{ a.e. }} \newcommand{\st}{\text{ s.t. }} \newcommand{\forr}{\text{ for }} \newcommand{\as}{\text{ as }} %--Differential-- \newcommand{\d}{\mathop{\kern0pt\mathrm{d}}\!{}} \newcommand{\dt}{\d t} \newcommand{\dx}{\d x} \newcommand{\dy}{\d y} \newcommand{\ptil}{\partial} \newcommand{\pt}{\ptil _t} \newcommand{\pfr}[2]{\frac{\partial #1}{\partial #2}} \newcommand{\dfr}[2]{\frac{\mathrm{d} #1}{\mathrm{d} #2}} \newcommand{\ddt}{\dfr{}t} \newcommand{\pthf}[2]{\pth{\frac{#1}{#2}}} %--Integral-- \newcommand{\intR}{\int _0 ^\infty} \newcommand{\intRn}{\int _{\Rn}} \newcommand{\intRd}{\int _{\Rd}} \newcommand{\fint}{-\!\!\!\!\!\!\int} \newcommand{\intset}[1]{\int _{\set{#1}}} %--Greek-- \newcommand{\e}{\varepsilon} \newcommand{\vp}{\varphi} %--Norm-- \newcommand{\nmL}[2]{\nor{#2} _{L ^#1}}

Research Blog

Couette flow

2022, August 16

\newcommand{\cN}{\mathcal N} \newcommand{\cA}{\mathcal A}

Couette Frame

Let (x, y) \in \T \times \R. Couette flow refers to the shear flow u _c (t, x, y) = y e _1. If u solves a transport-diffusion equation

\pt u + b \cdot \grad u = \nu \La u + f,

that is, u is transported by some flow b and diffuse at viscosity \nu, then we can make a change of variable z = x - t y, and

\tilde u (t, z, y) = u (t, x, y), \tilde b (t, z, y) = b (t, x, y) - u _c (t, x, y), \tilde f (t, z, y) = f (t, x, y).

Then

\begin{align*} \pt u (t, x, y) &= \pfr{}t \tilde u (t, x - t y, y) = \pt \tilde u - y \ptil _z \tilde u, \\ \ptil _x u (t, x, y) &= \pfr{}x \tilde u (t, x - t y, y) = \ptil _z \tilde u, \\ \ptil _y u (t, x, y) &= \pfr{}y \tilde u (t, x - t y, y) = -t \ptil _z \tilde u + \ptil _y \tilde u, \end{align*}

therefore

\begin{align*} (\pt + u _c \cdot \grad) u (t, x, y) &= \pt \tilde u (t, z, y), \\ (\ptil _y + t \ptil _x) u (t, x, y) &= \ptil _y \tilde u (t, z, y). \end{align*}

Moreover, denote \grad _L = (\ptil _z, \ptil _y - t \ptil _z) and \La _L = \grad _L \cdot \grad _L = \ptil ^2 _z + (\ptil _y - t \ptil _z) ^2, then \grad u = \grad _L \tilde u, \La u = \La _L \tilde u. Hence, \tilde u solves the system

\begin{align*} \pt \tilde u + \tilde b \cdot \grad _L \tilde u = \nu \La _L \tilde u + \tilde f. \end{align*}

Fourier transform

From now on, we drop the tildes. Suppose we adopt the Fourier transform (z, y) \to (k, \eta), then

\hat \grad _L = \begin{pmatrix} i k \\ i \eta - i t k \end{pmatrix} = i k \begin{pmatrix} 1 \\ \hat \eta - t \end{pmatrix}, \qquad \hat \La _L = - k ^2 (1 + (\hat \eta - t) ^2) = - k ^2 \ang{t - \hat \eta} ^2.

where \hat \eta := \eta / k (we are interested in the non-zero mode k \neq 0).

Inviscid Damping

Inviscid damping refers to the phenomenon that some sort of decay exists in the Couette frame albeit without viscosity. Consider the following inviscid equation

\pt u = -\ptil _z ^2 (- \La _L) \inv u + f.

At the first sight, one may think the right hand side seems to be a positive zeroth order Riesz transform of u, hence there should be some exponential growth. However, let \cN be a Fourier multiplier with symbol \hat \cN = \ang{t - \hat \eta}, then

\pt u = -\ptil _z ^2 (- \La _L) \inv u + f = \cN ^{-2} u + f.

Let \cA be a time-dependent Fourier multiplier defined as \cA \at{t = 0} = \Id, \dot \cA = [\pt, \cA] = - 2 \cA \cN ^{-2}, then

\pt \cA u = \dot \cA u + \cA \pt u = - 2 \cA \cN ^{-2} u + \cA \cN ^{-2} u + \cA f = - \cA \cN ^{-2} u + \cA f.

This means

\ddt \frac{\nor{\cA u} _{H ^s} ^2}2 + \nor{\cN ^{-1} \cA u} _{H ^s} ^2 \le (\cA f, \cA u) _{H ^s} \le \half \nor{\cN ^{-1} \cA u} _{H ^s} ^2 + \half \nor{\cN \cA f} _{H ^s} ^2.

Moreover, note that

\begin{cases} \pt \hat \cA = - 2 \ang{t - \hat \eta} ^{-2} \hat \cA \\ \hat \cA (0) = 1 \end{cases}

which has a solution that is uniformly bounded in \hat \eta, so \cA is a bounded multiplier from both above and below, hence

\ddt {\nor{\cA u} _{H ^s} ^2} + \nor{\cN ^{-1} u} _{H ^s} ^2 \le C \nor{\cN f} _{H ^s} ^2.

Hence

\nor{u} _{H ^s} ^2 (t) + \int _0 ^t \nor{\cN \inv u} _{H ^s} ^2 (\tau) \d \tau \le C \nor{u _{in}} _{H ^s} ^2 + C \int _0 ^t \nor{\cN f} _{H ^s} ^2 (\tau) \d \tau.

Note that \hat \cN = \ang{t - \hat \eta} \le \ang{\hat \eta} \ang{t}, so

\nor{u} _{H ^s} ^2 (t) + \int _0 ^t \ang \tau ^{-2} \nor{u} _{H ^{s - 1}} ^2 (\tau) \d \tau \le C \nor{u _{in}} _{H ^s} ^2 + C \int _0 ^t \ang{\tau} ^2 \nor{f} _{H ^{s + 1}} ^2 (\tau) \d \tau.

Here we used that for \alpha \in \mathbb R,

\ang t ^\alpha \nor{u} _{H ^{s - |\alpha|}} \le \nor{\cN ^\alpha u} _{H ^s} \le \ang t ^\alpha \nor{u} _{H ^{s + |\alpha|}}.

Enhanced Dissipation

Consider the following equation

\pt u = \nu \La _L u + f.

The naive energy estimates shows the following estimate on the energy dissipation,

\ddt \frac{\nor u _{H ^s} ^2}2 + \nu \nor{\grad _L u} _{H ^s} ^2 = (u, f) _{H ^s}.

Now, denote \hat \nu = k ^2 \nu, then -\nu \La _L = \hat \nu \cN ^2, thus (with abuse of notation we regard \hat \nu as a multiplier as well)

\pt u + \hat \nu \cN ^2 u = f.

To control the enhanced dissipation, we separate the dissipation by

\pt u + \max \{\hat \nu \cN ^2, \hat \nu ^\frac13\} u = (\hat \nu ^\frac13 - \hat \nu \cN ^2) _+ u + f.

Let \cA be a time-dependent Fourier multiplier defined as \cA \at{t = 0} = \Id, and similar as before let \dot \cA = [\pt, \cA] = - \cA (\hat \nu ^\frac13 - \hat \nu \cN ^2) _+, then

\pt \cA u + \max \{\hat \nu \cN ^2, \hat \nu ^\frac13\} \cA u = \cA f.

So

\ddt \frac{\nor{\cA u} _{H ^s} ^2}2 + \nor{\hat \nu ^{\frac16} \cA u} _{H ^s} ^2 \le (\cA f, \cA u) _{H ^s} \le \half \nor{\hat \nu ^{\frac16} \cA u} _{H ^s} ^2 + \half \nor{\hat \nu ^{-\frac16} \cA f} _{H ^s} ^2

Again, note that (\hat \nu ^\frac13 - \hat \nu \cN ^2) _+ is supported near \abs{t - \hat \eta} < \hat \nu ^{-\frac13}, which again makes \cA uniformly bounded from both sides. Therefore,

\ddt {\nor{\cA u} _{H ^s} ^2} + \nu ^\frac13 \nor{|\ptil _z| ^\frac13 u} _{H ^s} ^2 \le (\cA f, \cA u) _{H ^s} \le \nu ^{-\frac13} \nor{|\ptil _z| ^{-\frac13} f} _{H ^s} ^2

Now \nor{\abs{\ptil _z} ^\frac13 u} _{H ^s} ^2 \ge c \nor{\cA u} _{H ^s}, so

\nor{u} _{H ^s} ^2 (t) \le C \nor{u _{in}} _{H ^s} ^2 e ^{-c \nu ^\frac13 t} \int _0 ^t e ^{c \nu ^\frac13 \tau} \nu ^{-\frac13} \nor{f} _{H ^s} ^2 (\tau) \d \tau.

Reference: C. Zhai and W. Zhao, “Stability Threshold of the Couette Flow for Navier–Stokes Boussinesq System with Large Richardson Number \boldsymbol{\gamma}^{\boldsymbol{2}} \boldsymbol{\gt} \frac{\boldsymbol 1}{\boldsymbol 4},” SIAM J. Math. Anal., vol. 55, no. 2, pp. 1284–1318, Apr. 2023, doi: 10.1137/22M1495160. arXiv:2204.09662

Recent Posts

22 Jan 2021

De Giorgi

14 Apr 2020

Lorentz Space