Pseudo-Gevrey Smoothing for the

Passive Scalar Equations near Couette

Jacob Bedrossian∗ Siming He† Sameer Iyer‡ Fei Wang§
arXiv:2405.19233v1 [math.AP] 29 May 2024

May 30, 2024

Abstract
In this article, we study the regularity theory for two linear equations that are important
in fluid dynamics: the passive scalar equation for (time-varying) shear flows close to Couette
in T × [−1, 1] with vanishing diffusivity ν → 0 and the Poisson equation with right-hand side
behaving in similar function spaces to such a passive scalar. The primary motivation for this
work is to develop some of the main technical tools required for our treatment of the (nonlinear)
2D Navier-Stokes equations, carried out in our companion work. Both equations are studied
with homogeneous Dirichlet conditions (the analogue of a Navier slip-type boundary condition)
and the initial condition is taken to be compactly supported away from the walls. We develop
smoothing estimates with the following three features:
(1) Uniform-in-ν regularity is with respect to ∂x and a time-dependent adapted vector-field
Γ which approximately commutes √ with the passive scalar equation (as opposed to ‘flat’
derivatives), and a scaled gradient ν∇;
(2) (∂x , Γ)-regularity estimates are performed in Gevrey spaces with regularity that depends
on the spatial coordinate, y (what we refer to as ‘pseudo-Gevrey’);
(3) The regularity of these pseudo-Gevrey spaces degenerates to finite regularity near the cen-
ter of the channel and hence standard Gevrey product rules and other amenable properties
do not hold.
Nonlinear analysis in such a delicate functional setting is one of the key ingredients to our
companion paper, [5], which proves the full nonlinear asymptotic stability of the Couette flow
with slip boundary conditions. The present article introduces new estimates for the associated
linear problems in these degenerate pseudo-Gevrey spaces, which is of independent interest.

Contents
1 Introduction 2
1.1 Passive Scalar Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Poisson’s Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Main Result: Parabolic Regularity 8

2.1 Preliminary Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Parabolic Energy Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Main Theorem: Parabolic Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
∗
Department of Mathematics, University of California, Los Angeles, CA 90095, USA [email protected]
†
Department of Mathematics, University of South Carolina, Columbia, SC 29208, USA [email protected]
‡
Department of Mathematics, University of California, Davis, Davis, CA 95616, USA [email protected]
§
School of Mathematical Sciences, CMA-Shanghai, Shanghai Jiao Tong University, Shanghai 200240, China
[email protected]

1
3 Main Result: Elliptic Regularity 16
3.1 Preparatory Objects & Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Presentation of Elliptic Energy Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.3 Main Theorem: Elliptic Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4 Parabolic Regularity Estimates 19

4.1 Definitions of the Mode-by-mode Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.2 Commutator Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.3 The α/µ/γ Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.4 Proof of Proposition 2.5 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.5 Technical Lemmas: Auxiliary Bounds for Energy Estimates . . . . . . . . . . . . . . . . . . . 30

5 Commutator Estimates 34
(a,b,c)
5.1 Bounds on the Building Blocks Sm,n ωk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
(m,n)
5.2 Commutator Estimates, Cq,k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
(m,n)
5.3 Commutator Estimates, Cvisc,k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
(m,n)
5.4 Commutator Estimates, Ctrans,k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.5 Proof of Proposition 2.6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.6 Technical Lemmas: Gevrey Norm Estimates on Physical Side . . . . . . . . . . . . . . . . . . 58
5.7 Combinatorial lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

6 Elliptic Regularity Estimates 63

6.1 Setup for Elliptic Bounds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
(I,·)
6.2 Bounds on Eell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
(ℓ)
6.3 Bounds on Jell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
(E)
6.4 Bounds on Fell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.5 Proof of Proposition 3.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
6.6 Technical Lemmas: ICC-Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

References 129

1 Introduction
We study two distinct families of linear equations, each closely related to the solutions of the 2D
Navier-Stokes solutions near the Couette flow in a periodic channel.
Our investigation begins with an analysis of the passive scalar equation defined over the channel
domain T × [−1, 1]:

∂t ω + (y + U0x (t, y))∂x ω = ν∆ω + f, (1.1a)

ω(t, x, y = ±1) = 0,
ω(t = 0, x, y) = ωin (x, y).

The perturbation of the background shear flow, U0x (t, y), is assumed small in a suitable regularity
norm (with smallness independent of ν). In this paper, U0x (t, y) will be a passive quantity. However,
our eventual application of the methods developed in this paper will be for the nonlinear Navier-
Stokes equations in our companion work [5]. For this reason, we do not assume arbitrary regularity
on U0x (t, y). Rather, a central difficulty here is to track precisely how the results of the present
paper depend on the regularity of U0x (t, y). In a similar spirit, the external forcing can be considered
as nonlinearities in the 2D Navier-Stokes equations. For this reason, we will frequently refer to ω
as the ‘vorticity’.

2
 We make the following assumptions on the initial data ωin (x, y) (of course 1/4 is arbitrary and
can be replaced with any number < 1)
 
1 1
supp(ωin ) ⊂ − , , (1.2)
4 4
ωin ∈ L2 (T × [−1, 1]). (1.3)
The second equation we consider is the Poisson equation with right-hand side parameterized by
t, also set on T × [−1, 1]:
(∂x2 + ∂y2 )ψ(t, x, y) = w(t, x, y), ψ(t, x, y = ±1) = 0. (1.4)
Here the one-parameter family of abstract source terms {w(t, ·)}t≥0 should be considered to satisfy
similar a priori estimates as that of solutions to (1.1a). Thanks to the Biot-Savart law, if the source
w is the vorticity of the fluid, the skew-gradient of the solution ψ (stream function) yields the
velocity in the Navier-Stokes.
1.1 Passive Scalar Equation
Our main objective in the study of the passive scalar, (1.1a), is to prove smoothing estimates.
Smoothing estimates in general refer to a gain in regularity over the assumed initial datum which
potentially can become available after a trade-off (for example, integration in time, spatial local-
ization, etc...). In fact for the Navier-Stokes equations, even if we only have a L2 initial datum, the
solution becomes analytic immediately for any positive time t > 0 ( see [9,11] and references therein
√
for instance). However, the radius of analyticity is proportional to ν, too small to be useful for
our companion paper [5]. Regarding the smoothing effects we are considering here, more relevant
work could be found in [26–29, 33], where basically a local solution analytic near the boundary and
Sobolev regular far away is considered. Actually, our initial datum falls into this regime. Com-
pared with these results, we develop a novel scheme to give a better description of this regularity
discrepancy in a much larger time scale (up to ν −1/3 ). For local smoothing effects in Lebesgue
spaces, we refer to [22, 30] and references therein. We now proceed to indicate the precise type of
smoothing we will prove in this work.
Smoothing Estimates: Inspiration from Heat Equation: Of course, since only L2 regular-
ity on ωin is assumed we cannot expect higher regularity than L∞ 2
t ((0, ∞); Lx,y (T × [−1, 1])) over the
channel (uniformly in t). However, we might expect regularity gain as we depart from the support
of ωin , which is compactly supported on the interior, as set in (1.2). To see this, let us consider
the heat equation, where simple calculations serve as the starting point for our technique (which is
more elaborate for the passive scalar equation and the Navier-Stokes equations due to reasons we
will describe). Consider ∂t h − ν∆h = 0 on T × R, where h|t=0 = ωin . One can easily observe from
the Greens function formula that h ∈ C ∞ ([0, ∞) × T × (−1/4 − δ, 1/4 + δ)c ) for all δ > 0 uniformly
in ν. However, for our purposes, it will be better to see how to deduce this kind of regularization
using an energy estimate. For example, consider deducing a uniform-in-ν H 1 estimate:
√
• Perform the basic energy estimate to prove k ν∇hk2L2 . kωin k2L2 .
tx

• Differentiate the equation to obtain ∂t (∂y h) − ν∆(∂y h) = 0.

c
• Fix a cutoff χE (y) such that supp(χE ) ⊂ − 14 , 14 .
• Multiply by ∂y hχ2E to obtain
∂t √
ˆ
k∂y hχE k2L2 + k ν∇∂y hχE k2L2 . ν |∂y h|2 (χ2E )′′ ,
2 T×R

3
 and notice the right-hand side integrates as long as ωin ∈ L2 (the time integral is bounded by
kωin k2L2 by the first step). There is no appearance of the H 1 norm of the initial datum, ωin ,
due to the cutoff χE .
This is an example of a smoothing estimate which gains regularity pointwise in time (uniformly in
ν) due to the spatial localization. We are insisting on this energy method approach because these
are, in general, more robust (and specifically, they can be readily adapted to the full nonlinear
setting in our companion work, [5]).
Smoothing Estimates for the Passive Scalar Equation: In this paper, the main goal is
to establish analogues of the above smoothing estimates for the passive scalar equations on the
bounded domain T × [−1, 1] (with Dirichlet boundary conditions). However, doing so for passive
scalar equations (versus the heat equation) as well as a domain with boundary creates several major
difficulties and presents restrictions on the types of spaces we can hope for smoothing estimates.
We describe these here, and in so doing we introduce the reader to the main ingredients that will
appear in our (relatively complicated) functionals.

• Gevrey smoothing and the {χn (y)}: In contrast to the H 1 Sobolev smoothing discussed above
for the heat equation, we seek a much stronger type of smoothing, namely in Gevrey regularity.
This requires an infinite cascade of cutoff functions which replace the one χE , which are
defined in (2.4) – (2.5a). Of course due to the infinite gain in regularity in a finite distance
y ∈ [−1, 1], we need to decompose the physical domain in a very precise (eventually infinitely
small) manner which is achieved through the introduction of {xn , yn }, (2.2) – (2.3). The
motivation to use Gevrey regularity comes from the high regularity requirements that arise
when studying uniform-in-ν inviscid damping results (see e.g. [6, 7, 17, 20, 21, 31]).

• Gevrey coefficients, Bm,n (t): The Gevrey norms we work with are nearly analytic (as quan-
tified by our choice s > 1 in (2.1)), and are represented by the coefficient defined in (2.9). It
is common to use Gevrey or analytic regularity with a time decaying radius that is bounded
below. These factors give negative definite ‘Cauchy-Kovalevskaya’ (CK) terms that are useful
in nonlinear applications to handle certain critical nonlinearities. In our case, this is the
choice of λ = λ(t) in (2.8).

• Adapted Vector Fields (Γ, ∂x ): Unlike the simple case of the heat equation, flat ∂y derivatives
do not commute with the linearized Navier-Stokes equations, and therefore ∂y derivatives of
ω are expected to grow in t, making uniform-in-ν estimates difficult or impossible (depending
on precisely what one is after). Moreover, the nonlinear analysis quantifies regularity in a
very specific manner on the Fourier-side and it will be convenient to choose the compatible
regularity for the analysis here. Therefore, we modify the notion of derivative that we obtain
“smoothing” estimates for (1.1a). In the setting of T × [−1, 1], the Fourier analysis methods
of [6, 7, 20] will not be readily available close to the boundary, and therefore we are forced
to design a physical-side commuting vector field method. Indeed, one of the core ideas is
to design an approximate commuting vector-field which is adapted to the transport operator
appearing on the left-hand side of (1.1a). In comparison with the inviscid damping literature
[6, 7, 20], our vector field Γ mimics the operation of changing to (z, v) coordinates, taking
∂v of the resulting profile, and then transforming back (which ensures the two notions of
regularity work together well). Vector field methods have been used to quantify uniform
mixing estimates in both fluid mechanics and kinetic theory, see e.g. [2,10,15,38] for example.
However, previous works have both been in finite regularity and have used vector fields which
do not have non-trivial, time-variable coefficients.

4
 Let us now describe the objects in slightly more detail. Define the quantity v(t, y) to solve
the following evolution equation:

∂t (t(v − y)) = U0x (t, y) + νt∂y2 v, (1.5a)

∂y v|y=±1 = 1.

Correspondingly, we define our two (approximately) commuting vector fields:

1
∂x , Γ := ∂v + t∂x := ∂y + t∂x . (1.6)
∂y v

The vector-field Γ is an approximate commuting vector field for (1.1a):

2
[Γ, (∂t + (y + U0x )∂x − ν∆)]ω = G∂v Γω − ν∂v (vy2 − 1)∂v ω, (1.7)

where the quantity G is given by

U0x (t, y) − (v − y)
G(t, y) =∂t v − ν∂y2 v = . (1.8)
t
In addition to G, we will need two additional quantities which appear in our analysis that
measure the background shear flow:

H(t, y) =∂y v(t, y) − 1, (1.9a)

∂y U0x (t, y) − (∂y v(t, y) − 1)
H(t, y) =∂y G = .
t
The quantities G, H appear in (1.7) (for instance by writing vy2 − 1 = H(H + 2)). We
introduce H as it is more convenient to work with ∂y G instead of G itself. We refer to the
functions G, H, H as ‘coordinate system functions’.
√ √
• ν∇: The presence of the ν∇ derivatives is used to control various L∞ -based Gevrey norms
that appear during the analysis. When implementing the classical Ḣy1 -energy estimate for
transport diffusion-type equations, the transport term typically drives the Sobolev norm to
√
grow linearly. This extra growing term, when properly weighted by ν, can be absorbed by
the dissipation term from the L2 -based Gevrey energy estimates.

• Weighted Spaces & Spread of Vorticity, W (t, y): Another feature we want to quantitatively
capture is how quickly the vorticity can spread away from its initial support. For this reason,
we design the weight function W (t, y), (2.7).

• Co-normal weighted spaces, q(y): In order to work in Gevrey spaces on T × [−1, 1] and not
control overly-complicated and potentially singular boundary terms at y = ±1, we need to
weaken our Gevrey spaces by including degenerate weights of q(y)n , where q, defined in (2.6),
vanishes linearly as y → ±1.

• Decaying Anisotropic Regularity: Introducing the three ingredients of co-normal weights, our
adapted vector-field, and Gevrey regularity creates commutator terms which appear very
challenging to estimate uniformly in time, essentially due to dangerous combinations of 1q ,
(n + m) (this will imply a kind of derivative loss), and t appearing in commutator terms.
To handle these commutators, we introduce a novel iterative scheme which we refer to as

5
 the Iterated Co-normal Commutator (ICC) method. This scheme of estimating commutators
applies to both elliptic and parabolic equations. Therefore, we postpone the discussion of the
ICC method to the elliptic subsection below. However, we remark that part of the scheme uses
an anisotropic radius of Gevrey regularity which decays like 1/t in the Γ derivative direction
(see (2.10)).

• Time-varying Background Shear Flow: In order for our work to have maximal applicability to
nonlinear problems (in particular, to our companion paper [5]), we do not take the background
shear flow, U0x (t, y), as having arbitrarily high regularity. Rather, we very precisely quantify
functionals, (2.19), (2.20) and (2.21a), and estimates on these functionals that are required
for our results to hold. These estimates suffice to close the nonlinear analysis in [5].

Of the many aspects discussed above, the inclusion of the weight W (t, y), the decreasing radius
λ(t) and obtention of corresponding CK terms, and the precise quantification of norms on U0x (t, y)
are choices we have made in order to strengthen our results and to broaden their potential appli-
cability to other problems (in particular, to the nonlinear setting in [5]). On the other hand, the
use of the approximate vector-field Γ, the use of co-normal weights q(y), and the implementation
of the ICC method with the time decaying weight ϕ(t), are adapted to the study of the passive
scalar equation on T × [−1, 1].
We now give an informal statement of our main smoothing result on the passive scalar equations.
The precise statement requires the definitions of various functionals and can be found below in
Theorem 2.1.
Theorem 1.1 (Informal Version of the passive scalar equation theorem). Assume the initial datum,
ωin , satisfies (1.2) – (1.3). Assume further that the coordinate system functions satisfy appropriate
bounds. Then, as quantified by the coordinate system functionals and functionals on the source
term, f , the solution ω to (1.1a) satisfies Gevrey smoothing bounds:
a) On (− 12 , 12 )c , ω(t, ·), the solution is (uniformly) Gevrey regular in (∂x , Γ), for any Gevrey
class arbitrarily close to analytic.

b) On the transition zone: [− 12 , − 41 ) ∪ ( 14 , 12 ], the solution is “Pseudo-Gevrey regular” (in the

sense that the regularity depends on the location within the infinitely small slices as defined
by (2.2) – (2.4)).
In all cases, these Gevrey bounds are degenerate, weighted in both space and time. These estimates
hold for all t . ν −1/3−η for some sufficiently small η > 0.
Remark 1.2. We believe with some minor tweaking, the results can be extended ∀t . ν −1+ζ for
any ζ > 0. After the time-scale ν −1 , the qualitative behavior should be significantly different. We
did not pursue this sharper estimate as it is not required for our companion work [5]. See also our
earlier work [4], which obtains the results used to send t & ν −1+ζ in the nonlinear problem. This
naturally implies some results at the linear level, however, we do not seek pseudo-Gevrey regularity
on such long time-scales.
1.2 Poisson’s Equation
The elliptic system (1.4) is a family of Poisson equations with time dependent source terms. A
concrete example that we have in mind is the following vorticity stream function relation in the
Navier-Stokes equation

−∆ψ = ω, ψ y=±1
= 0.

6
The skew gradient of the stream function gives rise to the velocity of the fluid. We note that the
w in (1.4) corresponds to the −ω in the companion paper [5].
The central theme in analyzing the system (1.4) is to derive sharp inviscid damping estimates
in a bounded domain setting. Since the pioneering work of Orr [32], it has been known that the
inviscid damping mechanism is closely related to the hydrodynamic stability problem. To introduce
this mechanism, it is the easiest if we assume that the forcing w solves the transport equation on
the cylinder T × R:

∂t w + y∂x w = 0.

We observe that the quantities (∂y + t∂x )n w are transported by the flow and hence are bounded
over time, i.e., k(∂y + t∂x )n wk2L2 ≤ kwin k2Ḣ n . A simple Fourier transform (x, y) → (k, η), together
with the Plancherel identity yields that
Xˆ X ˆ (η + kt)4 |w(t, k, η)|2 1 kwin k2H 2
|ψ(t, k, η)|2 dη ≈ dη . k(∂y + t∂x )wk2
H2 . (1.10)
.
(k2 + η 2 )2 (η + kt)4 k 4 t4 k 4 t4
k6=0 k6=0

Hence, one can obtain decay for the solution ψ by paying regularity. Such estimates have been
justified in the nonlinear Euler equations [6,20,21,31] and the Navier-Stokes equations [7] uniformly-
in-ν without boundaries. See [1, 3, 8, 12–14, 16, 18, 19, 23–25, 34–37, 39] and the references therein for
some of the many works on the linearized Euler and Navier-Stokes problems.
In general, if there are no extra assumptions regarding the location of the forcing, the invis-
cid damping estimate derived in (1.10) is basically sharp. However, if the external forcing w is
compactly supported in space, and one is focusing on the solution ψ in the regions away from the
support of w, better decay estimates can be derived. This type of estimate, in a bounded domain
setting in these particular Gevrey spaces we are using, constitutes the first main result of our elliptic
results.
Next, we introduce the set of differential operators that capture various commutator terms that
arise when one commutes the vector fields Γ (1.6) and the co-normal weights q. We call these
differential operators the ICC-operators, which have the following general form:
 
(a,b,c) m+n a b c
S fk ∼ ∂y |k| fk . (1.11)
q

The basic intuition is that when commuting one copy of the vector field Γ and the weight q n , one
naturally introduces an extra factor of the form ∼ nq in the expression. Hence, in the analysis of
bounded domains, the boundary weight m+n q plays an equivalent role as a derivative in y. Hence,
2
an L -norm control over the quantity (1.11) is essentially a bound on the Ḣ a+b+c norm of the
solutions.
With these preliminaries, we are ready to state the informal elliptic estimates.

Theorem 1.3 (Informal Version of Elliptic Estimates). Assume that the external forcing w satisfies
similar estimates as the solutions studied in Theorem 1.1. Assume further that the coordinate system
functionals satisfy appropriate bounds. Then the solution ψ to (1.4) satisfies Gevrey smoothing
bounds:

c
a) On − 12 , 12 , ψ(t, ·), is small in a high Gevrey regularity space. It consists of a component
that undergoes inviscid damping . t1N for any N > 0 and a component that is exponentially
small in ν, i.e., . exp{−cν −1/9 };

7
 b) On the transition zone: [− 12 , − 41 ) ∪ ( 14 , 12 ], the solution is “Pseudo-Gevrey regular” (in the
sense that the regularity depends on the location within the infinitely small slices as defined
by (2.2) – (2.4)).
In all cases, these Gevrey bounds are degenerate, weighted in both space and time. These estimates
hold for all t . ν −1/3−η for some sufficiently small η > 0.

2 Main Result: Parabolic Regularity

Our main result will quantify the regularity gain for the solution, ω, to (1.1a) by controlling several
energy-dissipation-CK functionals of ω in terms of corresponding functionals on f , functionals on
U0x (t, y), and kωin kL2 .
2.1 Preliminary Objects
Cutoff Functions: First, we fix a Gevrey index s > 1. We define the following associated
parameters, which will arise in our analysis:

s > 1, 0 < σ < s − 1, σ∗ := (s − 1) − σ > 0. (2.1)

For technical purposes, we assume that σ∗ ≥ 10σ. Given the parameter σ defined above, we design
now a sequence of cut-off functions χn (y), n ∈ N. We first choose the following parameters:
3
x1 = , (2.2)
8
cσ
xn+1 =xn + 1+σ , n ≥ 1,
n
cσ
yn =xn + . (2.3)
100n1+σ
P 1
Above, the constant cσ is chosen so that cσ n≥1 n1+σ < 18 . We now define cut-offs, χn (y), adapted
to these scales:
(
0, −xn < y < xn
χn (y) = , n ≥ 1. (2.4)
1, {−1 < y < −yn } ∪ {yn < y < 1}

The following properties follow from our choice of sequences, {xn }, {yn }:
   
∞ 1 1
∩n=1 {χn = 1} ⊃ −1, − ∪ ,1 ,
2 2
supp(∇k χn+1 ) ⊂{χn = 1} for all k ∈ N, n ∈ N,
|∂yj χn | .n j(1+σ)
χn−1 . (2.5a)

Weight Functions: We now define our co-normal weight function


1

99(y + 1), −1 < y < −1 + 100 ;
q(y) = 1, 1 1 (2.6)
−1 + 50 < y < 1 − 50 ;

 1
99(1 − y), 1 − 100 < y < 1,

where q is taken to be monotone and smooth in the connecting regions. We now define our vorticity
localization weight-function:
(|y| − 1/4 − Lǫ arctan(t))2+
W (t, y) = . (2.7)
Kν(1 + t)

8
Above, K is a large but universal constant and the product Lǫ is a small but universal constant. We
highlight that the extra damping effect from the term Lǫ arctan(t) is only applied in our companion
paper [5]. Hence, one can safely drop this term and obtain all the estimates in this paper.
We define the Gevrey radius as follows:
1
λ(t) = λ0 (1 + (1 + t)− 100 ). (2.8)

Here λ0 ∈ (0, 1] is an order one small number chosen depending on the size of the solution (ǫ) and
the Gevrey index s. Subsequently, we denote our Gevrey weight:
 m+n s
λ
Bm,n (t) := . (2.9)
(m + n)!

We will also need the following decaying in time weight function

1 1
ϕ(t) := 1 = . (2.10)
(1 + t2 ) 2 hti

One simple consequence of this weight is the following:

sup ϕ(t)|t| . 1. (2.11)

t

However, the more delicate aspect of this weight is it does not decay too fast so as to force our
exterior norms to lose too many derivatives (measured by an appropriate Gevrey index) over the
1
critical time-scale t ∈ [0, ν − 3 − ). This will proven rigorously in Lemma 6.24. Comparing the
temporal functions λ(t) and ϕ(t), it is useful to record the following

ϕ̇(t) 1 ϕ̇(t)
& ⇒ 1 . hti ,
ϕ(t) hti ϕ(t)
λ̇(t) 1 λ̇(t)
& 2
⇒ 1 . hti2 .
λ(t) hti λ(t)

We introduce the notations

am,n (t) :=Bm,n (t)ϕ(t)1+n , (2.12a)

an (t) :=a0,n (t).

The final (minor) set of weights we will use are defined as follows: let {θn }∞
n=0 be a sequence of
positive numbers such that

θn+1 ≤ θn , n = 0, 1, . . . ,
θn+1
= δDrop , n < n∗ , (2.13)
θn
θn = 1, n ≥ n∗ ,

where the parameters n∗ and δDrop will be chosen (precisely in (5.41) and (5.42)) based on universal
constants.

9
2.2 Parabolic Energy Functionals

Vorticity Norms: We will have three families of energy-dissipation-CK functionals to measure

ω, the solution to (1.1a): the “γ” version corresponding to L2 level of regularity, the “α” version
corresponding to Hy1 level of regularity, and the “µ” version corresponding to Hx1 level of regularity.
We note that these (γ, α, µ) functionals are then propagated at the Gevrey level (summing over
(m, n) below). We note that our Gevrey spaces do not directly correspond to taking standard
derivatives ∂x , ∂y . This is due to a variety of reasons, chief among which is the inclusion of a
weakening weight q n as well as the sliding cut-offs χm+n . Therefore, it is important to note that
simply controlling the E (γ) functional yields essentially no useful direct H 1 information.
For an abstract function f = f (x, y) and its x-Fourier transform fk (y), we define the fm,n -
notation

fm,n :=∂xm Γn f, fm,n;k := (ik)m Γnk fk . (2.14)

Here, Γk is defined as

Γk := vy−1 ∂y + ikt. (2.15)

Then introduce now the energy functionals:

∞ X
X ∞
E (γ) [f ] := θn2 a2m,n kq n fm,n eW χm+n k2L2 , (2.16a)
m=0 n=0
X∞ X ∞
√
E (α) [f ] := θn2 a2m,n k ν∂y (q n fm,n )eW χm+n k2L2 ,
m=0 n=0
X∞ X ∞
√
E (µ) [f ] := θn2 a2m,n k ν∂x (q n fm,n )eW χm+n k2L2 . (2.16b)
m=0 n=0

Correspondingly, we have the dissipation functionals:

∞ X
X ∞
(γ) √
D [f ] := θn2 a2m,n k ν∇(q n fm,n )eW χm+n k2L2 , (2.17a)
m=0 n=0
X∞ X ∞
D (α) [f ] := θn2 a2m,n kν∂y ∇(q n fm,n )eW χm+n k2L2 ,
m=0 n=0
X∞ X ∞
D (µ) [f ] := θn2 a2m,n kν∂x ∇(q n fm,n )eW χm+n k2L2 . (2.17b)
m=0 n=0

Next, we have the CK functionals, which are generated from a time derivative falling on a decreasing
quantity. In our analysis, we have three decreasing quantities ϕ(t) defined in (2.10), the Gevrey
radius λ(t) defined in (2.8), as well as our weight function W (t, y) defined in (2.7). Consequently,
we have three families of CK terms which will be used in different ways:
∞ X
X ∞
ϕ̇
CK(γ;ϕ) [f ] := (1 + n)θn2 a2m,n kq n fm,n eW χm+n k2L2 ,
ϕ
m=0 n=0
X∞ X ∞
ϕ̇ √
CK(α;ϕ) [f ] := (1 + n)θn2 a2m,n k ν∂y (q n fm,n )eW χm+n k2L2 ,
ϕ
m=0 n=0

10
 ∞ X
X ∞
ϕ̇ √
CK(µ;ϕ) [f ] := (1 + n)θn2 a2m,n k ν∂x (q n fm,n )eW χm+n k2L2 ,
m=0 n=0
ϕ
∞ X
X ∞
λ̇
CK(γ;λ) [f ] := (m + n)θn2 a2m,n kq n fm,n eW χm+n k2L2 ,
λ
m=0 n=0
X∞ X ∞
λ̇ √
CK(α;λ) [f ] := (m + n)θn2 a2m,n k ν∂y (q n fm,n )eW χm+n k2L2 ,
m=0 n=0
λ
∞ X
X ∞
λ̇ √
CK(µ;λ) [f ] := (m + n)θn2 a2m,n k ν∂x (q n fm,n )eW χm+n k2L2 ,
λ
m=0 n=0
X∞ X ∞ p 2
CK(γ;W ) [f ] := θn2 a2m,n −∂t W q n fm,n eW χm+n ,
L2
m=0 n=0
X∞ X ∞ p √ 2
CK(α;W ) [f ] := θn2 a2m,n −∂t W ν∂y (q n fm,n )eW χm+n ,
L2
m=0 n=0
X∞ X ∞ p √ 2
CK(µ;W ) [f ] := θn2 a2m,n −∂t W ν∂x (q n fm,n )eW χm+n .
L2
m=0 n=0

Roughly speaking, by comparing CK(·;ϕ) and CK(·;λ) we see a trade-off between regularity and
decay: (m + n) is in general stronger than (1 + n), but ϕ̇
ϕ ∼ hti
−1 which is larger than λ̇ . On the
λ
other hand, CK(·,W ) encodes more subtle spatial localization information through the quotient ∂WtW

which is absent from both CK(·;ϕ) and CK(·;λ) .

The aforementioned functionals all have their k-by-k counterparts for fm,n;k (2.14). We postpone
their definitions to Section 4.1.
Exterior Coordinate System Functionals: Next, for ι ∈ {α, γ}, we introduce the following
coordinate system functionals:
(ι)
X −1
EH := ν iι /2 θn2 (n + 1)2s−2 a2n+1 hti3+2s k∂yiι H n eW/2 χn k2L2 ,
n≥0
(ι)
X −1
EG :=θ02 hti4−2Kǫ k∂yiι Gχ0 k2L2 + θn2 a2n hti3+2s k∂yiι Gn eW/2 χn−1+iι k2L2 ,
n≥1
(ι)
X
EH := θn2 ν iι /2 θn2 a2n k∂yiι Hn eW/2 χn k2L2 ,
n≥0

with
(
1, ι=α
iι = ,
0, ι=γ
and K is a big constant. The corresponding dissipation functionals are given as
(ι)
X −1
DH = ν 1+iι /2 θn2 (n + 1)2s−2 a2n+1 hti3+2s k∂y1+iι H n eW/2 χn k2L2 ,
n≥0
(ι) 2 X −1
DH = νθ02 hti4−2Kǫ ∂y1+iι Gχ0 L2
+ νθn2 a2n hti3+2s k∂y1+iι Gn eW/2 χn−1+iι k2L2 ,
n≥1
(ι)
X
DG = θn2 a2n ν 1+iι /2 k∂y1+iι Hn eW/2 χn k2L2 ,
n≥0

11
and the CK terms are written as
(ι)
X (ι)
CKH := θn2 CKH,n , (2.21a)
n≥0
(ι)
X (ι)
CKG := θn2 CKG,n ,
n≥0
(ι)
X (ι)
CKH := θn2 CKH,n ,
n≥0

where
2
X
(ι) (ι;j)
CKH,n := CKH,n (t),
j=1
(ι;1) −1 p
CKH,n :=ν iι /2 (n + 1)2s−2 a2n+1 hti3+2s k∂yiι H n −∂t W eW/2 χn k2L2 ,
(ι;2) −1
CKH,n := − ν iι /2 (n + 1)2s−2 an+1 ȧn+1 hti3+2s k∂yiι H n eW/2 χn k2L2 ,
 
 4hti4−2Kǫ1 t−1 − (4 − 2Kǫ )t hti2−2ǫ ∂ iι Gχ 2 n=0
(ι) 1 y 0 L2
CKG,n := P
 2 CK(ι;j) (t)2 n ≥ 1,
j=1 G,n
(ι;1) 3+2s −1 p
CKG,n :=a2n hti k∂yiι Gn −∂t W eW/2 χn−1+iι k2L2 ,
(ι;2) −1
CKG,n := − an ȧn hti3+2s k∂yiι Gn χn−1+iι k2L2 ,
(ι)
p
CKH,n := − ν iι /2 an ȧn k∂yiι Hn eW/2 χn k2L2 + ν iι /2 a2n k∂yiι Hn −Wt eW/2 χn k2L2 .

We remind the readers that as in the functionals of ω, we have three kinds of CK terms coming
from the functions φ, λ, and W . However, to ease the notation, we do not differentiate them and
denote the sum as CK for each coordinate functions H, H, and G.
2.3 Main Theorem: Parabolic Estimates
Next, we explain the assumptions on the external velocity field U0x (mostly stated as assumptions
on the coordinate change y 7→ v(t, y) and the auxiliary unknowns G, H, H). Fix T > 0 arbitrary.
Then we assume that ∀λ > 0 and ∀s > 1, ∃ǫ > 0 (necessarily uniform as λ → 0) such that the
following estimates hold:

• Interior coordinate system estimates:

∞ 
!4
X 2 2−(m+n)
low n
sup B0,n χc1 ∂v (vy2 − 1)k2L2 ≤ ǫ2 ,
ke low
Bm,n := , (2.22a)
0≤t≤T n=0 (m + n)!

where χ
e1 is a Gevrey-s cutoff satisfying

1,
 |ξ| ≥ 83 − 80
1
,
χ
e1 (ξ) = 0, 3 1
|ξ| ≤ 8 − 40 , ec1 (ξ) = 1 − χ
χ e1 (ξ). (2.22b)


monotone, others,

We highlight that the χ e1 -cutoff function is defined in the new coordinate system v, so the
variable ξ takes values in v(t, −1) and v(t, 1).

12
 • Exterior coordinate system estimates:
ˆ T
max sup Eι(γ) + Dι(γ) (t)dt ≤ ǫ2 , (2.22c)
ι∈{H,H,G} 0≤t≤T 0
ˆ T
max sup Eι(α) + Dι(α) (t)dt ≤ ǫ2 .
ι∈{H,H,G} 0≤t≤T 0

• Sobolev regularity estimates up to the boundary:

1 1
k∂y U0x kL∞ ∞ ≤
t Ly
, kv − ykL∞ ∞ ≤
t Ly
; (2.22d)
16 160
kvy−1 kL∞ ∞ + kvy − 1k ∞ 3 .1.
t Ly Lt H y (2.22e)

The main result of this paper for the parabolic equation (1.1a) is the following theorem.
1
Theorem 2.1. Assume ωin satisfies (1.2) – (1.3). Fix s > 1 arbitrary, 100 > η > 0, and assume
that the hypotheses (2.22) hold. There exists a λ0 > 0 and an ǫ0 > 0 such that if 0 < λ ≤ λ0 and
0 ≤ ǫ ≤ ǫ0 , the following energy inequalities are valid for all times t ≤ ν −1/3−η and 0 < ν ≤ 1 (with
all implicit constants are uniform in ν and t),
ˆ t
(γ)
E [ω(t)] + D (γ) [ω] + CK(γ;ϕ) [ω] + CK(γ;W ) [ω]ds
0
ˆ t ˆ t ˆ tp  
(γ) (γ) 99 (γ) 99 (γ) (γ)
. E [ω(0)]| + |ISource |ds + ǫν E ds + ǫν E (γ) D (γ) + CK(γ) + DH + DH ds,
0 0 0
(2.23)
ˆ t 
(α)
E [ω(t)] + D (α) [ω] + CK(α;ϕ) [ω] + CK(α;W ) [ω] ds
0
X ˆ t X
(ι)
ˆ t X
. E (ι) [ω(0)] + |ISource |ds + ǫν 98 E (ι) ds
ι∈(α,γ,µ) 0 ι∈(γ,α,µ) 0 ι∈(α,γ)
ˆ t X p X  ′ ′ (ι′ ) (ι′ )

+ ǫν 99 E (ι) D (ι ) + CK(ι ) + DH + DH ds; (2.24)
0 ι∈(α,γ) ι′ ∈(α,γ)
ˆ t
(µ)
E [ω(t)] + D (µ) [ω] + CK(µ;ϕ) [ω] + CK(µ;W ) [ω]ds
0
ˆ t ˆ t ˆ tp  
(µ) (µ) 99 (µ) 99 (γ) (γ)
. E [ω(0)] + |ISource | + ǫν E ds + ǫν E (µ) D (µ) + CK(µ) + DH + DH ds,
0 0 0
(2.25)

where the source terms are defined as follows:

∞ X
X ∞
(γ)
ISource := a2m,n θn2 Re q n ∂xm Γn f, ωm,n χ2m+n e2W , (2.26)
m=0 n=0
X∞ X ∞
(α)
ISource := a2m,n θn2 νRe q n ∂xm ∂y Γn f, ∂y ωm,n χ2m+n e2W ,
m=0 n=0
X∞ X ∞
(µ)
ISource := a2m,n θn2 νRe q n ∂xm ∂x Γn f, ∂x ωm,n χ2m+n e2W . (2.27)
m=0 n=0

13
 A few remarks are in order.

Remark 2.2. The assumption that λ is chosen small does not really lose generality as s is arbitrary,
however, it does simplify some technicalities in the proof.

Remark 2.3. With some additional technical work, we could likely extend the result to all η < 2/3.
After the time t & ν −1 , the qualitative nature of the estimates should be changed significantly, as
W no longer controls any localization. Of course, after this time, the solution decays exponentially;
if one wants to obtain sharper decay estimates, one should employ hypocoercivity-based methods
that can capture the enhanced dissipation.

Remark 2.4. Formally at least, we expect that we could take ϕ also y-dependent, namely fixing
it ≡ 1 for (say) |y| < 24
25 and decaying like t
−1 only for y close to the boundary. This is because the

decaying ϕ is only used for dealing with commutators that include q ′ . However, as this technical
enhancement is not required for our companion work [5], we did not pursue this more precise
estimate.

Theorem 2.1 follows in two steps, which corresponds to Proposition 2.5 and Proposition 2.6
below. First, we take the Fourier transform in x of (1.1a) to obtain

∂t ωk + (y + U0x (t, y))ikωk = ν∆k ωk + fk , (2.28)

ωk (t, y = ±1) = 0, ωk (t = 0, y) = ωin;k (y).

Here we used the notation ∆k := −|k|2 + ∂y2 . In order to derive higher regularity bounds, we derive
the equation for the quantity ω̊m,n;k :

ω̊m,n;k := |k|m q n Γnk ωk . (2.29)

Similarly, we will use the notation f˚m,n;k for general functions. Here, we highlight that in this
notation, we have the co-normal weight q n and we have |k|m as opposed to (ik)m in (2.14). The
resulting equation reads as follows,
(m,n) (m,n) (m,n) (m,n)
∂t ω̊m,n;k + ik(y + U0x (t, y))ω̊m,n;k − ν∆k ω̊m,n;k = Fk + Ctrans,k + Cvisc,k + Cq,k ,

where the detailed expressions of the right hand side terms are in (4.5) in Lemma 4.2. The terms
(m,n)
C··· encode all the commutators arising from various sources.

Proposition 2.5 (Section 4). Assume that the hypotheses (2.22) hold. Let ι ∈ {γ, α, µ}. Then, if
λ (2.8) is small enough compared to universal constants, there exists universal constant C# such
that
d (ι) X  
(ι) (ι) (ι)
E [ω] + D (ι) [ω] + CK(ι;D) [ω] ≤ C# |Iq(ι) | + |Itrans | + |Ivisc | + |ISource | , ι ∈ {γ, µ};
dt
D∈(ϕ,λ,W )
(2.30)
   
d  X (ι) 1 X  (ι) X
E [ω] + cα E (α) [ω] + D [ω] + CK(ι;D) [ω]
dt 2
ι∈(γ,µ) ι∈(γ,µ) D∈(ϕ,λ,W )
 
X X  (ι) (ι) (ι)

+ cα D (α) [ω] + CK(α;D) [ω] ≤ C# |Iq(ι) | + |Itrans | + |Ivisc | + |ISource(2.31)
| .
D∈(ϕ,λ,W ) ι∈(α,γ,µ)

14
 1 1
Here cα ∈ [ 16 , 2 ] is a small universal constant. Moreover, the error terms appearing above are
defined precisely in (4.28a) – (4.28b), and can be described as follows:

Iq(ι) :=commutators of the form [q, ν∆],

(ι)
Itrans :=commutators of the form [Γ, ∂t + (y + U0 )∂x ],
(ι)
Ivisc :=commutators of the form [Γ, ν∆],

(ι)
and ISource has been defined above in (2.26) – (2.27).

The next proposition provides estimates on the first three error terms appearing in the above
estimate. We have

Proposition 2.6 (Section 5). Assume that the hypotheses (2.22) hold. There exists a constant
1
c∗ ≤ 256C #
(with C# being defined in (2.30),(2.31)) such that following estimates hold for the
γ-level:

|Iq(γ) | ≤c∗ D (γ) [ω], (2.33a)

p p p p  q q 
(γ) 99 (γ) (γ) (γ) (γ) (γ) (γ)
|Itrans | .ǫν E E + D + CK 1 + DH + DH , (2.33b)
p p p 
(γ)
|Ivisc | .ǫν 100 E (γ) D (γ) + CK(γ) + ǫν 100 D (γ) . (2.33c)

The following estimates are valid for the α-level:

|Iq(α) | ≤c∗ (D (α) [ω] + D (µ) [ω] + D (γ) [ω] + CK(α,W ) ), (2.34a)
p X p p p  X  q q 
(α) (ι) (ι)
|Itrans | .ǫν 98 E (α) E (ι) + D (ι) + CK(ι) 1 + DH + DH , (2.34b)
ι∈{α,γ} ι∈{α,γ}
p p p p p p  q q 
(α) 99 (α) (γ) (α)
|Ivisc | .ǫν E (α) E (α) + D (α) + D (γ) + D (µ) + CK DH + DH (2.34c)
.

The following estimates are valid for the µ-level:

|Iq(µ) | ≤c∗ D (µ) [ω], (2.35a)

p p p p  q q 
(µ) (µ) (µ)
|Itrans | .ǫν 99 E (µ) E (µ) + D (µ) + CK(µ) 1 + DH + DH , (2.35b)
p p p 
(µ)
|Ivisc | .ǫν 100 E (µ) D (µ) + CK(µ) + ǫν 100 D (µ) . (2.35c)

We conclude the second section by a rigorous proof of Theorem 2.1 assuming Proposition 2.5,
2.6.

Proof. For the E (γ) and E (µ) estimates, one can combine the bounds (2.30) (2.33), (2.35) and
integrate in time to obtain the bounds (2.23), (2.25) directly. Thanks to the expression (2.31), we
have that
 
d (α) X 1 X X
E [ω] + D (α) [ω] + CK(α;D) [ω] + D (ι) [ω] + CK(ι;D) [ω]
dt 2cα
D∈(ϕ,λ,W ) ι∈(γ,µ) D∈(ϕ,λ,W )

15
 1 d X (ι) C# X  ′ 
(ι′ ) (ι′ ) (ι′ )
≤− E [ω] + |Iq(ι ) | + |Itrans | + |Ivisc | + |ISource | .
cα dt cα
ι∈(γ,µ) ι′ ∈(α,γ,µ)

Applying the bounds (2.33), (2.34), and (2.35) yields that

 
d (α) X 1 X X
E [ω] + D (α) [ω] + CK(α;D) [ω] + D (ι) [ω] + CK(ι;D) [ω]
dt 2cα
D∈(ϕ,λ,W ) ι∈(γ,µ) D∈(ϕ,λ,W )
 
1 d X (ι) 4c∗ C#  X
≤− E [ω] + D (ι) [ω] + CK(α,W ) 
cα dt cα
ι∈(γ,µ) ι∈(α,γ,µ)
X (ι)
X X p X  ′ ′ (ι′ ) (ι′ )

98
+C |ISource | + ǫν E (ι) + Cǫν 99 E (ι) D (ι ) + CK(ι ) + DH + DH .
ι∈(γ,α,µ) ι∈(α,γ) ι∈(α,γ) ι′ ∈(α,γ)

1 1 1
Since cα ∈ [ 16 , 2 ], and c∗ ≤ 256C #
, we have that the second term can be absorbed by the left hand
side. Now we integrate in time and obtain the result (2.24).

3 Main Result: Elliptic Regularity

3.1 Preparatory Objects & Definitions
Cutoff Functions: To distinguish between the interior and exterior components of various so-
lutions and quantities, we define the following two families of cutoff functions:
a) the interior/exterior cutoff functions of the first kind χI /χE , which are Gevrey-s regular and
satisfy


1,
 ξ ∈ − 12 , 12 ,
χI (ξ) = 0, |ξ| > 43 , χE (ξ) = 1 − χI (ξ). (3.1)


monotone, others.

b) the interior/exterior cutoff functions of the second kind χ χc1 defined above in (2.22b), where
e1 /e
the cutoff χ
e1 can be viewed as a fattened version of χ1 in (2.4).
Elliptic Equations: The starting point for our elliptic estimates are the relation with a one-
parameter family of forcing {w(t)}t≥0 :

∆ψ = w, ψ(t, x, y = ±1) = 0.

In the paper [5], we decompose the solution ψ(t, x, y) in two different ways, denoted by ψ =
Ψ(I) + Ψ(E) or ψ = φ(I) + φ(E) , based on the type of information we need.
The (Ψ(I) , Ψ(E) ) Decomposition: The first decomposition is intended for obtaining estimates of
the solution ψ in the interior of the channel. Let us recall the ideas from the companion paper [5]
without digging into technical details. One uses the cutoffs χI /χE (3.1) to localized the forcing
w and the H-quantity (1.9a) that captures the deviation between v and y. Thanks to technical
considerations, one ends up with the following

2

∆Ψ(I) =wI − (1 + H I )2 − (1 + H)2 ∂v Ψ(I) − (1 + H)∂v H − (1 + H I )∂v H I ∂v Ψ(I) ,

2

∆Ψ(E) =wE + (1 + H I )2 − (1 + H)2 ∂v Ψ(I) + (1 + H)∂v H − (1 + H I )∂v H I ∂v Ψ(I) ,
Ψ(I) (x, ±1) = 0, Ψ(E) (x, ±1) = 0.

16
Here we used the notations
F I (t, y) = χI (v(t, y))F (t, y), F E (t, y) = χE (v(t, y))F (t, y),

where F is a general function. However, we would highlight that the notations Ψ(I) , Ψ(E) do not
represent truncated functions. The motivation in implementing this decomposition is that when
we write the Ψ(I) -equation in the special coordinate system (the nonlinear (z, v)-coordinate in [7]),
the solutions Ψ(I) are easy to estimate. We refer the readers to the companion paper [5] and [6, 20]
for further details. The treatment of the Ψ(E) -component is discussed in this paper.
The (φ(I) , φ(E) ) Decomposition: We also consider another decomposition (based on the cutoff
functions χ χc1 (2.22b)) to understand the key features of the solutions near the boundary. Now
e1 /e
(E) (I)
we decompose the solution as ψk = φk (t, y) + φk (t, v(t, y)), where these two quantities satisfy
the following elliptic equations
(I) (I)
(∂v2 − |k|2 ) φk (t, v) = χ ec1 (v)(∂v2 − (vy ∂v )2 )φk (t, v),
ec1 (v) wk (t, y(t, v)) + χ (3.2a)
(E) 2 (I)
(∂y2 2
− |k| ) φk (t, y) = χ
e1 (v(t, y)) wk (t, y) + χ
e1 (v(t, y))(∂v − ∂y2 )φk (t, v(t, y)), (3.2b)
(I) (E)
φk (t, v)|v=v(t,±1) = 0, φk (t, y)|y=±1 = 0.
Gevrey Coefficients: In order to measure the interior forcing, we define the following Gevrey
coefficients
!4
low 2−(m+n)
Bm,n = .
(m + n)!

If the generating vorticity is supported in the interior region, we can utilize a smoothing mechanism
of the Green’s function to derive stronger estimates. Let us consider the new Gevrey index,
!s
λbm+n
b :=2λ(t), B
λ bm,n := , b
am,n := Bbm,n ϕn+1 . (3.3)
(m + n)!

3.2 Presentation of Elliptic Energy Functionals

Interior Functionals: We define the classical Gevrey norms
∞
X X  2 X
low low
EInt [w] := Bm,n k|k|m (∂v + ikt)n (χI (t, v)wk )k2L2 =: low
EInt [wk ]. (3.4)
k∈Z m+n=0 k∈Z

Elliptic Functionals:
We measure the two decompositions with different Gevrey norms. We consider the following
(E)
norm for the exterior contribution Ψk (t, x, y)
∞
!
(E)
X X (2 e
λ) 2(m+n)/r
(E)
Fell (t) := 2/r
kχm+n |k|m q n Γnk Ψk k2L2 . (3.5)
m+n=0
(m + n)!
k∈Z\{0}

e > 0 is an arbitrary parameter. For the application in our companion paper [5], the Gevrey
Here λ
e will be chosen such that the functional F (E) dominates the kAR (· · · )kL2 norm. We now
radius 2λ ell
introduce a set of functionals to measure the solution φ(I) , defined in (3.2a):
X ∞
X
(I,out) (I)
Eell (t) := a2m,n kχ1 |k|m (∂v + ikt)n ∂vℓ φk k2L2v ,
b ℓ ∈ {0, 1, 2},
k∈Z\{0} m+n=0

17
 X 1000
X
(I,f ull) bm,n
2 (I)
Eell (t) := B k|k|m (∂v + ikt)n φk k2L2v .
k∈Z\{0} m+n=0

The ICC Pseudo-differential Operators: If the solution is supported near the boundary,
one needs to control various commutator terms involving q. It turns out that the following ICC-
operators are crucial to express these commutators
 
m+n a b c
(a,b,c)
Sm,n fk = ∂y |k| (|k|m q n Γnk fk ) 1Sℓn (a, b, c), a + b + c = ℓ ≤ 2; (3.6a)
q
 
m+n a b c
(a,b,c)
Jm,n fk = ∂v |k| (χm+n |k|m q n Γnk fk ) 1Sℓn (a, b, c), a + b + c = ℓ ≤ 3; (3.6b)
q
 a
n b n
∂v (χn q n ∂v f0 ) 1S̊ℓ (a, b), a + b = ℓ ≤ 3.
(a,b,0)
J0,n f0 =
q n

Here, the parameters a, b, c take values in N and ∂v is defined as follows

∂v := vy−1 ∂y .
The index sets Sℓn and S̊ℓn are defined as follows
 
Sℓn := (a, b, c) ∈ N3 a + b + c = ℓ ∩ (a, b, c) ∈ N3 a + b + c ≤ n or a = 0 , (3.7)
 
S̊ℓn := (a, b) ∈ N2 a + b = ℓ ∩ (a, b) ∈ N2 a + b ≤ n or a = 0 .
(a,b,c) (a,b,c)
Moreover, Sm,n fk = Jm,n fk ≡ 0 for n < 0. We use the following notations to denote the vectors
(··· ) (··· )
formed by the operations Sm,n , Jm,n :
(≤ζ) (a,b,c) (ζ) (a,b,c)
Sm,n fk := {Sm,n fk }a+b+c≤ζ , Sm,n fk := {Sm,n fk }a+b+c=ζ ; (3.8a)
(≤ζ) (a,b,c) (ζ) (a,b,c)
Jm,n fk := {Jm,n fk }a+b+c≤ζ , Jm,n fk := {Jm,n fk }a+b+c=ζ ; (3.8b)
(≤ζ) (a,b,0) (ζ) (a,b,0)
J0,n f0 := {J0,n f0 }a+b≤ζ , J0,n f0 := {Jm,n f0 }a+b=ζ .
Moreover, we measure these vectors with the classical L2 -norm of vectors:
2 X 2 2 X 2
(≤ζ) (a,b,c) (ζ) (a,b,c)
Sm,n fk 2 := Sm,n fk 2 , Sm,n fk 2 := Sm,n fk ;
L L L L2
a+b+c≤ζ a+b+c=ζ
2 X 2 2 X 2
(≤ζ) (a,b,c) (ζ) (a,b,c)
Jm,n fk := Jm,n fk , Jm,n fk := Jm,n fk ,
L2 L 2 L2 L2
a+b+c≤ζ a+b+c=ζ
(≤ζ) 2 X (a,b,0) 2 (ζ) 2 X (a,b,0) 2
J0,n f0 := J0,n f0 , J0,n f0 := J0,n f0 .
L2 L 2 L2 L2
a+b≤ζ a+b=ζ

It is worth highlighting that there are two components to these norms. Part of the norm is the
classical Sobolev norm which only involves ∂v = vy−1 ∂y and |k|. The other part of the norm involves
the boundary weight m+n q . For a + b + c > n, we will not consider the boundary weight factor
m+n
q
because the quantity might not be well-defined, but we can still consider the classical Sobolev part.
We will derive the estimates for the following functional
(ℓ)
X X X
(a,b,c) (E) 2
Jell (t) := a2m,n kJm,n φk kL2y , ℓ ∈ {1, 2, 3}. (3.10)
k∈Z\{0} m,n≥0 a+b+c=ℓ

We highlight that these norms control the usual H ℓ norm of the function. We further highlight
(ℓ)
that for the ℓ = 1, 2 case, we will obtain L∞
t Jell -estimates whereas for the ℓ = 3 case, we will only
(3)
have L2t Jell bounds.

18
3.3 Main Theorem: Elliptic Estimates
Note that as a consequence of the assumptions in made in (2.22), there holds
∞
X
2 (≤1)
B0,n ϕ2n kJ0,n (vy2 − 1)k2L2 (suppχe1 ) . exp{−ν −1/6 }.
n=0

Theorem 3.1 (Section 6). Assume that the hypotheses (2.22) hold and recall the definitions
(3.4).The Jell functionals satisfy the following bounds:
−1/9
   
(1) (γ) (α)
Jell .e−ν E (γ) [w] + EH + EH EInt low
[w] , (3.11)
−1/9
   
(2) (γ) (α)
Jell .e−ν E (γ) [w] + EH + EH EInt low
[w] , (3.12)
−1/9 −1/9
X    
(3) (ι) (ι)
Jell .e−ν D (γ) [w] + e−ν EH + DH E (γ) [w] + EInt
low
[w] . (3.13)
ι∈{α,γ}

The Eell functionals satisfy the following bounds:

(I,out) 1
Eell (t) .n E low [w]; (3.14)
hti2 (|k|t)2n Int
(I,f ull) 1 low
Eell (t) . 4 EInt [w]. (3.15)
hti
(E)
Finally, the Fell functional satisfies the following bounds:
      
(E) −ν −1/8 (γ) low (α) 2 (γ) 2
Fell (t) .e E [w] + EInt [w] 1 + EH + EH . (3.16)

Remark 3.2. The bounds above are split into three groups. Informally, the first, (3.11) – (3.13)
measure information moving from exterior to exterior. The second, (3.14) – (3.15) captures interior
to exterior. The third, (3.16) captures exterior to interior.

4 Parabolic Regularity Estimates

In this section, we prove the (γ, α, µ) energy inequalities appearing in Proposition 2.5.
4.1 Definitions of the Mode-by-mode Functionals
We introduce some notations which will be in use throughout this section. First, compared to
(2.16a) – (2.16b), (2.17a) – (2.17b), we want to introduce notations for each fixed index, (m, n)
and frequency k. Therefore, we supply the following:

Definition 4.1. Let k ∈ Z, and m, n ∈ N. Consider the quantity

ω̊m,n;k := |k|m q n Γnk ωk ,

where q, Γk are defined in (2.6) and (2.15), respectively. Further recall the coefficients am,n (2.12).
We define an energy to capture the size of ω̊m,n;k ,
(α) (µ) (γ)
Em,n;k :=cα Em,n;k + Em,n;k + Em,n;k
:= cα a2m,n νk∂y ω̊m,n;k eW χm+n k2L2 + a2m,n ν|k|2 kω̊m,n;k eW χm+n k2L2 + a2m,n kω̊m,n;k eW χm+n k2L2(4.1)
.

19
 1 1
Here χm+n and W are defined in (2.4) and (2.7), respectively. The parameter cα ∈ [ 16 , 4 ] is a
universal constant chosen in Proposition 4.9. The following quantities naturally appear when one
takes the time derivatives of the energy (4.1) above:
a) Diffusion terms D:
(α) (µ) (γ)
Dm,n;k :=Dm,n;k + Dm,n;k + Dm,n;k
:=a2m,n ν 2 k∇k ∂y ω̊m,n;k eW χm+n k2L2 + a2m,n ν 2 |k|2 k∇k ω̊m,n;k eW χm+n k2L2
+ a2m,n νk∇k ω̊m,n;k eW χm+n k2L2 . (4.2)

Here ∇k := (ik, ∂y ).
b) Cauchy-Kovalevskaya terms CK:

(ι;ϕ) (1 + n)ϕ̇ (ι)

CKm,n;k := − Em,n;k , ι ∈ {α, µ, γ}, (4.3a)
ϕ
(α;W )
p 2
CKm,n;k :=a2m,n ν −∂t W ∂y ω̊m,n;k eW χm+n 2 , (4.3b)
L
(µ;W )
p 2
CKm,n;k :=a2m,n ν|k|2 −∂t Wω̊m,n;k eW χm+n 2 , (4.3c)
L
(γ;W )
p 2
2 W
CKm,n;k :=am,n −∂t Wω̊m,n;k e χm+n 2 . (4.3d)
L

This concludes the subsection. The remaining part of the section is organized as follows: In
subsection 4.2, we derive the fundamental equation for the higher conormal derivatives of the
solution ω̊m,n;k ; in subsection 4.3, we estimate the time derivative of each component of the energy
(4.1); in subsection 4.4, we conclude the proof of the key proposition; in the last subsection 4.5, we
provide some technical lemmas.
4.2 Commutator Structures
We derive now the equation satisfied by ω̊m,n;k .
Lemma 4.2. Let ωk satisfy the equation (2.28). Then ω̊m,n;k = |k|m q n Γnk ωk satisfies the following
system,
(m,n) (m,n) (m,n) (m,n)
∂t ω̊m,n;k + ik(y + U0x (t, y))ω̊m,n;k − ν∆k ω̊m,n;k = Fk + Ctrans,k + Cvisc,k + Cq,k , (4.4)

where the above commutators are defined by

(m,n)
Fk :=|k|m q n Γnk fk ;
Xn  
(m,n) n ℓ
Ctrans,k m n
:= − |k| q ∂v G Γn−ℓ+1
k ωk ;
ℓ
ℓ=1
X n  
(m,n) n ℓ 2
Cvisc,k :=ν|k|m q n ∂v vy2 ∂v Γkn−ℓ ωk ; (4.5a)
ℓ
ℓ=1
(m,n) n n(n + 1) n
Cq,k :=−2ν|k|m q ′ ∂y ω̊m,n;k + ν|k|m (q ′ )2 2
ω̊m,n;k − ν|k|m q ′′ ω̊m,n;k
q q q
3
X (m,n)
=: Cq,k,i . (4.5b)
i=1

In order to prove this lemma, we need to develop some general result about commutators which
will be useful also in future sections.

20
Lemma 4.3. Let A and B be two linear operators. Define the operation

adnA (B) := [A, [A, [A, ...[A, B]]...], n ∈ {1, 2, 3, ...}.

| {z }
n

Then
X
n−1
n

n n−ℓ
[A , B] = adA (B)Aℓ . (4.6)
ℓ
ℓ=0

Proof. We prove the relation through induction. First of all, we observe that the relation (4.6)
holds for n = 1 through definition. Next we assume that the relation (4.6) holds for n − 1 ≥ 1, and
show that the relation (4.6) holds for n. We accomplish this task by expressing the commutator
[An , B] as follows:

[An , B] =An B − ABAn−1 + ABAn−1 − BAn = A[An−1 , B] + [A, B]An−1

X n − 1
n−2
n−1−ℓ
=A adA (B)Aℓ + [A, B]An−1
ℓ
ℓ=0
X n − 1
n−2 X n − 1
n−2
n−ℓ ℓ n−1−ℓ
= adA (B)A + adA (B)Aℓ+1 + [A, B]An−1
ℓ ℓ
ℓ=0 ℓ=0
n−2 
X n−1  X n − 1
n−1
n−ℓ ℓ n−ℓ
= adA (B)A + adA (B)Aℓ + [A, B]An−1 .
ℓ ℓ−1
ℓ=0 ℓ=1

Here we have used the induction hypothesis to rewrite [An−1 , B]. Now we note that
     
n−1 n−1 n
+ = .
ℓ ℓ−1 ℓ

Application of this relation yields that the following relation holds

X
n−1
n
  
n
 
n−1

n n−ℓ ℓ
[A , B] = adA (B)A + − + adA (B)An−1 + [A, B]An−1
ℓ n−1 n−2
ℓ=0
X n 
n−1
n−ℓ
= adA (B)Aℓ .
ℓ
ℓ=0

Lemma 4.4. The following commutator relations hold:

X
n−1
n

n−ℓ
n  
X n n−ℓ+1
[∂y , Γnk ] = − ikt ∂v ℓ
(vy − 1)Γk + ∂v (vy − 1)Γℓk (4.7a)
ℓ ℓ
ℓ=0 ℓ=1
X 
n−1
n n−ℓ−1 ′′ 2 n−ℓ ′′

[∂yy , Γnk ] = − 2∂v v ∂v + ∂v v ∂v Γℓk ; (4.7b)
ℓ
ℓ=0
 
n ′ n n
[∂y , q ]f =q q f ;
q

21
 2
     
n ′ 2n + n n2 n ′ n n ′′ n n
[∂yy , q ]f = − (q ) q f + 2q ∂y (q f ) + q q f ; (4.7c)
n2 q2 q q
 
  q′ n n
∂v , q n f = q f ; (4.7d)
vy q
h i       
2 q′ n q ′′ n n q′ 1 n n
∂v , q n f =2 n
∂y (q f ) + q f − ∂y vy q f
(vy )2 q (vy )2 q (vy )2 vy q
 
(q ′ )2 n2 + n n2 n
− q f ; (4.7e)
(vy )2 n2 q2
h i q′ q ′′ q′
2
∂v , q f =2 2 ∂y f + 2 f − 3 vyy f. (4.7f)
vy vy vy

Here we recall the notation ∂v = vy−1 ∂y .

Proof. We recall Lemma 4.3 and compute the [Γk , ∂y ]
1 1 1 1
[Γk , ∂y ]f = [vy−1 ∂y , ∂y ]f = ∂y ( )vy ( ∂y f ) = − ∂y (vy − 1) × ∂y f.
vy vy vy vy

Hence the commutator is a product of the ∂v -derivative of the vy − 1, f . This will help us in the
future estimates. Now all the higher commutators are simple:
1
adnΓk (∂y )f = −( ∂y )n (vy − 1) × vy−1 ∂y f.
vy
Now the commutator is
X
n−1
n
 n−1  
X n 1
n−ℓ
[∂y , Γnk ] =− ℓ
adΓk (∂y )Γk = ( ∂y )n−ℓ (vy − 1) × vy−1 ∂y Γℓk
ℓ ℓ vy
ℓ=0 ℓ=0
n−1 
X n 1  n  
X
n−ℓ ℓ n 1
= − ikt ( ∂y ) (vy − 1)Γk + ( ∂y )n−ℓ+1 (vy − 1)Γℓk .
ℓ vy ℓ vy
ℓ=0 ℓ=1

Next we compute the [Γk , ∂yy ] using the Γ0 -derivative of v ′′ = ∂yy v = vy v1y ∂y vy :

1 1 1 1 1
[∂yy , Γk ]f = [∂yy , vy−1 ∂y ]f = ∂yy ( ∂y f ) − ∂yyy f = −2v ′′ ( ∂y )2 f − ( ∂y )v ′′ ( ∂y )f.
vy vy vy vy vy

1 1 1
adnΓk (∂yy )f = 2( ∂y )n−1 v ′′ · ( ∂y )2 f + ( ∂y )n v ′′ · (vy−1 ∂y )f.
vy vy vy

nX

n−1
[∂yy , Γnk ] =− adΓn−ℓ (∂yy )Γℓk
ℓ k
ℓ=0
X n 
n−1 
=− 2(vy−1 ∂y )n−ℓ−1 v ′′ (vy−1 ∂y )2 + (vy−1 ∂y )n−ℓ v ′′ vy−1 ∂y Γℓk
ℓ
ℓ=0
X n 
n−1 
=− 2(vy−1 ∂y )n−ℓ−1 v ′′ (Γ − ikt)2 + (vy−1 ∂y )n−ℓ v ′′ (Γ − ikt) Γℓk .
ℓ
ℓ=0

22
The proof of the last two commutator relation is more direct. First of all,
n
[∂y , q n ]f = ∂y (q n f ) − q n ∂y f = nq n−1 q ′ f = q ′ q n f.
q
Next we consider the commutator
     
n n n ′ n n ′ 2 n(n − 1) n ′′ n n
[∂yy , q ]f =∂yy (q f ) − q ∂yy f = 2q q ∂y f + (q ) q f +q q f
q q2 q
       
n n n n(n − 1) n ′′ n n
=2q ′ [q , ∂y ]f + 2q ′ ∂y (q n f ) + (q ′ )2 q f + q q f
q q q2 q
2 2
   
n n −n n n n n
= − 2(q ′ )2 2 q n f + (q ′ )2 2
q f + 2q ′ ∂y (q n f ) + q ′′ q f
q q q q
2
   
′ 2n + n n ′ n n ′′ n n
= − (q ) q f + 2q ∂y (q f ) + q q f .
q2 q q
The proof of (4.7d) is direct. Next we focus on the expression (4.7e),
" 2 #  2  2
1 n 1 n n 1
∂y , q f = ∂y (q f ) − q ∂y f
vy vy vy
 ′ 
q′ n n 1 1 q n n
=2 q ∂y f + ∂y q f
vy q vy vy vy q
   ′  
q′ n n 1 q n n (q ′ )2 n(n − 1) n
=2 q ∂y f + ∂ y q f + q f
(vy )2 q vy vy q (vy )2 q2
     
q′ n n (q ′ )2 n2 n q ′′ n n
=2 ∂y (q f ) − 2 q f + q f
(vy )2 q (vy )2 q 2 (vy )2 q
    
q′ 1 n n (q ′ )2 n(n − 1) n
− ∂ v
y y q f + q f
(vy )2 vy q (vy )2 q2
′
  ′′
 
q n q n n
=2 2
∂y (q n f ) + 2
q f
(vy ) q (vy ) q
    
q′ 1 n n (q ′ )2 n2 + n n2 n
− ∂y vy q f − q f .
(vy )2 vy q (vy )2 n2 q2

With these commutator identities in hand, we can now prove Lemma 4.2.

Proof of Lemma 4.2. We organize the proof in two steps. In the first step, we commute the vector
field Γnk with the equation, and then we commute the boundary weight q(y)n with the resulting
equation.
Step # 1: Commute Γnk -vector field. We apply the vector field Γnk on the equation (2.28), and
rewrite the equation in the form:
∂t Γnk ωk + ik(y + U0x (t, y))Γnk ωk − ν∆k Γnk ωk =Γnk fk
− [Γnk , ∂t + ik(y + U0x (t, y)) − ν∂yy ]ωk . (4.8)
The goal is to rewrite the last commutator term. First of all, we recall the equation (1.5a) and
compute the commutator as follows
 
m m−1 −1 1
adΓk (∂t )f =adΓk vy ∂y + ikt, (vty )∂y − ik f
(vy )2

23
  
1
=adm−1
Γk vy−1 ∂y + ikt, (νvyyy + Gy )∂y − ik f
(vy )2
m m
=ν∂v (vyy ) ∂v f + ∂v G ∂v f − ik1m=1 .

Next we invoke the relation (4.6) to derive the commutator expression

n  
X 
n n n−ℓ n−ℓ
[Γk , ∂t ]ωk = ν∂v (vyy ) ∂v Γℓk ωk + ∂v G ∂v Γℓk ωk
ℓ
ℓ=0
− iknΓkn−1 ωk .

Next we compute the commutator [Γnk , ik(y + U0x )]. To this end, we apply the relation (4.6), the
definition of G (1.8) and calculate that

[Γnk ,ik(y + U0x )]ωk

X n
n−1 X n
n−1
 −1 

n−ℓ
= x ℓ
adΓk (ik(y + U0 ))Γk ωk = adn−ℓ−1
Γk vy ∂y + ikt, ik(y + U0x ) Γℓk ωk
ℓ ℓ
ℓ=0 ℓ=0
X n
n−1   vy 
 X n 
n−1

n−ℓ
n−ℓ−1 ikt ℓ
= adΓk ∂y G + Γk ω k = vy−1 ∂y G iktΓℓk ωk + nikΓkn−1 ωk .
ℓ vy t ℓ
ℓ=0 ℓ=0

Combining these two commutator representations and the commutator expression (4.7a) and the
relation (4.7b), we have that
n−1  
X n
n−ℓ
[Γnk , ∂t + ik(y + U0x ) − ν∂yy ]ωk = vy−1 ∂y G Γℓ+1 ωk
ℓ
ℓ=0
X
n−1 
n 
n−ℓ−1 
2
−ν 2 vy−1 ∂y vyy vy−1 ∂y Γℓk ωk .
ℓ
ℓ=0

When combined with (4.8), this yields the following

∂t Γnk ωk + ik(y + U0x (t, y))Γnk ωk − ν∆k Γnk ωk

Xn   n  
X
n ℓ n ℓ 2
=Γnk fk − ∂v G Γn−ℓ+1
k ω k ν ∂v vy2 ∂v Γkn−ℓ ωk . (4.9)
ℓ ℓ
ℓ=1 ℓ=1

Step # 2: Derivation of (4.4). Next, we insert our co-normal weights, q(y)n . The computation
is straight forward. We observe that from the equation (4.9), one derives the following relation

∂t q n Γnk ωk + ik(y + U0x )q n Γnk ωk − ν∆k (q n Γnk ωk )

n  
X X n  
n ℓ n−ℓ+1 n ℓ 2
n n
=q Γk fk − q n
∂v G Γk ωk +νq n
∂v vy2 ∂v Γkn−ℓ ωk + ν[q n , ∂yy ]Γnk ωk .
ℓ ℓ
ℓ=1 ℓ=1

Multiplying this equation by |k|m and invoking the commutator relation (4.7c) yields the result
(4.4).

We will now need to differentiate (4.4) once in ∂y in order to derive the appropriate system for
∂y ω̊m,n;k . Indeed, doing so results in

∂t ∂y ω̊m,n;k + ik(y + U0x (t, y))∂y ω̊m,n;k − ν∆k ∂y ω̊m,n;k

24
 (m,n) (m,n) (m,n) (m,n)
=∂y Tk + ∂y Ctrans,k + ∂y Cvisc,k + ∂y Cq,k − ik(1 + ∂y U0x )ω̊m,n;k . (4.10)

(m,n)
To estimate the commutator term Cq , we further decompose it in the following lemma.
(m,n)
Lemma 4.5. The spacial derivative of the Cq,k term reads as follows,

n 2 2 3
(m,n) (2) n (3) n
∂y Cq,k =νΥ(1)
n ∂ ω m,n;k + νΥ ∂y ω m,n;k + νΥ ωm,n;k , (4.11)
q y n
q2 n
q3
where the coefficients are defined as follows:

Υ(1)
n :=−2q ,
′
(4.12a)
n + 3 ′ 2 3 ′′
Υ(2)
n := (q ) − q q,
n n
(3) 2n + 3 ′′ ′ n+1 1
Υn := q q q − 2 2 (q ′ )3 − 2 q ′′′ q 2 . (4.12b)
n2 n n
Proof. The proof is a straightforward computation.

4.3 The α/µ/γ Estimates

In this section we obtain energy bounds for frozen (m, n; k). We start with the L2 -component of
(·),(·)
the functional. Throughout the lemmas of this section, Im,n;k will be local variables: for the proof
of each lemma, we use these notations to represent various terms appeared in the decomposition.
However, after the proof is accomplished, these variables will be refreshed to represent other things.

Lemma 4.6 (The γ-term Estimates). The time evolution of the γ-component of the functional
Em,n;k (4.1) can be estimated as follows:

d 2
(γ) (γ;ϕ) (γ;W )

am,n kω̊m,n;k eW χm+n k2L2 + Dm,n;k + 2CKm,n;k + CKm,n;k
dt
λ2s λ2s (γ)
. D
(γ)
m,n−1;k 1 n≥1 + D 1n=0,m+n≥1
(m + n)2σ∗ m2σ∗ m−1,0;k
D   E
(m,n) (m,n) (m,n) (m,n)
+ a2m,n Re ω̊m,n;k , Tk + Ctrans;k + Cvisc;k + Cq,k e2W χ2m+n . (4.13)

The dissipation term D (γ) is defined in (4.2) and the CK(γ) terms are defined in (4.3). Moreover,
(m,n) (m,n) (m,n) (m,n)
the parameter σ∗ is defined in (2.1) and the terms Tk , Ctrans;k , Cvisc;k , Cq,k are defined in
(4.4).

Proof. We freeze (m, n; k) and assume that m + n ≥ 1. Thanks to the definition of CK(γ) (4.3a),
(4.3d), we observe that

d 2
d (γ;ϕ) (γ;B)
am,n kω̊m,n;k eW χm+n k2L2 = a2m,n kω̊m,n;k eW χm+n k2L2 − 2CKm,n;k − 2CKm,n;k . (4.14)
dt dt
x
Recalling the weight (2.9) and the equation (4.4), we compute the identity (letting U 0 (t, y) :=
y + U0x (t, y)),

1 d
a2m,n kω̊m,n;k eW χm+n k2L2
2 dt
=a2m,n Re ω̊m,n;k , ∂t ω̊m,n;k e2W χ2m+n + a2m,n Re ω̊m,n;k , ω̊m,n;k ∂t (e2W )χ2m+n

25
 x
= − a2m,n Re ω̊m,n;k , ikU 0 ω̊m,n;k e2W χ2m+n + a2m,n Re ω̊m,n;k , ν∆k ω̊m,n;k e2W χ2m+n


+ a2m,n Re ω̊m,n;k , ω̊m,n;k ∂t e2W χ2m+n
D   E
(m,n) (m,n) (m,n) (m,n)
+ a2m,n Re ω̊m,n;k , Tk + Ctrans;k + Cvisc;k + Cq,k e2W χ2m+n
4
X (i)
=: Im,n;k . (4.15)
i=1

First of all, we note that integration by parts together with the boundary condition ω̊m,n;k (t, y =
(1)
±1) = 0 yields that Im,n;k = 0. We thus turn to the remaining terms above.
(2) (3)
Next we estimate the terms Im,n;k + Im,n;k in (4.15). We recall the boundary condition
ω̊m,n;k (t, y = ±1) = 0 and integrate by parts to get
(2) (3)
Im,n;k + Im,n;k
2
=a2m,n Re ω̊m,n;k , ν∂y2 ω̊m,n;k e2W χ2m+n − νa2m,n |k|2 ω̊m,n;k eW χm+n L2
+ a2m,n Re ω̊m,n;k , ω̊m,n;k ∂t (e2W )χ2m+n
√ p 2
= − a2m,n νk∇k ω̊m,n;k eW χm+n k2L2 − 2a2m,n ω̊m,n;k −∂t W eW χm+n
L2
− a2m,n Re 2W
ω̊m,n;k , ν∂y ω̊m,n;k ∂y (e )χ2m+n − a2m,n νℜ ω̊m,n;k , ∂y ω̊m,n;k e2W ∂y (χ2m+n )
p 2
X
√ 2 2 (2,i)
=: − a2m,n νk∇k ω̊m,n;k eW χm+n k2L2 − 2am,n ω̊m,n;k −∂t W eW χm+n + Im,n;k .(4.16)
L2
i=1

(2,1)
The term Im,n;k can be estimated with (4.30) as follows
(2,1)
Im,n;k ≤a2m,n νk|∂y W | ω̊m,n;k eW χm+n k2 k∂y ω̊m,n;k eW χm+n k2
1 1 p 2
≤ a2m,n νk∂y ω̊m,n;k eW χm+n k22 + a2m,n ω̊m,n;k −∂t W eW χm+n . (4.17)
4 8 2

Here we have chosen K large enough in (4.30).

Next, we recall the definitions of am,n , Bm,n (2.12), (2.9), the parameter choice (2.1) and the
property of the cutoff χm+n (2.5a)
√ √
|Im,n;k | .(m + n)1+σ a2m,n k ν∂y ω̊m,n;k eW χm+n kL2y k νω̊m,n;k eW χm+n−1 kL2y 1n≥1 1m+n≥1
(2,2)

√ √
+ m1+σ a2m,0 k ν∂y ω̊m,0;k eW χm kL2y k νω̊m,0;k eW χm−1 kL2y 1m≥1
 
√ √ m+n
σ 2 W
.(m + n) am,n k ν∂y ωm,n;k e χm+n kL2y ν ω̊m,n;k eW χm+n−1 1n≥1
q L2y
√ √
+ am,0 k ν∂y ω̊m,0;k eW χm kL2y am−1,0 λs m−σ∗ k ν|k|ωm−1,0;k eW χm−1 kL2y 1m≥1 .

(1,0,0)
In the next section, we will develop several technical estimates concerning the quantity Sm,n ωk :=
q ω̊m,n;k 1S1n (1, 0, 0) (3.6a). Without digging into the details, we directly invoke (5.1a) to obtain
m+n

that
√ q
(2,2) σ W s −σ−σ∗ (γ)
|Im,n;k | .(m + n) am,n k ν∂y ω̊m,n;k e χm+n kL2y λ (m + n) Dn−1,m;k
1 √ λ2s (γ)
≤ a2m,n k ν∂y ω̊m,n;k eW χm+n k2L2y + C D . (4.18)
8 (m + n)2σ∗ n−1,m;k

26
Combining (4.14), (4.15), (4.16), (4.17), and (4.18) yields (4.13).
Finally, we observe that the estimates above applied to the m + n = 0 case with mild modifi-
cation. This concludes the proof of the lemma.

We now want to commute our γ estimate by a factor of ν|k|.

Lemma 4.7 (The µ-term Estimate). The time evolution of the µ-component of the functional
Em,n;k (4.1) can be estimated as follows:
d 2
(µ) (µ;ϕ) (µ;W )
am,n ν|k|2 kω̊m,n;k eW χm+n k2L2 + Dm,n;k + 2CKm,n;k + CKm,n;k
dt
λ2s λ2s (µ)
. D
(µ)
m,n−1;k 1 n≥1 + D 1n=0 1m≥1
(m + n)2σ∗ m2σ∗ m−1,0;k
D   E
(m,n) (m,n) (m,n) (m,n)
+ a2m,n ν|k|2 Re ω̊m,n;k , Tk + Ctrans;k + Cvisc;k + Cq,k e2W χ2m+n . (4.19)

The dissipation term D (µ) is defined in (4.2) and the Cauchy-Kovalevskaya terms CK(µ) are defined
(m,n) (m,n) (m,n) (m,n)
in (4.3a) and (4.3c). The parameter σ∗ is defined in (2.1). The terms Tk , Ctrans;k , Cvisc;k , Cq,k
are defined in (4.4).
Proof. This follows by scaling (4.13) by ν|k|2 , and using the bound
(γ) (µ)
ν|k|2 Dm,n−1;k (t) . Dm,n−1;k (t).

Lemma 4.8 (The α-term Estimate). The time evolution of the α-component of the functional
Em,n;k (4.1) can be estimated as follows:
d
(α) (α;ϕ) (α;W )
cα a2m,n νk∂y ω̊m,n;k eW χm+n k2L2 + cα Dm,n;k + 2cα CKm,n;k + cα CKm,n;k
dt
λ2s λ2s
≤C D m,n−1;k 1 n≥1 + C Dm−1,0;k 1m≥1,n=0 + 2cα k1 + ∂y U0x kL∞
(γ)
Dm,n;k
(m + n)2σ∗ m2σ∗ t,y

Cλ2s  
+ 1n=1
(α) (α,W )
D m,0;k + CK m,0;k
(m + 1)2σ∗
D   E
(m,n) (m,n) (m,n) (m,n)
+ Ca2m,n νRe ∂y ω̊m,n;k , ∂y Tk + ∂y Ctrans,k + ∂y Cvisc,k + ∂y Cq,k e2W χ2m+n .(4.20)

Here the D (α) , CK(α) terms are defined in (4.2), (4.3a), (4.3b). Here we also recall the definitions
in (4.10).
e x := y +U x ),
Proof. We freeze (m, n; k) and compute upon invoking (4.3), (4.10) (below we define U 0 0

cα d 2

am,n νk∂y ω̊m,n;k eW χm+n k2L2
2 dt
cα
= νRe ∂y ω̊m,n;k , ∂y ω̊m,n;k ∂t (a2m,n e2W )χ2m+n + cα a2m,n νRe ∂t ∂y ω̊m,n;k , ∂y ω̊m,n;k e2W χ2m+n
2 D E
X (α;ι)
≤ − cα CKm,n;k − cα a2m,n νRe ∂y ω̊m,n;k , ikUe0x ∂y ω̊m,n;k e2W χ2m+n
ι∈{ϕ,W }

+ cα a2m,n νRe ∂y ω̊m,n;k , ν∆k ∂y ω̊m,n;k e2W χ2m+n

D E
e0x ω̊m,n;k e2W χ2m+n
− cα a2m,n νRe ∂y ω̊m,n;k , ik∂y U

27
 D   E
(m,n) (m,n) (m,n) (m,n)
+ cα a2m,n νRe ∂y ω̊m,n;k , ∂y Tk + ∂y Ctrans,k + ∂y Cvisc,k + ∂y Cq,k e2W χ2m+n
X 4
X
(α;ι) (i)
=: − cα CKm,n;k + Im,n;k . (4.21)
ι∈{ϕ,W } i=1

(1)
We first note that the term Im,n;k in (4.21) is zero,
ˆ 1
(1)
Im,n;k = −cα a2m,n νRe |∂y ω̊m,n;k |2 ik(y + U0x (t, y))e2W χ2m+n dy = 0. (4.22)
−1

(2)
Next we address the term Im,n;k in (4.21). Integration by parts yields the following decomposition

(2) 2
Im,n;k :=cα a2m,n νRe ∂y ω̊m,n;k , ν∂y2 (∂y ω̊m,n;k )e2W χ2m+n − cα a2m,n ν 2 |k|2 ∂y ω̊m,n;k eW χm+n L2y

= − cα a2m,n ν 2 k∂y2 ω̊m,n;k eW χm+n k2L2y − cα a2m,n ν 2 |k|2 k∂y ω̊m,n;k eW χm+n k2L2y
− 2cα a2m,n ν 2 Re ∂y ω̊m,n;k , ∂y2 ω̊m,n;k (∂y W )e2W χ2m+n
 
+ cα a2m,n ν 2 Re ∂y ω̊m,n;k ∂y2 ω̊m,n;k e2W χ2m+n
y=±1
− cα a2m,n ν 2 Re 2 2W 2
∂y ω̊m,n;k , ∂y ω̊m,n;k e ∂y (χm+n )
X3
(2,i)
=: − cα a2m,n ν 2 k∂y ∇k ω̊m,n;k eW χm+n k2L2y + Im,n;k . (4.23)
i=1

We recall the negative CK term in (4.21) and the dissipative term in (4.23), and apply the property
(2,1)
of the W -weight (4.30) to estimate the Im,n;k -term as follows

(2,1)
|Im,n;k | ≤2a2m,n ν 2 k|∂y W |∂y ω̊m,n;k eW χm+n kL2y k∂y2 ω̊m,n;k eW χm+n kL2y
8 p
≤a2m,n √ −∂t W ν 1/2 ∂y ω̊m,n;k eW χm+n 2 kν∂y2 ω̊m,n;k eW χm+n kL2y
K Ly
C (α;W ) C
≤ 1 CKm,n;k + 1 a2m,n ν 2 k∇k ∂y ω̊m,n;k eW χm+n k2L2y . (4.24)
K2 K2
If K is chosen larger than some universal constants, these two terms can be absorbed by the negative
terms in (4.21) and (4.23).
(2,2)
Next we address the important boundary contribution from above, Im,n;k , for which we invoke
Lemma 4.14, identity (4.35), which shows that

λ2s  
|Im,n;k | . 1n=1
(2,2) (α) (α)
Dm,0;k + CK m,0;k .
(m + 1)2σ∗
(2,3)
Next, we treat the term Im,n;k in (4.23). The idea is similar to the one applied in (4.18).
(1,1,0)
However, this time, we need to invoke a technical estimate (5.1c) on the function Sm,n fk :=
q ∂y (|k| q Γk ωk )1S2n (1, 1, 0) (3.6a). Combining it with the bounds on the cut-off functions
m+n m n n

(2.5a), we have that

  
(2,3)
|Im,n;k | ≤Ccα am,n kν∂y2 ω̊m,n;k eW χm+n kL2y am,n (m + n)1+σ kν∂y ω̊m,n;k eW χm+n−1 kL2y

28
 cα 2
≤ a ν 2 k∂y2 ω̊m,n;k eW χm+n k2L2y
8 m,n
λ2s 2
+ Ca2m−1,0;k ν 2 2σ ∂y |k|ωm−1,0;k eW χm−1 L2 1m≥1,n=0
m y
2s
λ ϕ 2
2
+ Ca2m,0;k ν 2 ∂y (q(vy−1 ∂y + ikt)ω̊m,0;k )eW χm L2y
1m≥0,n=1
(m + 1)2σ∗
2
m+n
+ Ca2m,n ν 2 (m + n)2σ ∂y ω̊m,n;k eW χm+n−1 1n≥2
q L2y

cα Cλ2s  
≤ a2m,n ν 2 k∂y2 ω̊m,n;k eW χm+n k2L2y + D
(α)
1
m,n−1;k n≥1 + D
(µ)
1
m,n−1;k n≥1
8 (m + n)2σ∗
Cλ2s (α)
+ 2σ∗ Dm−1,0;k 1m≥1 1n=0 . (4.25)
m
(3)
Finally, we estimate the term Im,n;k in (4.21) as follows,

(3) 1
|Im,n;k | ≤cα a2m,n k1 + ∂y U0x kL∞
t,y
kν 1/2 |k|ω̊m,n;k eW χm+n kL2 kν 2 ∂y ω̊m,n;k eW χm+n kL2
(γ)
≤cα a2m,n k1 + ∂y U0x kL∞
t,y
Dm,n;k . (4.26)

Combining the decomposition (4.21), (4.23), and the estimates (4.22), (4.24), (4.25), (4.26), we
have obtained (4.20). This concludes the proof of the lemma.

4.4 Proof of Proposition 2.5

Proposition 2.5 is formalized as follows:
1
Proposition 4.9. If λ (2.8) is small enough compared to universal constants and cα = 8 in (4.1),
the following estimate holds,
∞ ∞ ∞
d X 1 X X (ι) 1 X X X (ι,ζ)
Em,n;k + Dm,n;k + CKm,n;k
dt 9 9
m+n=0 m+n=0 ι∈{α,µ,γ} m+n=0 ι∈{α,µ,γ} ζ∈{ϕ,W }
∞
X X (ι)
. Im,n;k , (4.27)
m+n=0 ι∈{α,µ,γ}

where we define the following inner products

D  E
(γ) (m,n) (m,n) (m,n) (m,n)
Im,n,k :=a2m,n Re Tk + Ctrans;k + Cvisc;k + Cq,k , ω̊m,n;k e2W χ2m+n , (4.28a)
D   E
(α) (m,n) (m,n) (m,n) (m,n)
Im,n,k :=a2m,n νRe ∂y Tk + Ctrans;k + Cvisc;k + Cq,k , ∂y ω̊m,n;k e2W χ2m+n ,
D   E
(µ) (m,n) (m,n) (m,n) (m,n)
Im,n,k :=a2m,n νRe |k| Tk + Ctrans;k + Cvisc;k + Cq,k , |k|ω̊m,n;k e2W χ2m+n . (4.28b)

Proof. To obtain the estimate (4.27), we sum over (4.13), (4.19) and (4.20)
∞ ∞
d X X (α)
X (α,ζ)
Em,n;k + cα Dm,n;k + cα CKm,n;k
dt
m+n=0 m+n=0 ζ∈{ϕ,W }
∞
X X X X
(ι) (ι,ζ)
+ Dm,n;k + CKm,n;k
m+n=0 ι∈{µ,γ} ι∈{µ,γ} ζ∈{ϕ,W }

29
  
∞
X X X
(γ) D (ι) (ι,ζ)
≤2cα k1 + ∂y U0x kL∞
t,y
Dm,n;k + Cλ2s m,n;k +C CKm,n;k 
m+n=0 ι∈{α,µ,γ} ζ∈{ϕ,W }
∞
X X (ι)
+C Im,n;k .
m+n=0 ι∈{α,µ,γ}

Then we recall the assumption (2.22d), choose cα = 18 and λ (2.8) small enough so that the first
two terms on the right hand side are absorbed by the corresponding terms on the left hand side.
This concludes the proof.

4.5 Technical Lemmas: Auxiliary Bounds for Energy Estimates

We first record the following important inequality relating the losses we incur from our Gevrey
cut-off scale, χm+n (2.4), with the Gevrey weights Bm,n :
Lemma 4.10. Fix ℓ ∈ {1, 2}. The weights {Bm,n }m+n≥ℓ (2.9) satisfy the following bound,

(m + n)ℓ(1+σ) Bm,n ≤C(s)λℓs (m + n)−ℓσ∗ Bm−ℓ1 ,n−ℓ2 , ℓ1 + ℓ2 = ℓ. (4.29)

Here the parameters σ, σ∗ are defined in (2.1) and the constant C(s) ≤ 4 for s ∈ [1, 2]. Moreover,
the (ℓ1 , ℓ2 ) is chosen such that m − ℓ1 ≥ 0, and n − ℓ2 ≥ 0.
Remark 4.11. For m + n < ℓ, the coefficients Bm,n are of order 1 and one can estimate the
coefficient directly.
Proof. Recalling the Bm,n -definition (2.9) and the relation σ + σ∗ = s − 1, which is a consequence
of the definition (2.1), we have that
 m+n s  m+n s
1+σ 1+σ λ s−σ∗ λ
(m + n) Bm,n =(m + n) = (m + n)
(m + n)! (m + n)!
 m+n−1 s
λ
=(m + n)−σ∗ λs = (m + n)−σ∗ λs Bm−ℓ1 ,n−ℓ2 .
(m + n − 1)!
Here ℓ1 + ℓ2 = 1. By invoking the estimate obtained again, we have that for m + n ≥ 2,

(m + n)2+2σ Bm,n ≤ λs (m + n)1+σ (m + n)−σ∗ Bm−ℓ′1 ,n−ℓ′2 1ℓ′1 +ℓ′2 =1

 1+σ+σ∗
m+n
≤λ 2s
(m + n)−2σ∗ Bm−ℓ1 ,n−ℓ2 1ℓ1 +ℓ2 =2
m+n−1
C(s)λ2s
≤ Bm−ℓ1 ,n−ℓ2 1ℓ1 +ℓ2 =2 .
(m + n)2σ∗
This concludes the proof of the lemma.

Lemma 4.12. Recall the function W from (2.7),

(|y| − 1/4 − Lǫ arctan(t))2+

W (t, y) = .
Kν(1 + t)
The function W satisfies the following estimate:

1 (|y| − 1/4 − Lǫ arctan(t))2+ 2Lǫ(|y| − 1/4 − Lǫ arctan(t))+

∂t W + Kν|∂y W |2 ≤ − − ≤ 0.
(4.30)
8 2Kν(1 + t)2 Kν(1 + t)(1 + t2 )

30
Here K, L ≥ 1 are large constants. Alternatively, there exists a constant C, independent of the
parameters K, L such that
C
ν|∂y W |2 ≤ − ∂t W ;
K
ǫ C
|∂y W | ≤ − ∂t W.
(1 + t)2 L

Proof. We take the time derivative and spatial derivative to obtain that

K (|y| − 1/4 − Lǫ arctan(t))2+ 2Lǫ(|y| − 1/4 − Lǫ arctan(t))+

∂t W + ν|∂y W |2 = − −
8 Kν(1 + t)2 Kν(1 + t)(1 + t2 )
2
(|y| − 1/4 − Lǫ arctan(t))+
+ Kν
2K 2 ν 2 (1 + t)2
(|y| − 1/4 − Lǫ arctan(t))2+ 2Lǫ(|y| − 1/4 − Lǫ arctan(t))+
≤− − ≤ 0.
2Kν(1 + t)2 Kν(1 + t)(1 + t2 )

Lemma 4.13. The following two estimates hold for arbitrary M ∈ N,

M
X M
X (γ)
a2m,n νk∂y (χm+n ω̊m,n;k )eW k2L2 . Dm,n;k ; (4.32)
m+n=0 m+n=0
M
X M
X M
X
(α) (µ)
a2m,n ν 2 k∇k ∂y (χm+n ω̊m,n;k )eW k2L2 . Dm,n;k + Dm,n;k . (4.33)
m+n=0 m+n=0 m+n=0

Proof. We start by proving (4.32). We recall that χ0 ≡ 1, am,n = Bm,n ϕn+1 and the χ′m+n estimate
(2.5a) to obtain
M
X
a2m,n νk∂y (χm+n ω̊m,n;k )eW k2L2
m+n=0
M
X M
X
. a2m,n νkχm+n (∂y ω̊m,n;k )eW k2L2 + a2m,n νk[∂y , χm+n ]ω̊m,n;k eW k2L2
m+n=0 m+n=0
M
X M
X
.
(γ)
Dm,n;k + 1n≥1 a2m,n (m + n)2+2σ νkχm+n−1 (∂v + ikt)ωm,n−1;k eW k2L2
m+n=0 m+n=0
M
X
+ a2m,0;k m2+2σ νkχm−1 |k|ωm−1,0;k eW k2L2 .
m=1

Here we apply the definition Γ = ∂v + ikt in the last inequality. Now, we invoke a technical relation
of Bm,n (4.29) and end up with the result:
M
X M
X M
X −1
(γ) (γ)
a2m,n νk∂y (χm+n ω̊m,n;k )eW k2L2 . Dm,n;k + λ2s (m + n)−2σ∗ Dm,n;k
m+n=0 m+n=0 m+n=0

proving (4.32).

31
 P 2
Next we focus on (4.33). First of all, we observe that the am,n νkk∂y (χm+n ω̊m,n;k )eW k2L2 can
be estimated in the same fashion as in (4.32). Now we focus on the following
M
X
a2m,n ν 2 k∂yy (χm+n ω̊m,n;k )eW k2L2
m+n=0
M
X M
X
. a2m,n ν 2 kχm+n (∂yy ω̊m,n;k )eW k2L2 + a2m,n ν 2 k[χm+n , ∂yy ]ω̊m,n;k eW k2L2
m+n=0 m+n=0
M
X M
X
(α)
. Dm,n;k + a2m,n (m + n)4+4σ ν 2 kχm+n−2 ω̊m,n;k eW k2L2
m+n=0 m+n=2
M
X X
+ a2m,n (m + n)2+2σ ν 2 kχm+n−1 ∂y ω̊m,n;k eW k2L2 + a2m,n ν 2 k[χm+n , ∂yy ]ω̊m,n;k eW k2L2
m+n=2 m+n=1
4
X
=: Tj . (4.34)
j=1

The T1 term is consistent with the result. We focus on the T2 -term. By invoking the definition
am,n = Bm,n ϕn+1 , the property (2.11), the bound kvy−1 kL∞ t,y
+ kvy kL∞ Wy1,∞ ≤ C, and the technical
t
relation (4.29), we obtain that there exists an ℓ ∈ {0, 1, 2} such that the following bound holds
M
X
T2 . a2m,n (m + n)4+4σ 1{m−ℓ≥0, n−2+ℓ≥0} ν 2 kχm+n−2 (∂v + ikt)2−ℓ |k|ℓ ωm−ℓ,n−2+ℓ;k eW k2L2
m+n=0
M
X
.λ 4s
a2m−ℓ,n−2+ℓ;k 1{m−ℓ≥0, n−2+ℓ≥0} (m + n)−4σ∗ ϕ4−2ℓ ν 2
m+n=0

× kχm+n−2 (∂v + ikt)2−ℓ |k|ℓ ωm−ℓ,n−2+ℓ;k eW k2L2

M
X −2 M
X −2
(α) (µ)
.λ4s a2m,n ν 2 kχm+n ∇2k ω̊m,n;k eW k2L2 . λ4s (Dm,n;k + Dm,n;k ).
m+n=0 m+n=0

The estimate for the T3 -term is similar and we obtain that

M
X −1
(α) (µ)
T3 . λ2s (Dm,n;k + Dm,n;k ).
m+n=0

For the T4 -term, we have that

X X
T4 . a2m,n ν 2 k[χ1 , ∂yy ]ω̊m,n;k eW k22 = a2m,n ν 2 (k∂y ω̊m,n;k eW k22 + kω̊m,n;k eW k22 )
m+n=1 m+n=1
(α) (µ)
.λ2s (D0,0;k + D0,0;k ).

Combining the estimates above with the decomposition (4.34), we obtain (4.33).

We record here a lemma which establishes that the top order boundary contribution from our
α estimate, vanishes.

32
Lemma 4.14. For all n ∈ N, k ∈ Z, we have

Re∂y ω̊m,n;k ∂y2 ω̊m,n;k |y=±1 = 0, n 6= 1. (4.35)

For n = 1, the boundary contribution can be estimated as follows

a2m,1 ν 2 Re∂y (ω̊m,1;k )∂y2 (ω̊m,1;k )e2W χ2m+1 y=±1

. a2m,1 ν 2 |∂y ω̊m,0;k |2 e2W
y=±1
λ2s (α) λ2s (α,W )
. D + CKm,0;k . (4.36)
(m + 1)2σ∗ m,0;k (m + 1)2σ∗

Here, the D-term and the CK-term are defined in (4.2), (4.3b).

Proof. For n > 1, the result is a direct consequence of the fact that


∂y ω̊m,n;k = ∂y (q n Γnk ωk ) = (n − 1)q n−1 q ′ Γnk ωk + q n ∂y Γn ωk = 0.
y=±1 y=±1 y=±1

For n = 0, we recall that the boundary condition ωk (t, y = ±1) = 0 implies ∂yy ωk (t, y = ±1) = 0
and guarantees that the left hand side of (4.35) is zero. The n = 1 case is the most delicate one, we
invoke the properties q(±1) = q ′′ (±1) = 0, ωk (t, y = ±1) = 0 and ∂yy ωk (t, y = ±1) = 0 to obtain
that

Re∂y ω̊m,1;k ∂yy ω̊m,1;k e2W χ2m+1

y=±1



=2Req ′ vy−1 ∂y + ikt (|k|m ωk )q ′ ∂y vy−1 ∂y + ikt (|k|m ωk )e2W χ2m+1
y=±1
 
−vyy
=2Req ′ vy−1 ∂y (|k|m ωk )q ′ ∂y (|k| ωk ) + vy ∂yy (|k| ωk ) + ikt∂y (|k| ωk ) e2W χ2m+1
m −1 m m
vy2 y=±1
vyy vyy
= − 2 3 |q ′ |2 |∂y (|k|m ωk )|2 e2W χ2m+1 = −2 3 |q ′ |2 |∂y ω̊m,0;k |2 e2W χ2m+1 .
vy y=±1 vy y=±1


Here we have used the fact that Re ikt|∂y (|k|m ωk )|2 = 0 in the last line. Combining this relation
and the fact that k v1y k∞ + kvyy k∞ ≤ C yields the first inequality in (4.36).
Now we develop the second inequality in (4.36). We use the following trace theorem, which is
a natural consequence of integration by parts:

a2m,1 ν 2 |∂y ω̊m,0;k |2 e2W χ2m+1

y=±1
.am,1 ν k∂y ω̊m,0;k e χm+1 k2L2
2 2 W
+ a2m,1 ν 2 k∂y (∂y ω̊m,0;k eW χm+1 )k2L2 .

We further expand the second term with the property (2.5a) as follows

a2m,1 ν 2 |∂y ω̊m,0;k |2 e2W χ2m+1

y=±1
.am,1 ν k∂y ω̊m,0;k e χm+1 k2L2
2 2 2 W
+ a2m,1 ν 2 (m + 1)2+2σ k∂y ω̊m,0;k eW k2L2 (supportχm+1 )
+ a2m,1 νk∂y ω̊m,0;k |ν 1/2 ∂y W |eW χm k22 =: T1 + T2 + T3 .

33
Now we distinguish between two cases: k 6= 0 and k = 0. If k 6= 0, the first two terms can be
bounded by D (α) . Then the relation (4.30) yields that

λ2s (α)
p
a2m,1 ν 2 |∂y ω̊m,0;k |2 e2W χ2m+1 . Dm,0;k + a 2
m,1 νk∂y ω̊ m,0;k −∂t W eW χm k22
y=±1 (m + 1)2σ∗
λ2s (α) λ2s (α,W )
. 2σ
Dm,0;k + 2σ+2σ
CKm,0;k .
(m + 1) ∗ (m + 1) ∗

On the other hand, if k = 0, then m = 0 and we observe that by (4.30),

p 1 1
−∂t W ≥ ≥ , y ∈ supportχ1 , t ≤ ν −1/3−η .
CKν(1 + t)2 CKν 1/3−2η
Hence,
p (α,W )
T2 . a20,1 ν 2 k∂y ω̊0,0;0 −∂t W eW k2L2 (supportχ1 ) . νλ2s CK0,0;0 .

The T1 , T3 terms are estimated in an identical fashion as before. Combining the estimates in the
cases k 6= 0 and k = 0, we obtain the second inequality in (4.36). Hence the proof is concluded.

5 Commutator Estimates
In this section, we collect all of the commutator bounds we need in order to close our energy scheme.
In particular, the main result of this section will be the proof of Proposition 2.6.
(a,b,c)
5.1 Bounds on the Building Blocks Sm,n ωk
This subsection is devoted to the proof of the following Lemma 5.1. Along the way, we need a
technical combinatorial Lemma 5.21.

Lemma 5.1. If 1 > λ > 0 is smaller than a universal constant, the quantities Sm,n ωk , defined in
(3.8a) satisfies the following bounds:
n−1
X  −2(σ+σ∗ )
2 (m + n)! (γ)
a2m,n ν (1,0,0)
Sm,n ωk eW χm+n−1 . s 2(n−ℓ)
(Cλ ) Dm,ℓ;k , (5.1a)
L2 (m + ℓ)!
ℓ=0
M
X M
X −1
2 (γ)
a2m,n ν (1,0,0)
Sm,n ωk eW χm+n−1 .λ 2s
Dm,ℓ;k . (5.1b)
L2
m+n=0 m+n=0

For the j = 2 level,

n−1
X  −2σ−2σ∗
2 (m + n)! (α) (µ)
a2m,n ν 2 (2)
Sm,n ωk eW χm+n−1 . s 2n−2ℓ
(Cλ ) (Dm,ℓ;k + Dm,ℓ;k(5.1c)
),
L2 (m + ℓ)!
ℓ=0
M
X 2
M
X −1  
(α) (µ)
a2m,n ν 2 Sm,n
(2)
ωk eW χm+n−1 .λ2s Dm,n;k + Dm,n;k . (5.1d)
L2
m+n=0 m+n=0

Here D-terms are defined in (4.2). The implicit constants and the constant C’s depend only on
s ∈ (1, 2].

34
Proof. We decompose the proof into several components.
Step # 1: Proof of (5.1a), (5.1b). We proceed by induction. For the base case, n = 0, we
(1,0,0)
trivially have Sm,0 ωk = 0. We now record the inductive identity, valid for all n ≥ 1,

(1,0,0) m+n

Sm,n ωk := ω̊m,n;k = (m + n)q n−1 |k|m Γnk ωk = (m + n)q n−1 vy−1 ∂y + ikt Γkn−1 |k|m ωk
q


=(m + n)vy−1 ∂y q n−1 Γkn−1 |k|m ωk − (m + n)(n − 1)vy−1 q ′ q n−2 Γkn−1 |k|m ωk
+ (ikt)(m + n)q n−1 Γkn−1 |k|m ωk
(m + n) (1,0,0)
=(m + n)vy−1 ∂y ωm,n−1;k − (n − 1)vy−1 q ′ Sm,n−1 ωk + (ikt)(m + n)ωm,n−1;k .
m+n−1
We recall the definition of am,n (2.12), the bound tϕ ≤ C (2.11), with the bound (4.29) and estimate
the right hand side above as follows,
1
(1,0,0)
am,n ν 2 kSm,n ωk eW χm+n−1 kL2
1
.(m + n)am,n vy−1 L∞
ν 2 ∂y ω̊m,n−1;k eW χm+n−1
L2
1 (1,0,0)
+ (m + n)am,n vy−1 q ′ L∞
ν 2 Sm,n−1;k eW χm+n−1 2
L
1
+ (m + n)Bm,n ϕm+n−1 (ϕt) ν 2 |k|ωm,n−1;k eW χm+n−1 2
L
q
(γ) 1 (1,0,0)
s −σ−σ∗ s −σ−σ∗
.λ (m + n) Dm,n−1;k + λ (m + n) am,n−1;k ν 2 Sm,n−1 ωk eW χm+n−2 .
L2

Here we have invoked the bound (2.22e) to control vy−1 L∞ . Now we have that there exists a
constant C, independent of m, n, such that iteration of the above inequality yields that for λ small
compared to universal constants,
n−1  
X (m + n)! −2(σ+σ∗ ) (γ)
a2m,n νkSm,n
(1,0,0)
ωk eW χm+n−1 k2L2 ≤ (Cλs )2(n−ℓ) Dm,ℓ;k .
(m + ℓ)!
ℓ=0

This is (5.1a). Next, we have that by Lemma 5.21,

M
X XM MX−m n−1
X  −2(σ+σ∗ )
s 2(n−ℓ) (m + n)! (γ)
a2m,n νkSm,n
(1,0,0)
ωk eW
χm+n−1 k2L2 . (Cλ ) Dm,ℓ;k
m+n=0 m=0 n=0
(m + ℓ)!
ℓ=0
M M −m−1
X X MX −m M
X −1
(γ) (γ)
. (Cλs )2(n−ℓ) Dm,ℓ;k . λ2s Dm,n;k . (5.2)
m=0 ℓ=0 n=ℓ+1 m+n=0

This is (5.1b). Hence the first step is concluded.

(1,0,1)
Step # 2: Estimate of Sm,n ωk . These estimates are natural consequences of multiplying the
(γ) (µ)
bounds (5.1a), (5.1b) by ν|k|, and applying the definitions (4.2) ν|k|Dm,n;k = Dm,n;k . The explicit
estimate is as follows
n−1  
2 X (m + n)! −2(σ+σ∗ ) (µ)
a2m,n ν 2 Sm,n
(1,0,1)
ωk eW χm+n−1 2 . (Cλs )2(n−ℓ) Dm,ℓ;k , (5.3)
L (m + ℓ)!
ℓ=0
M
X M
X −1
2 (µ)
a2m,n ν 2 Sm,n
(1,0,1)
ωk eW χm+n−1 .λ2s Dm,ℓ;k . (5.4)
L2
m+n=0 m+n=0

35
This concludes Step # 2.
(2,0,0)
Step # 3: Estimate of Sm,n ωk . We proceed inductively. The base case is n = 2, we estimate
the term as follows:
 2 
(2,0,0)
Sm,2 ωk =(m + 2)2 ∂v + 2ikt∂v − k2 t2 |k|m ωk .

Now we have that by the definition of am,2 (2.12), the Gevrey coefficient bound (4.29), the vy -bound
(α) (µ)
(2.22e), tϕ ≤ C (2.11), and the definition of Dm,n;k , Dm,n;k (4.2), the following estimate holds

(2,0,0)
am,2 ν Sm,2 ωk eW χm+1
2
.am,2 (m + 2) ν ∇k ∂y |k|m ωk eW χm+1
2
+ Bm,2 (m + 2)2 ϕ2 ν(tϕ)k|k|∂y |k|m ωk eW χm+1 k2
2
 1/2
(α) (µ)
+ Bm,2 (m + 2)2 ϕν(tϕ)2 k|k|2+m ωk eW χm+1 k2 . Dm,0;k + Dm,0;k . (5.5)

For n > 2, we open up two Γk -derivatives to obtain the expression

 
(2,0,0) 2
Sm,n ωk =(m + n)2 q n−2 ∂v + 2ikt∂v − k2 t2 Γkn−2 |k|m ωk
h i
2 2
=(m + n)2 ∂v (q n−2 Γkn−2 |k|m ωk ) + (m + n)2 q n−2 , ∂v Γkn−2 |k|m ωk

 
+ 2(m + n)2 ikt∂v q n−2 Γkn−2 |k|m ωk + 2(m + n)2 ikt q n−2 , ∂v Γkn−2 |k|m ωk


− (m + n)2 |k|2 t2 q n−2 Γkn−2 |k|m ωk
5
X
=: T3;j . (5.6)
j=1

Here, ‘3’ stands for Step # 3 and ‘j’ denotes the ‘j’-th term. We start by estimating the T3;1 term
in (5.6). By further expanding the expression, we have that
 
W 2 1 vyy
am,n νkT3;1 e χm+n−1 k2 = am,n ν (m + n) ∂yy ωm,n−2;k − 3 ∂y ω̊m,n−2;k eW χm+n−1
vy2 vy 2
2 W W


.(m + n) am,n ν k∂yy ω̊m,n−2;k e χm+n−2 k2 + k∂y ω̊m,n−2;k e χm+n−2 k2 .

Note that the definition of am,n (2.12) and the estimate (4.29) yields that

Cλ2s Bm,n−2 ϕm+n−2 Cλ2s

(m + n)2 am,n =(m + n)2 Bm,n ϕm+n ≤ ≤ am,n−2;k .
(m + n)2(σ+σ∗ ) (m + n)2(σ+σ∗ )
(α)
Combining this estimate and the vy -bound (2.22e), and the definition of the dissipation Dm,n;k (4.2)
yields that
(α)
W Cλ2s am,n−2 W
Cλs Dm,n−2;k
am,n νkT3;1 e χm+n−1 k2 ≤ νk∇ k ∂y ω̊ m,n−2;k e χ m+n−2 k2 ≤ .(5.7)
(m + n)2(σ+σ∗ ) (m + n)2(σ+σ∗ )

To estimate the T3;2 term, we apply the commutator relation (4.7e) (n−2 ≥ 2), (4.7f) (n−2 = 1),
and the definition of am,n (2.12), the bound (4.29), the bound of kvy−1 k∞ , kvyy k∞ , and the bound
ϕt ≤ C (2.11), to derive that,

am,n νkT3;2 eW χm+n−1 k2

36
 X
.1n≥4 (m + n)2 am,n ν kSm,n−2 ωk eW χm+n−1 k2 + 1n=3 (m + 3)2 am,3 νkΓk ωk eW χm+2 kH 1
(a,b,c)

a+b+c=2,
a6=0
q
λ2s am,n−2 ν X λ2s 1n=3
.1n≥4
(a,b,c) (α) (µ)
kSm,n−2 ωk eW χm+n−1 k2 + Dm,0;k + Dm,0;k . (5.8)
(m + n)2(σ+σ∗ ) (m + 3)2σ+2σ∗
a+b+c=2,
a6=0

Next we recall the definition of am,n (2.12), the bound (4.29), the bound of kvy−1 k∞ (2.22e), and
the bound ϕt ≤ C (2.11), and then estimate the T3;3 as follows

am,n νkT3;3 eW χm+n−1 k2 . (m + n)2 Bm,n ϕm+n ν(ϕt)k|k|∂y ω̊m,n−2;k eW χm+n−2 k2

q
λ2s Bm,n−2 ϕm+n W λ2s (α)
. 2(σ+σ )
νk|k|∂y ω̊m,n−2;k e χm+n−2 k2 . 2(σ+σ )
Dm,n−2;k . (5.9)
(m + n) ∗ (m + n) ∗

To estimate the T3;4 term, we invoke the commutator relation (4.7d) to get that
 
2 q ′ n − 2 m n−2 n−2
T3;4 = − 2(m + n) ikt |k| q Γ ωk
vy q
(m + n)2 (n − 2) k q ′ (1,0,1) q′
=−2 i t Sm,n−2 ωk 1n≥4 − 2(m + 3)2 ikt (∂v + ikt)|k|m ωk 1n=3 .
m+n−2 |k| vy vy

Then we have that by the bound of kvy−1 k∞ , kq ′ k∞ , the estimate (5.3), the definition of am,n (2.12),
the bound (4.29), and the bound ϕt ≤ C (2.11), the following estimate holds

am,n νkT3;4 eW χm+n−1 k2

.(m + n)2 am,n νtkSm,n−2 ωk eW χm+n−1 k2 1n≥4 + (m + 3)2 am,3 νkkΓk |k|m ωk eW χm k2 1n=3
(1,0,1)

q
λ2s am,n−2 λ2s
. νkS
(1,0,1)
m,n−2 k ω eW
χ k 1
m+n−1 2 n≥4 + Dm,0;k + Dm,0;k 1n=3
(α) (µ)
(5.10)
.
(m + n)2(σ+σ∗ ) (m + 3)2(σ+σ∗ )

The estimate of the T3;5 contribution is similar

am,n νkT3;5 eW χm+n−1 k2 . (m + n)2 am,n νt2 k|k|2 ωm,n−2;k eW χm+n−1 k2

λ2s am,n−2 2 W λ2s am,n−2 (µ)
. νk|k| ω m,n−2;k e χ k
m+n−2 2 . (D )1/2 . (5.11)
(m + n)2(σ+σ∗ ) (m + n)2(σ+σ∗ ) m,n−2;k

Summarizing the decomposition (5.6), the estimates (5.5), (5.7), (5.8), (5.9), (5.10) and (5.11), we
obtain that
q
(2,0,0) (α) (µ)
am,2 ν Sm,2 ωk eW χm+1 . λ2s Dm,0;k + Dm,0;k ;
2

am,n ν (2,0,0)
Sm,n e χm+n−1 1n>2
ωk W
2
λ2s  1/2 λ2s  1/2
. D
(α)
m,n−2;k + D
(µ)
m,n−2;k 1n>2 + D
(α)
m,0;k + D
(µ)
m,0;k 1n=3
(m + n)2σ+2σ∗ (m + 3)2σ+2σ∗
λ2s am,n−2 X
kSm,n−2 ωk eW χm+n−3 k2 1n>2 .
(a,b,c)
+ 2σ+2σ
ν (5.12)
(m + n) ∗
a+b+c=2
a6=0

37
 (1,1,0) (1,1,0) (1,1,0)
Step 4: Iteration relation of Sm,n ωk . First of all, we recall that Sm,1 ωk = Sm,0 ωk = 0.
For n = 2, we recall the commutator relation (4.7d) and derive the following

(1,1,0) (m + 2)


Sm,2 ωk = ∂y q 2 Γ2k |k|m ωk = (m + 2)q −1 ∂y q[q, Γk ]Γk |k|m ωk + q(∂v + ikt)(qΓk |k|m ωk )
q


=(m + 2)q −1 ∂y −(q ′ vy−1 q)Γk ωk + (qvy−1 )∂y (qΓk |k|m ωk ) + (m + 2)qikt(qΓk |k|m ωk )


=(m + 2) −q −1 ∂y (q ′ vy−1 q)Γk |k|m ωk − q ′ vy−1 ∂y (∂v + ikt)|k|m ωk


+ (m + 2) vy−1 ∂yy (ω̊m,1;k ) + q −1 ∂y (qvy−1 )(q ′ Γk |k|m ωk + q∂y (∂v + ikt)|k|m ωk )
3
X
′ −1


+ (m + 2) q ikt(q ω̊m,1;k ) + ikt∂y ω̊m,1;k =: T4;j . (5.13)
j=1

We invoke the bound of kvy−1 k∞ , kvyy k∞ , kq ′ k∞ , the estimate (5.3), the definitions of am,n (2.12)
and D (4.2), and the bounds (2.11), (4.29), (5.1a), to obtain the following estimates

am,2 νkT4;1 eW χm+1 k2


(1,0,0)
.(m + 2)am,2 ν k∂y (q ′ vy−1 q)k∞ kSm,1 ωk eW χm+1 k2 + kq ′ vy−3 vyy k∞ k∂y |k|m ωk eW χm+1 k2

′ −2 m W ′ −1 m W
+ kq vy k∞ k∂yy |k| ωk e χm+1 k2 + tkq vy k∞ k|k|∂y |k| ωk e χm+1 k2
q
s λ 2s D (α)
λ am,1 (1,0,0) W m,0;k λ2s Bm,0 ϕ2 (tϕ)
. νkS m,1 ω k e χ m+1 k2 + + νk|k|∂y ω̊m,0;k eW χm+1 k2
(m + 2)σ+σ∗ (m + 2)2σ+2σ∗ (m + 2)2σ+2σ∗
λs (α)
. σ+σ
(Dm,0;k )1/2 .
(m + 2) ∗

Application of similar arguments yields the estimates for the other two terms in (5.13), i.e.,

am,2 νkT4;2 eW χm+1 k2


(1,0,0)
. (m + 2)am,2 ν k∂yy ω̊m,1;k eW χm+1 k2 + kSm,1 ωk eW χm+1 k2 + k∂y |k|m ωk eW χm+1 k2

+ k∂y (qvy−1 )k∞ kvy−1 k∞ k∂yy |k|m ωk eW χm+1 k2 m W
+ tk|k|∂y |k| ωk e χm+1 k2
(α) (α) (µ)
. (Dm,1;k + Dm,0;k + Dm,0;k )1/2 ;

 
W (1,0,0) W W
am,2 νkT4;3 e χm+1 k2 .(m + 2)am,2 ν tk|k|Sm,1 ωk e χm+1 k2 + tk|k|∂y ω̊m,1;k e χm+1 k2
(µ) (α)
. (Dm,1;k + Dm,0;k )1/2 .

Combining the estimates we developed and the decomposition (5.13), we obtain that
1
X X
(1,1,0) (ι)
a2m,2 ν 2 kSm,2 ωk eW χm+1 k22 . Dm,ℓ;k . (5.14)
ℓ=0 ι∈{α,µ}

For n > 2, we have that

m+n m+n

(1,1,0)
Sm,n ωk = ∂y (q n Γnk |k|m ωk ) = ∂y q[q n−1 , Γk ]Γkn−1 |k|m ωk + qΓk (q n−1 Γkn−1 |k|m ωk ) .
q q

38
Hence we recall the commutator relation (4.7d) and obtain that
 ′ 
(1,1,0) m+n q n−1 n−1 m

n−1 n−1 m
Sm,n;k ωk = ∂y − (n − 1)q Γk |k| ωk + q ∂v + ikt (q Γk |k| ωk )
q vy
(m + n)(n − 1) ′ −1

(m + n − 1)2
=− (q∂y q vy ) ω̊m,n−1;k
(m + n − 1)2 q2
| {z }
(2,0,0)
=Sm,n−1 ωk

(m + n)(n − 1)q ′ (m + n − 1)
− ∂y ω̊m,n−1;k
(m + n − 1)vy q
| {z }
(1,1,0)
=Sm,n−1 ωk
m+n
m+n−1
+ ∂y qvy−1 ∂y ω̊m,n−1;k +(m + n)vy−1 ∂yy ω̊m,n−1;k
m+n−1 q
| {z }
(1,1,0)
=Sm,n−1 ωk

m + n ′ k (m + n − 1)
+ (m + n)ikt∂y ω̊m,n−1;k + qi t |k|ω̊m,n−1;k .
m + n − 1 |k| q
| {z }
(1,0,1)
=Sm,n−1 ωk

The treatment of these terms are similar to the treatment in Step 1. We invoke the bound of
kvy−1 k∞ , kqvyy k∞ , kq ′ k∞ , the estimate (5.3), the definition of am,n (2.12), and the bound (4.29),
to obtain the following estimates for n ≥ 3,
(1,1,0)
am,n νkSm,n ωk eW χm+n−1 k2
λs (α) (µ) λs am,n−1 X (1+a,b,0)
1/2
. (Dm,n−1;k + D m,n−1;k ) + ν kSm,n−1 ωk eW χm+n−2 k(5.15)
2.
(m + n)σ+σ∗ (m + n)σ+σ∗
a+b=1

Step 5: Iteration. Application of (5.3), (5.4), (5.12), (5.14) and (5.15) gives us (5.1c). Now an
argument as in (5.2) yields (5.1d).
(m,n)
5.2 Commutator Estimates, Cq,k
(m,n)
We now obtain commutator bounds on the various contributions of Cq,k which appear in our
γ/µ/α energy scheme. First, we have

Lemma 5.2. The following bound is valid:

D E X n−1  −2σ−2σ∗
(m,n) 1 (γ) (m + n)! (γ)
a2m,n Re ω̊m,n;k , Cq,k e2W χ2m+n ≤ Dm,n;k + CB (Cλs )2n−2ℓ (5.16)
Dm,ℓ;k .
B (m + ℓ)!
ℓ=0

Here B > 1 is an arbitrary constant and C is universal constant.

Proof. We distinguish between the n = 1 case and the n ≥ 2 case.

(1,0,0)
For the n = 1 case, we invoke the definition of Sm,1 ωk (3.6a) and have that

(m,1) (1,0,0)
Cq,k = ν[q, ∂yy ]Γk |k|m ωk = − 2νq ′ ∂y Γk |k|m ωk − νq ′′ Γk |k|m ωk = −2νq ′ ∂y Γk |k|m ωk − νq ′′ Sm,1 ωk .
(5.17)

39
 (1,0,0)
Next we invoke the definition D (4.2), together with the bound of Sm,1 ωk (5.1a), to estimate the
left hand side of (5.16) as follows
D E
(m,1)
a2m,1 Re ω̊m,1;k , Cq,k e2W χ2m+1
D 
 E
(1,0,0)
=a2m,1;k ν Re ω̊m,1;k , −2νq ′ ∂y q −1 ω̊m,1;k − νq ′′ Sm,1 ωk e2W χ2m+1
.a2m,1 νkq −1 ω̊m,1;k eW χm+1 k2 k∂y ω̊m,1;k eW χm+1 k2 + a2m,1;k νkq −1 ω̊m,1;k eW χm+1 k22
(γ) (1,0,0) (γ)
.Dm,0;k + a2m,1;k νkSm,1 ωk eW χm+1 k22 . Dm,0;k .
This is estimate (5.16) for n = 1.
We invoke (4.5b)n≥2 to produce the identity
D E
(m,n)
a2m,n Re ω̊m,n;k , Cq,k e2W χ2m+n
   
2 ′n ′ 2 n(n + 1) ′′ n 2W 2
= am,n Re ω̊m,n;k , −2νq ∂y ωm,n;k + ν(q ) ωm,n;k − νq ωm,n;k e χm+n
q q2 q
3
X
=: Tj . (5.18)
j=1

We now estimate each of the terms appearing above. First, thanks to the definition (3.6a), the
estimate (5.1a) and the combinatorial estimate (5.34), we observe that
1 1
|T1 | .a2m,n kν 2 ∂y ω̊m,n;k eW χm+n kL2 kν 2 Sm,n;k eW χm+n kL2
q
(γ) 1
(1,0,0)
. Dm,n;k am,n kν 2 Sm,n ωk eW χm+n kL2
q n−1  −2σ−2σ∗ ! 12
X
(γ) s 2n−2ℓ (m + n)! (γ)
. Dm,n;k (Cλ ) Dm,ℓ;k .
(m + ℓ)!
ℓ=0

An application of the Young’s inequality yields that the first term can be estimated as in (5.16).
The second and last terms are estimated in a similar fashion,
n−1  
2 1
W 2
X
s 2n−2ℓ (m + n)! −2σ−2σ∗ (γ)
|T2 | .am,n kν 2 Sm,n;k e χm+n kL2 . (Cλ ) Dm,ℓ;k ;
(m + ℓ)!
ℓ=0
1 1
|T3 | .a2m,n kν ω̊m,n;k eW χm+n kL2 kν Sm,n
2
(1,0,0)
2 ωk eW χm+n kL2
n−1
X  −2σ−2σ∗
2 1
(1,0,0) W 2 s 2n−2ℓ (m + n)! (γ)
.am,n kν Sm,n ωk e χm+n kL2 .
2 (Cλ ) Dm,ℓ;k .
(m + ℓ)!
ℓ=0

Combining the estimates above and the decomposition (5.18) yields (5.16) for n ≥ 2. As a result,
the proof of the lemma is completed.

Lemma 5.3. The following bound is valid:

D E
(m,n)
a2m,n ν Re ∂y ω̊m,n;k , ∂y Cq,k e2W χ2m+n
n−1 X
X  −σ∗
1 (α) s 2n−2ℓ (m + n)!
Dm,ℓ;k + Cλ2s CKm,0;k 1n=1 . (5.19)
(ι) (α,W )
≤ Dm,n;k + CB (Cλ )
B (m + ℓ)!
ℓ=0 ι∈{α,µ}

Here B > 1 is an arbitrary constant and C is universal constant.

40
Proof. We distinguish between the n = 1 case and the n ≥ 2 case.
(1)
To estimate the n = 1 case, we expand the ∂y Cq,k with (5.17) as follows

(1)
∂y Cq,k = −2νq ′ ∂yy Γk ωk − 3νq ′′ ∂y Γk ωk − νq ′′′ Γk ωk . (5.20)

Based on this, we decompose the left hand side as follows

D E
2 2 (m,1) 2
am,1;k ν Re ∂y ω̊m,1;k , ∂y Cq,k χm+1
≤a2m,1;k ν 2 Re ∂y ω̊m,1;k , 2νq ′ ∂yy Γk |k|m ωk e2W χ2m+1
+ a2m,1;k ν 2 Re ∂y ω̊m,1;k , 3νq ′′ ∂y Γk |k|m ωk e2W χ2m+1
3
X
+ a2m,1;k ν 2 Re ∂y ω̊m,1;k , νq ′′′ Γk |k|m ωk e2W χ2m+1 =: T1;j . (5.21)
j=1

To estimate the first term, we implement an integration by parts to estimate the T1;1 -term in (5.21)
as follows

T1;1 ≤Ca2m,1 ν 2

× ℜ −q ′ ∂yy ω̊m,1;k − q ′′ ∂y |k|m ωk , ∂y Γk |k|m ωk e2W χ2m+1
  y=1
+ Re ∂y ω̊m,1;k q ′ ∂y Γk |k|m ωk e2W χ2m+1 + Re q ′ ∂y |k|m ωk , ∂y Γk |k|m ωk (∂y W )e2W χ2m+1
y=−1

′ m m
+ Re q ∂y |k| ωk , ∂y Γk |k| ωk (χ′m+1 )e2W χm+1
4
X
=: T1;1j . (5.22)
j=1

The estimate of the first term is a direct consequence of the definitions (4.2), (2.12) and the bounds
that tϕ ≤ C (2.11), kvy−1 k∞ + kvyy k∞ ≤ C (2.22e),

T1;11 ≤Ca2m,1 ν 2
× Re q ′ ∂yy ω̊m,1;k + q ′′ ∂y ω̊m,1;k , (−vy−2 vyy ∂y |k|m ωk + vy−1 ∂yy |k|m ωk + ikt∂y |k|m ωk )e2W χ2m+1
1 (α) (α)
≤ Dm,1;k + CBλ2s Dm,0;k .
B
Here B > 0 is an arbitrary constant. To estimate the second term in (5.22), we apply the boundary
conditions ωk (t, y = ±1) = ∂yy ωk (t, y = ±1) = 0 and the estimate (4.36), kvy−1 k∞ + kvyy k∞ ≤ C,
to obtain that


y=1
T1;12 .a2m,1 ν 2 |k|2m Re|q ′ |2 vy−1 ∂y ωk + iktωk vy−1 ∂yy ωk − vy−2 vyy ∂y ωk + ikt∂y ωk e2W χ2m+1
y=−1

 −2  y=1
.a2m,1 ν 2 |k|2m Re vy−1 ∂y ωk vy vyy ∂y ωk + ikt∂y ωk e2W χ2m+1
y=−1
y=1
(α) (α,W )
.a2m,1 ν 2 vy−3 vyy |∂y ω̊m,0;k |2 e2W χ2m+1 . λ2s Dm,0;k + λ2s CKm,0;k .
y=−1

41
Upon invoking the estimates (4.30), (2.11), and the definitions (2.12), (4.2), we have that the third
term in (5.22) can be estimated as follows,
(α) (α,W )
T1;13 .λ2s Dm,0;k + λ2s CKm,0;k .
For the fourth term in (5.20), we invoke the relations (2.1), (2.5a) and (4.29) to obtain
T1;14 .a2m,1 ν k∂y ω̊m,0;k eW χm k2 (m + 1)1+σ k∂y (∂v + ikt)ω̊m,0;k eW χm k2
λs am,0 ν 1/2
.λs am,0 ν 1/2 k∂y ω̊m,0;k eW χm k2 (k∂y ∂v ω̊m,0;k eW χm k2 + k∂y |k|ω̊m,0;k eW χm k2 )
(m + 1)σ∗
λ2s (α)
. D .
(m + 1)σ∗ m,0;k
This concludes the proof of the estimate (5.19) in the n = 1 case.
In the n ≥ 2 case, we decompose the integrand on the left-hand side of (5.19) according to the
identity (4.11), which gives
D E
(m,n)
a2m,n ν Re ∂y ω̊m,n;k , ∂y Cq,k e2W χ2m+n
 
n
≤a2m,n ν 2 Re ∂y ω̊m,n;k , Υ(1) ∂ 2
ω m,n;k e2W 2
χ
q n y m+n
 
2 2 n2 (2) 2W 2
+ am,n ν Re ∂y ω̊m,n;k , 2 Υn ∂y ωm,n;k e χm+n
q
  3
X
2 2 n3 (3) 2W 2
+ am,n ν Re ∂y ω̊m,n;k , 3 Υn ωm,n;k e χm+n =: T2;j .
q
j=1

Here, the {Υ(i) }3i=1 are defined in (4.11) and (4.12a)–(4.12b). We bound each of these contributions
(1,1,0)
individually, starting with T2;1 . By invoking the definition of D (α) (4.2) and (3.6), the Sm,n ωk -
estimate (5.1c) and the combinatorial estimate (5.34), we obtain that
n
T2;1 .a2m,n kΥ(1)
n kL∞ ν ∂y ω̊m,n;k e χm+n
W
kν∂y2 ω̊m,n;k eW χm+n kL2
q L2
 q
(1,1,0) (α)
. am,n kνSm,n ωk eW χm+n kL2 Dm,n;k
 −2σ−2σ∗  !1
n−1
X  2q
s 2n−2ℓ (m + n)! (α) (µ) (α)
. (Cλ ) Dm,ℓ;k + Dm,ℓ;k Dm,n;k .
(m + ℓ)!
ℓ=0

Next, we apply similar argument to obtain that

2
n
T2;2 .a2m,n kΥ(2)
n L k ∞ ν ∂ ω̊
y m,n;k eW
χ m+n . a2m,n kΥ(2) (1,1,0) W 2
n kL∞ kνSm,n ωk e χm+n kL2
q L2
n−1
X  −2σ−2σ∗  
(m + n)! (α) (µ)
. (Cλs )2n−2ℓ Dm,ℓ;k + Dm,ℓ;k .
(m + ℓ)!
ℓ=0
(a,b,c)
Finally, by invoking the definition of D (α) (4.2) and Sm,n ωk (3.6a), the estimates (5.1a), (5.1c)
and the combinatorial estimate (5.34), we obtain that
n n2
T2;3 .a2m,n kΥ(3) W
n kL∞ ν ∂y ω̊m,n;k e χm+n ν ωm,n;k eW χm+n
q L2 q2 L2

42
 .a2m,n kΥ(3) (1,1,0) W (2,0,0) W
n kL∞ kνSm,n ωk e χm+n kL2 kνSm,n ωk e χm+n kL2
n−1
X  −2σ−2σ∗  
s 2n−2ℓ (m + n)! (α) (µ)
. (Cλ ) Dm,ℓ;k + Dm,ℓ;k .
(m + ℓ)!
ℓ=0

Combining all the estimates developed thus far, we obtain the result (5.19).

We now need the µ version, which follows largely in an analogous manner to the γ term. We
have

Lemma 5.4. The following bound is valid:

D E n−1  
(m,n) 2W 2 1 (µ) X (m + ℓ)! 2σ∗ (µ)
a2m,n ν Re |k|ω̊m,n;k , |k|Cq,k e χm+n ≤ Dm,n;k + CB s 2n−2ℓ
(Cλ ) Dm,ℓ;k .
B (m + n)!
ℓ=0

Here B > 1 is an arbitrary constant and C, C are universal constants.

(1,0,0)
Proof. This follows in an identical manner to Lemma 5.2, with the bounds on Sm,n ωk being
replaced by the bound (5.3).
(m,n)
5.3 Commutator Estimates, Cvisc,k
Recall that the commutator coming from the viscous term (4.5a) reads as follows
n  
X n  
X
(m,n) n ℓ 2 n ℓ
2
Cvisc,k := νq n
∂v vy2 ∂v Γkn−ℓ |k|m ωk = νq n
∂v vy2 − 1 ∂v Γkn−ℓ |k|m ωk .
ℓ ℓ
ℓ=1 ℓ=1

Then we have the following lemma for the γ level estimate.

(m,n)
Lemma 5.5 (Viscous commutator for γ estimate). Let Cvisc,k be defined as above. It holds
X X D (m,n)
E
|am,n |2 ω̊m,n;k , Cvisc,k e2W χ2m+n
m,n≥0 k6=0
p p p 
. ǫν 100 E (γ) D (γ) + CK(γ) + ǫν 100 D (γ) .

Proof. For each fixed m, n ∈ N, we integrate by parts to obtain

D E
−1 2 (m,n) 2W 2
ν |am,n | ω̊m,n;k , Cvisc,k e χm+n
X n  D E
n l 2
= |am,n |2 ω̊m,n;k , |k|m q n ∂v (vy2 − 1)∂v Γkn−l ωk e2W χ2m+n
l
l=1
X n  D E
n l
=− |am,n |2 ∂v ω̊m,n;k , |k|m q n ∂v (vy2 − 1)∂v Γkn−l ωk e2W χ2m+n
l
l=1
X n   
n 2 m n l 2 ∂y n−l 2∂y W 2W 2
− |am,n | ω̊m,n;k , |k| q ∂v (vy − 1) Γk ωk e χm+n
l vy vy
l=1
X n  D   E
n l
− |am,n |2 ω̊m,n;k , |k|m ∂v q l ∂v (vy2 − 1) q n−l ∂v Γkn−l ωk e2W χ2m+n
l
l=1

43
 Xn  D E
n l
− |am,n |2 ω̊m,n;k , |k|m q l ∂v (vy2 − 1)∂v q n−l ∂v Γkn−l ωk e2W χ2m+n
l
l=1
Xn   
n 2 m n l 2 n−l 2W 2∂y χm+n
− |am,n | ω̊m,n;k , |k| q ∂v (vy − 1)∂v Γk ωk e χm+n
l vy
l=1
Xn     
n 2 m n 1 l 2 n−l 2W 2
− |am,n | ω̊m,n;k , |k| q ∂y ∂v (vy − 1)∂v Γk ωk e χm+n
l vy
l=1
= V1 + V2 + V3 + V4 + V5 + V6 . (5.23)

The term V1 : We divide the sum into two cases:

Xn  D E
n l
V1 = − |am,n |2 ∂v ω̊m,n;k , |k|m q n ∂v (vy2 − 1)∂v Γkn−l ωk e2W χ2m+n
l
l=1
 
X X  D E
= − +  n l
|am,n |2 ∂v ω̊m,n;k , |k|m q n ∂v (vy2 − 1)∂v Γkn−l ωk e2W χ2m+n
l
n/2≤l≤n 0<l<n/2

For the first piece, we use Hölder’s inequality to arrive at

X n D l
E
V11 : = |am,n |2 ∂v ω̊m,n;k , |k|m q n ∂v (vy2 − 1)∂v Γkn−l ωk e2W χ2m+n
l
n/2≤l≤n
X  
n l
≤ |am,n |2 ∂v ω̊m,n;k eW χm+n L2 q l−1 ∂v (vy2 − 1)χl−1 χ e1 2
l L
n/2≤l≤n

× |k|m q n−l+1 ∂v Γkn−l ωk eW χm+n .

L∞

By the Sobolev inequality and |vy | & 1, it follows

|k|m q n−l+1 ∂v Γkn−l ωk eW χm+n

L∞

. |k|m q n−l+1 ∂v Γkn−l ωk eW χm+n−l+1 2

L
 
m n−l+1 n−l W
+ |k| ∂y q ∂v Γk ωk e χm+n−l+1
L2

. |k|m q n−l+1 ∂v Γkn−l ωk eW χm+n−l+1 2

L
 
m n−l+1 n−l W
+ |k| ∂y q ∂v Γk ωk e χm+n−l+1
L2

+ |k|m q n−l+1 ∂v Γkn−l ωk eW ∂y χm+n−l+1

L2

+ |k|m q n−l+1 ∂v Γkn−l ωk ∂y W eW χm+n−l+1

L2
= I1 + I2 + I3 + I4 .

The I1 term is estimated by

I1 . |k|m q n−l+1 Γn−l+1

k ωk eW χm+n−l+1
L2

+ |k|m |kt|q n−l+1 Γkn−l ωk eW χm+n−l+1

L2

44
 . |k|m q n−l+1 Γn−l+1
k ωk eW χm+n−l+1
L2

+ t |k|m+1 q n−l Γkn−l ωk eW χm+n−l+1

L2

. |k|m q n−l+1 Γn−l+1

k ωk eW χm+n−l+1
L2

+ ν −1/3−η |k|m+1 q n−l Γkn−l ωk eW χm+n−l+1

L2

where we used the definition of Γ, the boundedness of q, and t ≤ ν −1/3−η . The treatment of I2 is
slightly different:
 
I2 = |k|m ∂y q n−l+1 (Γk − ikt)Γkn−l ωk eW χm+n−l+1 2
L
 
m n−l+1 n−l+1 W
≤ |k| ∂y q Γk ωk e χm+n−l+1 2
L
 
m n−l+1 n−l W
+ |k| ∂y q ktΓk ωk e χm+n−l+1 2
L
 
m n−l+1 n−l+1 W
≤ |k| ∂y q Γk ωk e χm+n−l+1 2
L
m n−l
+ |k| ∂y qq ktΓkn−l ωk eW χm+n−l+1 2
L
 
m+1 n−l n−l W
+ t |k| q∂y q Γk ωk e χm+n−l+1
L2
 
≤ |k|m ∂y q n−l+1 Γn−l+1
k ω W
k e χm+n−l+1 2 L
−1/3−η
+ν |k|m+1 q n−l Γkn−l ωk χm+n−l+1 2
L
 
−1/3−η m+1 n−l n−l W
+ν |k| q∂y q Γk ωk e χm+n−l+1
L2

where we used the boundedness of ∂y q as well as (5.24) in the last step. For the term I3 , from the
definition of χm+n with Γ, vy & 1, and (5.24), we have

I3 = |k|m q n−l+1 ∂v Γkn−l ωk eW ∂y χm+n−l+1

L2
n−l
. (m + n − l)1+σ |k|m q n−l+1 ∂y Γk ωk eW χm+n−l χ
e1
L2

. (m + n − l)1+σ |k|m ∂y (q n−l Γkn−l ωk )eW χm+n−l χ

e1
L2

+ (m + n − l)1+σ (n − l) |k|m+1 q n−l Γkn−l ωk eW χm+n−l χ

e1
L2

where χ
e1 is a fattened version of χ1 while supp χe1 ⊂ [−1, −1/2] ∪ [1/2, 1]. Next we turn to the last
term I4 . From the definition of the weight W , it is straightforward to check that
1
∂y W . ,
ν
from where using the bound for I1 we arrive at
1
|I4 | . |k|m q n−l+1 ∂v Γkn−l ωk eW χm+n−l+1
ν L2

.ν −1 |k|m q n−l+1 Γn−l+1

k ωk eW χm+n−l+1
L2

45
 + ν −4/3−δ |k|m+1 q n−l Γkn−l ωk eW χm+n−l+1 .
L2

Therefore, we obtain
X X X X X  
n
ν V11 . ν am,n a−1 −1 W
0,l am,n−l+1 |am,n | ∂v ω̊m,n;k e χm+n L2
l
m,n≥0 k6=0 m,n≥0 k6=0 n/2≤l≤n
l
× a0,l q l−1 ∂v (vy2 − 1)χl 2 am,n−l+1
L

−1
× ν W
ω̊m,n−l+1;k e χm+n−l+1 L2 + ν −4/3−δ ω̊m+1,n−l;k eW χm+n−l+1 L2

+ ∂y ω̊m,n−l+1;k eW χm+n−l+1 L2
+ ν −1/3−η q∂y ω̊m+1,n−l;k eW χm+n−l+1 L2
1+σ W
+(m + n − l) ∂y ω̊m,n−l;k e χm+n−l χ
e1 L2
1+σ W


+(m + n − l) (n − l) ω̊m+1,n−l;k e χm+n−l χ
e1 L2
.

Next we use Corollary 5.14, (5.33), Young’s inequality, and Hölder’s inequality to obtain
 1/2
X X p p p  X 2
n
ν V11 . D (γ) E (γ) + D (γ)  a20,n q n−1 ∂v (vy2 − 1)χm+n 
L2
m,n≥0 k6=0 n≥1

In view of that
ν −100 ≤ CeW/2 for t ≤ ν −1/3−η , (5.24)
by an argument similar to (5.1a), we further get

X X p p p  X X  −2σ
100  n−ℓ n!
ν V11 . ν D (γ) E (γ) + D (γ) C a20,ℓ
ℓ!
m,n≥0 k6=0 n≥1 ℓ≤n−1
  1/2
ℓ−1 2
ℓ−1
× ∂y q ∂v (vy2 − 1) eW/2 χℓ
L2
 1/2
p p p  X n
. ν 100 D (γ) E (γ) + D (γ)  a2n k∂y (q n ∂v (vy2 − 1))eW/2 χn k2L2  (5.25)
,
n≥0

where we have used Fubini in the last step. Using the Product Lemma 5.17 leads to
X X p p p  q 
99 (γ) (γ) (γ) (α) (α)
ν V11 . ν D E + D EH + EH
m,n≥0 k6=0
p p p 
. ǫν 99 D (γ) E (γ) + D (γ) .

Now we consider the case 0 < l < n/2:

X n D l
E
V1,2 : = |am,n |2 ∂v ω̊m,n;k , |k|m q n ∂v (vy2 − 1)∂v Γkn−l ωk e2W χ2m+n
l
0<l<n/2
X n l
. |am,n |2 ∂v ω̊m,n;k χm+n L2 q l ∂v (vy2 − 1)χm+n ∞
l L
0<l<n/2

46
 × |k|m q n−l ∂v Γkn−l ωk eW χm+n−l χ
e1 .
L2

For the last term in the above expression, noting that

[q ℓ , ∂y ] = −ℓq ℓ−1 ∂y q

and
q ℓ ∂y f = ∂y (q ℓ f ) + [q ℓ , ∂y ]f
for any ℓ ∈ N, we get

|k|m q n−l ∂v Γkn−l ωk eW χm+n 2

 L
n−l n−l
. |k| ∂y q Γk ωk eW χm+n
m
L2
m n−l−1 n−l
+ (n − l) |k| q ∂y qΓk ωk eW χm+n
L2
 
(1,0,0)
. |k|m ∂y q n−l Γkn−l ωk eW χm+n + |k| m
Sm,n−l ωk eW χm+n
L2 L2

where we used the boundedness of ∂y q and (3.6a) in the last step. Similar as before, the main
difficulty is to estimate
 
l l l
q l ∂v (vy2 − 1)χm+n ∞ . q l ∂v (vy2 − 1)χl+1 2 + ∂y q l ∂v (vy2 − 1)χl+1 2
L L L
l l l
. q ∂v (vy2 − 1)χl+1 + ∂y (q l
∂v (vy2 − 1))χl+1
L2 L2
l
+ (l + 1)1+σ q l ∂v (vy2
− 1)χl 2
L
1+σ

l l 2 l
. l + 1 q ∂v (vy − 1)χl 2 + ∂y (q l ∂v (vy2 − 1))χl . (5.26)
L L2

Collecting the estimates about V12 , by Corollary 5.14, (5.33), Young’s inequality, and Hölder’s
inequality, similarly as (5.25), we have
X X p p p  p p p 
ν V12 . ν 100 D (γ) E (γ) + D (γ) vy2 − 1 Y . ǫν 100 D (γ) E (γ) + D (γ) .
1,0
n,m≥0 k6=0

The term V2 : Like in the previous situation, we consider two cases:

Xn   
n 2 m n l 2 n−l 2∂y W 2W 2
V2 = − |am,n | ω̊m,n;k , |k| q ∂v (vy − 1)∂v Γk ωk e χm+n
l vy
l=1
 
X X  
= − +  n (· · ·) =: V21 + V22 . (5.27)
l
n/2≤l≤n 0<l<n/2

For the first piece, we use Cauchy-Schwarz inequality to obtain

X n  l 2∂y W 2W 2

V21 : = |am,n |2 ω̊m,n;k , |k|m q n ∂v (vy2 − 1)∂v Γkn−l ωk e χm+n
l vy
n/2≤l≤n
X n l
. |am,n |2 ω̊m,n;k eW χm+n L2 q l−1 ∂v (vy2 − 1)eW/2 χl−1 χ e1 2
l L
n/2≤l≤n

47
 2∂y W W/2
× |k|m q n−l+1 ∂v Γkn−l ωk e χm+n−l+1 .
vy L∞

For the second term in the above bound, we invoke (5.1a). For the third term, by Sobolev embed-
ding, we arrive at

2∂y W W/2
|k|m q n−l+1 ∂v Γkn−l ωk e χm+n−l+1
vy L∞

. |k|m q n−l+1 ∂v Γkn−l ωk ∂y W eW/2 χm+n−l+1

L∞

. |k|m q n−l+1 ∂v Γkn−l ωk ∂y W eW/2 χm+n−l+1 2

L
 
m n−l+1 n−l W/2
+ |k| ∂y q ∂v Γk ωk ∂y W e χm+n−l+1
L2

. |k|m q n−l+1 ∂v Γkn−l ωk ∂y W eW/2 χm+n−l+1 2

L
 
m n−l+1 n−l W/2
+ |k| ∂y q ∂v Γk ωk ∂y W e χm+n−l+1
L2
 
+ |k|m q n−l+1 ∂v Γkn−l ωk ∂y ∂y W eW/2 χm+n−l+1
L2
m n−l+1
+ |k| q ∂v Γkn−l ωk ∂y W eW/2 ∂y (χm+n−l+1 )
L2
= II1 + II2 + II3 + II4 .

For the II1 term, we use

ν −100 ∂y W . eW/2 , ν −100 ∂y2 W . eW/2 , ∀t ≤ ν −1/3−η ,

and (5.24) to obtain

II1 . |k|m q n−l+1 Γn−l+1

k ωk ∂y W eW/2 χm+n−l+1
L2
m+1 n−l+1 n−l W/2
+t |k| q Γk ωk ∂y W e χm+n−l+1
L2
100 W 99
.ν ω̊m,n−l+1;k e χm+n−l+1 L2
+ν ω̊m+1,n−l;k eW χm+n−l+1 L2

for t ≤ ν −1/2 . The second term II2 is treated in a similar fashion

 
II2 . |k|m ∂y q n−l+1 Γn−l+1
k ωk ∂y W eW/2 χm+n−l+1 2
L
 
m+1 n−l+1 n−l W/2
+ t |k| ∂y q Γk ωk ∂y W e χm+n−l+1
L2
100 W 99
.ν ∂y ω̊m,n−l+1,k e χm+n−l+1 L2
+ν ∂y ω̊m+1,n−l;k eW χm+n−l+1 L2
99 W
+ν ω̊m+1,n−l;k e χm+n−l+1 L2
.

By a similar argument as that of II2 , we arrive at

II3 . ν 100 ω̊m,n−l+1;k eW χm+n−l+1 L2

+ ν 99 ω̊m+1,n−l;k eW χm+n−l+1 L2
.

The last piece II4 is estimated in an analogous way

II4 . (m + n − l)1+σ |k|m q n−l+1 ∂v Γkn−l ωk ∂y W eW/2 χm+n−l

L2

48
 . ν 100 (m + n − l)1+σ ∂y ω̊m,n−l;k eW χm+n−l L2
100 1+σ W
+ν (m + n − l) (n − l) ω̊m,n−l;k e χm+n−l L2
.

As in the treatment of V1 , gathering the estimates, by Corollary 5.14, (5.33), Young’s inequality,
and Hölder’s inequality, we again arrive at
X X p p p 
ν V21 . ǫν 100 E (γ) D (γ) + CK(γ) .
m,n≥0 k6=0

The second term in (5.27) may be estimated by

X n   2∂y W 2W 2

2 m n l 2 n−l
V22 : = |am,n | ω̊m,n;k , |k| q ∂v (vy − 1)∂v Γk ωk e χm+n
l vy
0<l<n/2
X n  l 2∂y W
. |am,n |2 ω̊m,n;k eW χm+n L2 q l ∂v (vy2 − 1) χm+n
l vy L∞
0<l<n/2

× |k|m q n−l ∂v Γkn−l ωk eW χm+n−l χ

e1 .
L2

Following the arguments in the treatment of V12 step by step, we may arrive at
X X p p p 
ν V22 . ǫν 100 E (γ) D (γ) + CK(γ) .
m,n≥0 k6=0

The term V3 : We recall

n  D
X   E
n l
V3 = − |am,n |2 ω̊m,n;k , |k|m ∂v q l ∂v (vy2 − 1) q n−l ∂v Γkn−l ωk e2W χ2m+n
l
l=1

and get by Hölder’s inequality

Xn    
n l
|V3 | . |am,n |2 ω̊m,n;k eW/2 χm+n ∂v q l ∂v (vy2 − 1) eW/2 χl
l L∞ L2
l=1

× |k|m q n−l ∂v Γkn−l ωk eW χm+n−l χ

e1 .
L2

We use Sobolev’s inequality to obtain

 
ω̊m,n;k eW/2 χm+n . ω̊m,n;k eW/2 χm+n + ∂y ω̊m,n;k eW/2 χm+n
L∞ L2 L2

. ω̊m,n;k eW/2 χm+n + ∂y ω̊m,n;k eW/2 χm+n

L2 L2

+ ω̊m,n;k ∂y eW/2 χm+n + (m + n)1+σ ω̊m,n;k eW/2 χm+n−1 χ

e1 .
L2 L2

For the last term in the above expression, we have

e1 kL2 . (m + n)1+σ |k|m q n ∂v Γn−1 ωk eW/2 χm+n−1 χ

(m + n)1+σ kω̊m,n;k χm+n−1 χ e1
L2

+ t(m + n)1+σ |k|m+1 q n−1 Γn−1 ωk eW/2 χm+n−1 χ

e1
L2

. (m + n)1+σ ∂y ω̊m,n−1;k eW/2 χm+n−1 χ

e1
L2

49
 + (m + n)1+σ n ω̊m,n−1;k eW/2 χm+n−1 χ
e1
L2

+ t(m + n)1+σ ω̊m+1,n−1;k eW/2 χm+n−1 χ

e1
L2
100 1+σ W
.ν (m + n) ∂y ω̊m,n−1;k e χm+n−1 χ
e1 L2
ν 99
+ (m + n)1+σ n ω̊m,n−1;k eW χm+n−1 χ
e1 L2
t
99 1+σ W
+ ν (m + n) ω̊m+1,n−1;k e χm+n−1 χ
e1 L2
.

Then again by Corollary 5.14, (5.33), Young’s inequality, and Hölder’s inequality, similar as (5.25),
we arrive at
X X p p p 
ν V3 . ǫν 100 D (γ) E (γ) + D (γ) .
m,n≥0 k6=0

The term V4 : Like in the treatment of V3 term, we have

Xn  
n l
|V4 | . |am,n |2 ω̊m,n;k eW/2 χm+n ∞ q l−1 ∂v (vy2 − 1)eW/2 χl
l L L2
l=1

× (n − l) |k|m q n−l ∂v Γkn−l ωk eW χm+n−l χ

e1 ,
L2

from where we follow the treatment of V3 term line by line.

The term V5 : We recall that
Xn   
n 2 m n l 2 n−l 2W 2∂y χm+n
V5 = − |am,n | ω̊m,n;k , |k| q ∂v (vy − 1)∂v Γk ωk e χm+n
l vy
l=1

from where applying Hölder’s inequality gives

n  
X n l
|V5 | ≤ |am,n |2 (m + n)1+σ kω̊m,n;k χm+n kL∞ q l ∂v (vy2 − 1)eW χl
l L2
l=1

× |k|m q n−l ∂v Γkn−l ωk eW χm+n−l χ

e1
L2

The term V6 : We recall that

n  
X   
n 2 m n 1 l 2 n−l 2W 2
V6 = − |am,n | ω̊m,n;k , |k| q ∂y ∂v (vy − 1)∂v Γk ωk e χm+n
l vy
l=1

from where applying Hölder’s inequality gives

n  
X  
n 2 1 l
|V6 | ≤ |am,n | kω̊m,n;k χm+n kL∞ ∂y q l ∂v (vy2 − 1)eW χl
l vy L∞ L2
l=1

× |k|m q n−l ∂v Γkn−l ωk eW χm+n−l χ

e1 .
L2

Hence the proof is finished by summing up the estimates of V1 –V6 in (5.23) and we omit further
details.

For the α and µ level estimate, we get the following two lemmas.

50
Lemma 5.6 (Viscous commutator for α estimate). It holds
X X D 
(m,n)
 E
ν |am,n |2 ∂y ω̊m,n;k , ∂y Cvisc,k e2W χ2m+n
m,n≥0 k6=0
p p p p p p  q q 
99 (α) (γ) (α)
. ǫν E (α) (α) (α) (γ)
E + D + D + D + CK (µ) DH + DH .

Proof. For fixed m, n ≥ 0, note

D   E
(m,n)
ν |am,n |2 ∂y ω̊m,n;k , ∂y Cvisc,k e2W χ2m+n
Xn  D   E
n l 2
=ν 2
|am,n |2 ∂y ω̊m,n;k , |k|m ∂y q n ∂v (vy2 − 1)∂v Γkn−l ωk e2W χ2m+n
l
l=1
Xn   D E
n l 2
=ν 2
|am,n |2 ∂y ω̊m,n;k , |k|m n∂y qq n−1 ∂v (vy2 − 1)∂v Γkn−l ωk e2W χ2m+n
l
l=1
D E
l 2
+ |am,n |2 ∂y ω̊m,n;k , |k|m q n ∂y ∂v (vy2 − 1)∂v Γkn−l ωk e2W χ2m+n
D   E
2 m n l 2 2 n−l 2W 2
+ |am,n | ∂y ω̊m,n;k , |k| q ∂v (vy − 1)∂y ∂v Γk ωk e χm+n

= V1α + V2α + V3α .

The term V1α : By the boundedness of ∂y q, it is easy to obtain
X n  l
α 2
|V1 | . ν |am,n |2 ∂y ω̊m,n;k eW χm+n L2 n q l−1 ∂v (vy2 − 1)χn−1
l L∞
l≤n/2
2
× |k|m q n−l ∂v Γkn−l ωk eW χm+n 2
L
X n l
+ ν2 |am,n |2 ∂y ω̊m,n;k eW χm+n L2
nq l−2 ∂v (vy2 − 1)eW/2 χl−1 χ
e1
l L2
n≥l>n/2
2
× |k|m q n−l+1 ∂v Γkn−l ωk eW/2 χm+n−l+1
L∞
α α
= V1,1 + V1,2 .
By Sobolev embedding, we have
1/2   1/2
l l−1 l 2 l−1 l 2
q l−1 ∂v (vy2 − 1)χn−1 . q ∂v (v y − 1)χ l ∂y q ∂v (v y − 1)χ l
L∞ L2 L2

l 1/2 l
. q l−1 ∂v (vy2 − 1)χl 2 (l − 1) q l−2 ∂v (vy2 − 1)χl 2
L L
 l  1/2
l
+ q l−1 ∂y ∂v (vy2 − 1) χl +l 1+σ
q l−1
∂v (vy2 − 1)χl−1 .
L2 L2

For a term of the following type

l
q l−1 ∂v (vy2 − 1)
we invoke estimate (5.1a). While for terms of type
2  
l−2 l−1 l−2 ∂y l−2 1 l−2
q ∂y ∂v (vy2 − 1) = q ∂v (vy2 − 1) + q l−2
∂y ∂y ∂v (vy2 − 1),
vy vy

51
and
2
|k|m q n−l ∂v Γkn−l ωk eW χm+n
L2

we perform an argument similar to (5.1a) and (5.1c). Collecting the estimates about V1,1 α , by

Corollary 5.14, (5.33), Young’s inequality, Hölder’s inequality, and Fubini’s theorem, we have
 1/4
X X p p p p p  X (0,1,0)
α
V1,1 . ν 100 E (α) E (α) + D (α) + D (γ) + D (µ) hti  a2n kS0,n (vy2 − 1)eW/2 χn k2L2 
m,n≥0 k6=0 n≥0
 1/4
X (0,2,0)
× a2n kS0,n (vy2 − 1)eW/2 χn k2L2 
n≥0
p p p p p  q q 
99 (α) (α) (α) (γ) (µ) (γ) (α)
. ǫν E E + D + D + D DH + DH .

α , we note by
where we also used assumptions (2.22) and Product Lemma 5.17. For the term V1,2
Sobolev inequality
2
|k|m q n−l+1 ∂v Γkn−l ωk eW/2 χm+n−l+1
L∞
2 1/2  2
 1/2
. |k|m q n−l+1 ∂v Γn−l+1
k ωk eW/2 χm+n−l+1 2 |k|m ∂y q n−l+1 ∂v Γkn−l ωk eW/2 χm+n−l+1
L L2
1/2  
2 2
. |k|m q n−l+1 ∂v Γkn−l ωk eW/2 χm+n−l+1 |k|m ∂y q n−l+1 ∂v Γkn−l ωk eW/2 χm+n−l+1
L2 L2

2
+ |k|m q n−l+1 ∂v Γkn−l ωk ∂y eW/2 χm+n−l+1
L2
!1/2
2 1/2
+ (m + n − l + 1) |k|m q n−l+1 ∂v Γkn−l ωk χm+n−l 2 .
L

By the definition of Γ, we obtain

2
|k|m q n−l+1 ∂v Γkn−l ωk eW/2 χm+n−l+1
L2
n−l+1
≤ |k| q m n−l+1
∂v Γk ωk eW/2
χm+n−l+1 + t |k|m kq n−l+1 ∂v Γkn−l ωk eW/2 χm+n−l+1
L2 L2

. |k| q m n−l+1
∂v Γn−l+1
k ωk eW/2 χm+n−l+1 + t |k| kq m n−l+1
∂v Γkn−l ωk eW/2 χm+n−l+1
L2 L 2
 
. |k| q m n−l+1
∂v Γn−l+1
k ωk eW/2 χm+n−l+1 m
+ t |k| k∂y q n−l+1 n−l
Γk ω k eW/2
χm+n−l+1
L2 L2
m n−l n−l W/2
+ (n − l + 1)t |k| k∂y qq Γk ωk e χm+n−l+1
L2

. |k| q m n−l+1
∂v Γn−l+1
k ωk eW/2 χm+n−l+1 +ν −1/3−η
∂y ω̊m+1,n−l;k eW/2 χm+n−l+1
L2 L2
−1/3−η W/2
+ (n − l + 1)ν ω̊m+1,n−l;k e χm+n−l+1
L2

where we used (5.24). Similarly, we have

 2

|k|m ∂y q n−l+1 ∂v Γkn−l ωk eW/2 χm+n−l+1
L2

52
  
≤ |k|m ∂y q n−l+1 ∂v Γn−l+1
k ω k eW/2 χm+n−l+1 2
L
 
m n−l+1 n−l W/2
+ t |k| k∂y q ∂v Γk ωk e χm+n−l+1
L2
 
∂y q
. ∂y2 ω̊m,n−l+1;k eW/2 χm+n−l+1 + (n − l + 1) ∂y ω̊m,n−l+1;k eW/2 χm+n−l+1
L2 q L2

+ |k| q m n−l+1
∂y Γkn−l+1 ωk eW/2 χm+n−l+1 + t ∂y ω̊m+1,n−l+1;k e W/2
χm+n−l+1
L2 L2
2 W/2
+t ∂y (qω̊m+2,n−l;k ) e χm+n−l+1
L2

. ∂y2 ω̊m,n−l+1;k eW/2 χm+n−l+1 2

L
 
+ (n − l + 1) |k| ∂y q ∂v Γkn−l ωk eW/2 χm+n−l+1
n−l m
L2

+ (n − l + 1)t ∂y (ω̊m,n−l;k ) e W/2

χm+n−l+1 + |k| q m n−l+1
∂y Γkn−l+1 ωk eW/2 χm+n−l+1
L2 L2
W/2 2 W/2
+ t ∂y ω̊m+1,n−l+1;k e χm+n−l+1 +t ∂y (qω̊m+2,n−l;k ) e χm+n−l+1
L2 L2

α , we deduce
where we used (5.24) in the last step. Therefore, as for V1,2

X X p p p p p  q q 
α 99 (γ) (α)
V1,2 . ǫν E (α) E (α) (α) (γ)
+ D + D + D (µ) DH + DH .
m,n≥0 k6=0

The term V2α : We recall

n  D
X E
n l 2
V2α =ν 2
|am,n |2 ∂y ω̊m,n;k , |k|m q n ∂y ∂v (vy2 − 1)∂v Γkn−l ωk e2W χ2m+n .
l
l=1

Upon dividing the sum into two parts: l ≤ n/2 and l > n/2, we follow the treatment of V1α to get

|V2α | . V2,1
α α
+ V2,2
α are given as
where V2,i

X n  l
α 2
V2,1 =ν |am,n |2 ∂y ω̊m,n;k eW χm+n L2
|k|m q l ∂y ∂v (vy2 − 1)χl+1
l L∞
l≤n/2
2
× |k|m q n−l ∂v Γkn−l ωk eW χm+n−l
L2

and
X n  l
α 2
V2,2 =ν |am,n |2 ∂y ω̊m,n;k eW χm+n L2
|k|m q l−1 ∂y ∂v (vy2 − 1)eW/2 χl−1 χ
e1
l L2
l>n/2
2
× q n−l+1 ∂v Γkn−l ωk eW/2 χm+n−l+1 .
L∞
α term, we note by Sobolev inequality and Minkowski inequality,
To estimate the V2,1

l
|k|m q l ∂y ∂v (vy2 − 1)χl+1
L∞

53
 1/2   1/2
l l
. |k|m q l ∂y ∂v (vy2 − 1)χl+1 |k|m ∂y q l ∂y ∂v (vy2 − 1)χl+1
L2 L2
1/2  
l l
. |k|m q l ∂y ∂v (vy2 − 1)χl+1 |k|m ∂y q l ∂y ∂v (vy2 − 1) χl+1
L2 L2
!1/2
l
+ |k|m q l ∂y ∂v (vy2 − 1)∂y χl+1
L2

2
The treatment of q n−l+1 ∂v Γkn−l ωk χm+n α is the same as in V α and we omit
in the term V2,2 1,2
L∞
further details. Using Corollary 5.14, (5.33), Young’s inequality, and Hölder’s inequality, similarly
as (5.25), we conclude the estimate of V2α .
The term V3α : Using Γk = Γ0 + ikt= ∂v + ikt, V3α could be further rewritten as
Xn  D   E
n l
α
V3 = ν 2
|am,n |2 ∂y ω̊m,n;k , |k|m q n ∂v (vy2 − 1)∂y ∂v Γn−l+1
k ω k e2W 2
χ m+n
l
l=1
X n  D   E
n l
+ ν2 |am,n |2 ∂y ω̊m,n;k , |k|m q n ∂v (vy2 − 1)∂y ∂v Γkn−l iktωk e2W χ2m+n ,
l
l=1

then this term could be treated almost the same as the V2 term in Lemma 5.5, except the γ
norms are replaced by the correspond α norms as well as some commutator estimates elaborated
in Section 6, and we conclude the proof without giving more details.
(m,n)
Lemma 5.7 (Viscous commutator for µ estimate). Let Cvisc,k be defined as above. It holds
X X D (m,n)
E
ν |am,n |2 ikω̊m,n;k , ikCvisc,k e2W χ2m+n
m,n≥0 k6=0
p p p 
. ǫν 100 E (µ) D (µ) + CK(µ) + ǫν 100 D (µ) .

Proof. Noting
n  
X n  
X
(m,n) n ℓ 2 n ℓ
2
ikCvisc,k n
:= νq |k| m
∂v vy2 ∂v Γkn−ℓ ikωk =ν q ℓ ∂v vy2 − 1 q n−ℓ ∂v ikΓkn−ℓ |k|m ωk ,
ℓ ℓ
ℓ=1 ℓ=1

the proof follows almost identically as that of Lemma 5.5.

(m,n)
5.4 Commutator Estimates, Ctrans,k
We recall the transport commutator is defined as
n  
X
(m,n) n l
Ctrans,k := q n
∂v G |k|m Γn−l+1
k ωk .
l
l=1

Then we have the following estimates.

(m,n)
Lemma 5.8 (Transport commutator for γ estimate). Let Ctrans,k be defined as above. It holds
X X D E
(m,n)
θn2 |am,n |2 ω̊m,n;k , Ctrans,k e2W χ2m+n
m,n≥0 k6=0
p p p p  q q 
99 (γ) (γ) (γ)
. ǫν E (γ) E (γ) (γ)
+ D + CK 1 + DH + DH .

54
Proof. The transport commutator term could be written as
n  
X  
(m,n) n l−1 H
Ctrans,k = q n
∂v Γn−l+1
k |k|m ωk . (5.28)
l vy
l=1

Hence, we have
D E
(m,n)
|am,n |2 ω̊m,n;k , Ctrans,k e2W χ2m+n
n−1   
X 
n 2 n lH 2W 2
= |am,n | ω̊m,n;k , q ∂v ω̊m,n−l;k e χm+n .
l vy
l=0

Similar as the treatment of the viscous commutator term in section 5.3, we split the sum into two
parts:
 
X X   
  n 2 m l l H n−l n−l 2W 2 m,n m,n
+ |am,n | ω̊m,n;k , |k| q ∂v q Γk ωk e χm+n = Tγ,1 + Tγ,2 .
l vy
l≤n/2 n/2<l≤n−1

The first term is estimated by

X  
m,n 2 n lH
Tγ,1 ≤ |am,n | ω̊m,n;k eW χm+n L2
q l ∂v χ̃l+1 |k|m q n−l Γkn−l ωk eW χm+n−l
l vy L∞ L2
l≤n/2 y

where we denote
(
χl+1 , l≥1
χ
el+1 =
χe1 , l=0

with χ
e1 being a fattened version of χ1 while supp χ
e1 ⊂ [−1, −7/16] ∪ [7/16, 1]. The trouble term
l lH
el+1 ∞ is treated as in (5.26)
q ∂v vy χ
Ly

lH
lH lH
q l ∂v χ
el+1 . l1+σ + 1 q l ∂v χ
ẽ + ∂y (q l ∂v )χ
ẽl
vy L∞ vy l L2 vy L2
y

with
(
e χl , l ≥ 1
χ
el =
e
χ
e1 , l = 0

e
where χ
el is a fattened version of χ e
el while supp χ
el ⊂ [−1, −1/2] ∪ [1/2, 1]. Therefore, we get
X X X X X  
m,n n
θn2 Tγ,1 . |am,n |2 ω̊m,n;k eW χm+n L2 a−1 −1
m,n−l al+1 (l + 1)
2−2/s
l
m,n≥0 k∈Z m,n≥0 k∈Z l≤n/2

lH e
× am,n−l |k|m q n−l Γkn−l ωk eW χm+n 2 al+1 (l + 1)2/s−2 l1+σ + 1 q l ∂v χ
e
L vy l L2
X X X  
2 n
+ |am,n | ω̊m,n;k eW χm+n L2 a−1 −1
m,n−l al+1 (l + 1)
2−2/s
l
m,n≥0 k∈Z l≤n/2

55
  
lH e
× am,n−l |k|m q n−l Γkn−l ωk eW χm+n al+1 (l + 1) 2/s−2
∂y q ∂v l
el (5.29)
χ .
L2 vy L2

By Corollary 5.14, using Hölder’s and Young’s inequalities, we deduce

 1/2
X X X   2
2 m,n 2 (γ) 100  2s−2 2 H
θn Tγ,1 . θn E ν  (n + 1) an+1 ∂y eW/2 χn 
vy n L2
m,n≥0 k∈Z n≥0
 1/2 
X   2
H  
+ (n + 1)2s−2 a2n+1 eW/2 χn .
vy n L2
n≥0

Noting that
X ∞
1 1
= = (−1)l H l ,
vy 1+H
l=0

we have by Lemma 5.16 and Lemma 5.18 that

 1/2 !
X   2 q ∞ 
X 
 H  (γ) (α) i/2
(n + 1)2s−2 a2n+1 ∂y W/2
e χn ≤ν −1/2
DH EH
vy n L2
n≥0 i=0
q
(γ)
.ν −1/2 DH ,

and
 1/2
X   2 q q q q q 
 (n + 1)2s−2 2 H  (γ) (γ) (γ) (γ) (γ)
an+1 eW/2 χn ≤ EH +ν −1/2
EH DH + EH DH .
vy n L2
n≥0

Hence, we get
X X  q q 
m,n (γ) (γ)
θn2 Tγ,1 . ǫν 99 E (γ) 1 + DH + DH
m,n≥0 k∈Z

m,n
by the bootstrap assumption and the boundedness of the sequence {θn }. While for Tγ,2 , we have

X  
m,n n2 l H W/2
Tγ,2 . |am,n | ω̊m,n;k eW χm+n L2
q l ∂v e χn−1
l vy L2
n/2<l≤n−1

× |k|m q n−l Γkn−l ωk eW/2 χm+n .

L∞

By Sobolev embedding, we get

|k|m q n−l Γkn−l ωk eW/2 χm+n

L∞
1/2   1/2
. |k|m q n−l Γkn−l ωk eW/2 χm+n−l+1 |k|m ∂y q n−l Γkn−l ωk eW/2 χm+n−l+1
L2 L2

. ∂y ωm,n−l;k eW/2 χn−l + (m + n − l)1+σ ωm,n−l;k eW/2 χn−l

L2 L2

56
 + ν −1 ωm,n−l;k eW/2 χn−l
L2

A similar argument as (5.29), we arrive at

 1/2
X X p p p  X   2
m,n H
θn2 Tγ,2 . θn2 ν 100 E (γ) D (γ) + CK(γ)  (n + 1)2s−2 a2n+1 eW/2 χn 
vy n L 2
m,n≥0 k∈Z n≥0
p p p  q q q q q 
(γ) (γ) (γ) (γ) (γ)
. θn2 ν 99 E (γ) D (γ) + CK(γ) EH + EH DH + EH DH
p p p  q q 
99 (γ) (γ) (γ) (γ) (γ)
. ǫν E D + CK 1 + DH + DH

concluding the proof, where we again used the boundedness of the sequence {θn }.

Lemma 5.9 (Transport commutator for α estimate). Let the assumptions in Lemma 5.8 hold.
Then it follows
X X D E
2 2 (m,n) 2W 2
νθn |am,n | ∂y ω̊m,n;k , ∂y Ctrans,k e χm+n
m,n≥0 k6=0
p X p p p  X  q q 
(ι) (ι)
. ǫν 98 E (α) E (ι) + D (ι) + CK(ι) 1+ DH + DH . (5.30)
ι∈{α,γ} ι∈{α,γ}

Proof. From (5.28), we get

n     n  
(m,n)
X n l−1 H X n l−1 l−1 H  n−l+1 n−l+1 
∂y Ctrans,k = |k|m ∂y q l−1 ∂v q n−l+1 Γn−l+1
k ω k + |k|m
q ∂v ∂y q Γk ωk .
l vy l vy
l=1 l=1

Following the estimates in Lemma 5.8 line by line for the above two terms, upgrading the corre-
sponding regularity from γ to α, we get (5.30).
(m,n)
Lemma 5.10 (Transport commutator for µ estimate). Let Ctrans,k be defined as above. It holds
X X D E
(m,n)
θn2 |am,n |2 ω̊m,n;k , Ctrans,k e2W χ2m+n
m,n≥0 k6=0
p p p p  q q 
99 (µ) (µ) (µ)
. ǫν E (µ) E (µ) + D (µ) + CK 1 + DH + DH .

Proof. By the same argument as in Lemma 5.7, we obtain the desired bound in the statement.

5.5 Proof of Proposition 2.6

We now consolidate all the bounds from this section in order to prove Proposition 2.6.

Proof of Proposition 2.6. We subdivide the proof into three steps.

Proof of (2.33a), (2.34a), and (2.35a). We first invoke the bound (5.16) to obtain the following
(γ)
estimate for |Iq |
∞ X
X n−1
∞ XX  −2σ−2σ∗
1 (γ) s 2n−2ℓ (m + n)! (γ)
|Iq(γ) | ≤ D + CB (Cλ ) Dm,ℓ;k .
B (m + ℓ)!
m=0 n=1 k∈Z ℓ=0

57
Now we invoke the technical inequality (5.39)G =D(γ) and select the δDrop , n∗ in Lemma 5.22
··· ···
appropriately (to guarantee that the parameter B is large enough) to obtain the result.
This concludes the proof of (2.33a), whereas the proofs of (2.34a), (2.35a) are nearly identical.

Proof of (2.33b), (2.34b), and (2.35b). It is a direct consequence of Lemma 5.8, Lemma 5.9, and
Lemma 5.10. In fact, noting the sequence {θn } is bounded, we have
X X D E
(γ) (n)
|Itrans | ≤ |am,n |2 θn2 ωm,n;k , |k|m Ctrans,k e2W χ2m+n
n,m≥0 k6=0
X X D (n)
E
. |am,n |2 ωm,n;k , |k|m Ctrans,k e2W χ2m+n
n,m≥0 k6=0

3/2  
(γ) 3/2
p   E (γ) + EH
(γ) (γ) (γ) (γ)
. E D + CK + DH +
hti2
  ǫ3
(γ)
. ǫ D (γ) + CK(γ) + DH + 2 .
hti

And the proof of (2.34b) and (2.35b) follows similarly.

(γ)
Proof of (2.33c), (2.34c), and (2.35c). According to the definition of Ivisc , these inequalities are
exactly the statement of Lemma 5.5, Lemma 5.6, and Lemma 5.7, concluding the proof.

5.6 Technical Lemmas: Gevrey Norm Estimates on Physical Side

5.6.1 Binomial coefficients lemmas
We need the following lemmas about the binomial coefficients bound from our companion paper [5].

Lemma 5.11. For l ≤ n/2 and n ≥ 5, we have

     l(s−1)
−2 n −1 n − l −1 −1 1
hti am,n+1 am+1,l a0,j a0,n−l−j ≤ .
l j 2

Moreover, if m ≤ c0 n for some c0 > 0, then it holds

     l(s−1)  
−2 n −1 n − l −1 −1 1 c0 + 1/2 ms
hti am,n+1 a a0,j a0,n−l−j ≤ . (5.31)
l m+1,l j 2 c0 + 1

If m ≥ C0 n, then we have
     s−1  (n−l)s
−2 n −1 n − l −1 −1 1 1
hti am,n+1 a a0,j a0,n−l−j ≤ . (5.32)
l m+1,l j 2l C0 + 1

Remark 5.12. Note that in the above theorem, for aα,β , it is essentially the sum α + β that
matters. Hence, we are able to generalize it to a more general version:
     l(s−1)
−2 n −1 n − l −1 −1 1
hti am,n+1 aα1 ,β1 aα2 ,β2 aα3 ,β3 ≤ . (5.33)
l j 2

58
with

α1 + β1 = m + l + 1, α2 + β2 = j, α3 + β3 = n − l − j.

Of course, the corresponding inequalities for (5.31) and (5.32) also holds.
Remark 5.13. From Theorem 5.11, one may also obtain
     l(s−1)
−2 n −1 n − l −1 −1 a 1
hti am,n+1 a a0,j a0,n−l−j m ≤ C
l m+1,l j 2
for a parameter a ≤ n/4, where the constant C depends on a.
Two direct corollaries are the following.
Corollary 5.14. For l ≤ n/2 and n ≥ 5, we have
   l(s−1)
−1 n −1 −1 1
hti am,n am,l a0,n−l ≤ .
l 2
Moreover, if m ≤ c0 n for some c0 > 0, then it holds
   l(s−1)  
−1 n −1 −1 1 c0 + 1/2 ms
hti am,n a a ≤ .
l m,l 0,n−l 2 c0 + 1
If m ≥ C0 n, then we have
   s−1  (n−l)s
−1 n −1 −1 1 1
hti am,n a a ≤ .
l m,l 0,n−l 2l C0 + 1
Corollary 5.15. For l ≤ n/2 and n ≥ 5, we have
   l(s−1)
−1 n −1 −1 1
hti am,n+1 am+1,l a0,n−l ≤ .
l 2
Moreover, if m ≤ c0 n for some c0 > 0, then it holds
   l(s−1)  
n −1 1 c0 + 1/2 ms
hti−1 am,n+1 am+1,l a−1
0,n−l ≤ .
l 2 c0 + 1
If m ≥ C0 n, then we have
   s−1  (n−l)s
−1 n −1 −1 1 1
hti am,n+1 a a ≤ .
l m+1,l 0,n−l 2l C0 + 1

5.6.2 Product rule

We need the following product lemmas from our companion paper [5].
Lemma 5.16. For f |y=±1 = g|y=±1 = 0, we have

kf gkY0,−s . kf kY0,−s kgkY 1,0 + kf kY1,−s kgkY 0,0 .

Without the homogeneous boundary condition, it holds

   
kf gkY0,−s . kf kY0,−s kgkY 1,0 + kgkL2 + kf kY1,−s + kf kL2 kgkY 0,0 .

59
Lemma 5.17. Assume f |y=±1 = g|y=±1 = 0, we have

kf gkY1,0 . kf kY1,0 kgkY 1,0 .

Without the boundary condition requirement, it holds

  
kf gkY1,0 . kf kY1,0 + kf kL2 kgkY 1,0 + kgkL2 .

Lemma 5.18. Suppose f |y=±1 = g|y=±1 = 0, we have

kf gkY1,−s . kf kY1,−s kgkY 1,0 .

Without boundary value requirement, it holds

  
kf gkY1,−s . kf kY1,−s + kf kL2 kgkY 1,0 + kgkL2 .

5.6.3 Convolution type lemmas

We also need the following versions of convolution type lemma, proof of which may be found in our
companion paper [5].

Lemma 5.19. It holds

!1/2  1/2  1/2
X X X X X XX X
fn−m,k gm,k hn ≤ |fn,k |2  |gn,k |2   |hn |2  .
n≥0 0≤m≤n k∈Z n≥0 k∈Z n≥0 k∈Z n≥0

Lemma 5.20. It holds

  1/2  1/2
X X XX XX XX XX
fn−m,k−l gm,l hn,k ≤  |fn,k |  |gn,k |2   |hn,k |2  .
n≥0 0≤m≤n k∈Z l∈Z n≥0 k∈Z n≥0 k∈Z n≥0 k∈Z

5.7 Combinatorial lemma

To derive the estimates (5.1b) and (5.1d), we need the following combinatorial lemmas.

Lemma 5.21. The following inequality holds

n−1
X  −σ−σ∗
n−ℓ n!
C ≤ C, ∀n ∈ N. (5.34)
ℓ!
ℓ=0

Here the bound C only depends on C ≥ 1, σ, σ∗ .

n 1
o
Proof. We define a threshold L = max (2C) σ+σ∗ , 10 . We observe that if n < 2L2 + 2, then the
sum in (5.34) is bounded in terms of C, σ, σ∗ . Hence we assume that n ≥ 2L2 + 2 without loss of
generality. Now we further decompose the sum into two parts
n−1  −σ−σ∗ L n−1
!
X X X
n−ℓ n!
C = + = I1 + I2 . (5.35)
ℓ!
ℓ=0 ℓ=0 ℓ=L+1

60
Now we focus on the I2 term. In this regime, we can compute the ratio between consecutive terms
ℓ, ℓ + 1:


n! −σ−σ∗
Cn−ℓ ℓ! C
 −σ−σ∗ = .
n! (ℓ + 1)σ+σ∗
Cn−ℓ−1 (ℓ+1)!

1
Hence, if ℓ ≥ L, then the ratio is less than 2 and we can estimate the I2 as follows

n−1
X  n−1−ℓ
1 C
I2 ≤ σ+σ
≤ 1. (5.36)
2 n ∗
ℓ=L+1

Now we estimate the term I1 in (5.35),

 ⌊ n2 ⌋
2n −σ−σ∗ C2
I1 ≤ (1 + L) max C 2 (n(n − 1) · · · (ℓ + 1)) ≤ (1 + L) C ≤ (1 + L)C.
(5.37)
ℓ∈{0,...,L} ⌊n/2⌋σ+σ∗

Combining the discussion above and the decomposition (5.35) and the estimates (5.36), (5.37), we
have obtained (5.34).

Lemma 5.22. Given a constant B > 1, there exists δDrop (σ, B), n⋆ (σ, B) (defined in (2.13) such
that the following estimate holds
∞ X
X ∞ n−1
X  −2σ ∞ ∞
(m + n)! 1 XX 2
θn2 (Cλs )2n−2ℓ Gm,ℓ;k ≤ θn Gm,n;k . (5.38)
(m + ℓ)! B
m=0 n=1 ℓ=0 m=0 n=0

Here Gm,ℓ;k are positive quantities.

Proof. We denote the left hand side of (5.38) as follows

∞ X
X ∞ n−1
X  −2σ
(m + n)!
(5.38)L.H.S = 1 2 s 2n−2ℓ
n<n∗ θn (Cλ ) Gm,ℓ;k
m=0 n=1 ℓ=0
(m + ℓ)!
X∞ X ∞ n−1
X  −2σ
(m + n)!
+ 1n≥n∗ (Cλ ) s 2n−2ℓ
Gm,ℓ;k := T1 + T2 .
m=0 n=1 ℓ=0
(m + ℓ)!

For T2 , we froze (m, k), and study the sum

∞ n−1
X X  −2σ
s 2n−2ℓ (m + n)!
Sm,k := (Cλ ) Gm,ℓ;k .
n=n
(m + ℓ)!
∗ ℓ=0

By interchanging the order of summation, we obtain

nX
∗ −1 ∞  
X (m + n)! −2σ
s 2n−2ℓ
Sm,k = (Cλ ) Gm,ℓ;k
(m + ℓ)!
ℓ=0 n=n∗
X∞ X∞  −2σ
s 2n−2ℓ (m + n)! (rectangle) (triangle)
+ (Cλ ) Gm,ℓ;k = Sm,k + Sm,k .
(m + ℓ)!
ℓ=n∗ n=ℓ+1

61
To ease the notation going forward, we define
Az
A := (Cλs )2 , b := 2σ, K : N → R, K(z) = .
[(m + z)!]b

(triangle)
Estimation of Sm,k : This is the more challenging of these bounds due to the double infinite
sum in ℓ, n. To handle such a sum, we issue the following
Claim 5.23. Fix any C > 1. There exists L∗ sufficiently large relative to A, b, C and universal
constants such that the function K satisfies the following inequality
∞
X
K(n) ≤ CK(L∗ ). (5.39)
n=L∗

Proof of Claim 5.23. This follows almost immediately by comparing K(n) to a geometric function.
More precisely, define
∗
L∗ AL 1
R > 1, C0 = C0 (L∗ ) = R , τ (n) = C0 .
[(m + L∗ )!]b Rn

Such a choice of C0 ensures that

1
τ (L∗ ) = C0 = K(L∗ ).
R L⋆
We first note that, due to the standard geometric sum properties,
∞
X ∞
X 1 1 1 1
τ (n) = C0 n
= C 0 L∗ 1 = 1 τ (L∗ ).
R R 1− R 1− R
n=L∗ n=L ∗

We secondly notice that for L∗ sufficiently large, we have the order property

An RL⋆ AL⋆
K(n) = b
≤ τ (n) = , n ≥ L∗ . (5.40)
[(m + n)!] [(m + L⋆ )!]b

To establish, (5.40), we notice that K(L∗ ) = τ (L∗ ). Therefore, it suffices to check

K(n + 1) τ (n + 1)
≤ , n ≥ L∗ ,
K(n) τ (n)

which is equivalent to
A 1
b
≤ , n ≥ L∗ ,
(m + n + 1) R

which can clearly be achieved by choosing L∗ sufficiently large. Therefore, we have

∞
X ∞
X 1 1
K(n) ≤ τ (n) ≤ 1 τ (L∗ ) = 1 K(L∗ ).
n=L∗ n=L∗
1− R 1− R

62
 (triangle)
Returning now to the Sm,k -bound, we have by choosing n∗ sufficiently large
∞ ∞
! ∞
(triangle)
X 1 X X K(ℓ + 1)
Sm,k = Gm,ℓ;k K(n) ≤ Gm,ℓ;k
K(ℓ) K(ℓ)
ℓ=n∗ n=ℓ+1 ℓ=n∗
X∞ ∞
X
Aℓ+1 [(m + ℓ)!]b CA
=C b
Gm,ℓ;k ≤ Gm,ℓ;k . (5.41)
[(m + ℓ + 1)!] Aℓ (m + n∗ + 1)b
ℓ=n∗ ℓ=n∗

By taking n∗ large and summing over m, k, we obtain

X ∞ ∞
(triangle) 1 XX 2
Sm,k ≤ θn Gm,n;k .
3B
m=0 n=0

(rectangle) (triangle)
Estimation of Sm,k : This bound works in a nearly identical manner to Sm,k .

Bound of T1 : We have upon again invoking (5.16)

X n∗
∞ X n−1
X
T1 ≤Cn∗ θn2 Gm,ℓ;k
m=0 n=0 ℓ=0
n−1 
θn 2 2
X∞ X n∗ X X
≤Cn∗ θℓ Gm,ℓ;k
m=0 n=0 k∈Z ℓ=0
θℓ
X ∞ X n∗ n−1
X
2
≤Cn∗ δDrop θℓ2 Gm,ℓ;k
m=0 n=0 ℓ=0
X∞ X ∞
2
=δDrop Cn∗ θℓ2 Gm,ℓ;k , (5.42)
m=0 ℓ=0

which upon bringing δDrop small relative to n∗ , we obtain the right-hand side is bounded by one
third of the right hand side of (5.38).
By summing all the conclusions above, we conclude the proof of (5.38).

6 Elliptic Regularity Estimates

6.1 Setup for Elliptic Bounds
In this section, we study the elliptic equation:

∆k ψk = wk , ψk |y=±1 = 0.

When we implements analysis in different regions, we need different set of elliptic estimates. We
categorize them as interior elliptic estimates and exterior elliptic estimates.
Elliptic Estimates in the Interior Analysis: First of all, we recall the elliptic equation for
Ψ(E)

2
∆Ψ(E) =wE + (1 + H I )2 − (1 + H)2 ∂v Ψ(I)


+ (1 + H)∂v H − (1 + H I )∂v H I ∂v Ψ(I) , (6.1)
Ψ(E) (x, ±1) =0.

The first main goal in this section is to prove the following proposition.

63
Proposition 6.1 (Estimate of (6.1)). Consider the solutions to the equation (6.1). Under the
assumptions (2.22), the following estimate holds:
∞
!2/r
X e m+n
(2λ) (E)
kχm+n |k|m q n Γnk Ψk k2Hy1
(m + n)!
m+n=0
  
(α) (γ) 2 −ν −1/8
. E (γ) [wk ] + EInt
low
[wk ] 1 + EH + EH e . (6.2)

This estimate implies (3.16) in Theorem 3.1.

Elliptic Estimates in the Exterior Analysis:
For the elliptic estimates applied in the exterior region, we apply another decomposition of the
solution (3.2a), (3.2b). We observe that
   
(I) (E) 2 (I)
∆k φk (v(y)) + φk (y) = (|k|2 − ∂v ) φk (v(y)) + χ e1 (v(y))wk (y)
 
(I) (E)
= (eχc1 (v(y)) + χ
e1 (v(y))) wk (y) = wk (y), φk (v(y)) + φk (y) = 0.
y=±1

Here in the equation, we omit the argument t in the v(t, y), χ ec1 (t, v(t, y)) expressions
e1 (t, v(t, y)), χ
for the sake of notation simplicity. Furthermore, we define the following quantities
(I) (E)
ek (t, v(t, y)) :=(1 − χ
w e1 (t, v(t, y)))wk (t, y), ek (t, y) = χ
w e1 (t, v(t, y))wk (t, y), (6.3)
(I) (E)
Rk :=R.H.S. of (3.2a), Rk := R.H.S. of (3.2b).

Furthermore, we define an intermediate cut-off function χ∗ (t, ·) ∈ C ∞ ([v(t, −1), v(t, 1)]) such that
for a positive constant c > 0 and any t, the following conditions hold

 1, v ∈ support χ1 (y(t, ·));
χ∗ (t, v) = 0, ec1 (t, ·);
v ∈ support χ (6.4)

smooth, others.
ec1 (t, ·)) ≥ c > 0.
distance(support χ∗ (t, ·), support χ

Here y(t, v) is the inverse function of v(t, y) at fixed time t. Here the cutoff function χ∗ has a mild
dependence on time because the boundary of the domain is v(t, ±1). Finally, we have the relation
χn (y)χ∗ (t, v(t, y)) = χn (y) for n ≥ 1.
Now we present a more precise version of Theorem 3.1.
Proposition 6.2. Assume that the conditions in (2.22) hold. Further assume that k 6= 0.
a) Interior estimates: Recall the definition of b
am,n (3.3) and consider the solution to (3.2a). The
following estimate holds
∞
X (I) 1 (I) 1 low
b2m,n kχ∗ |k|m (∂v + ikt)n |k|l1 ∂vl2 φk k2L∞
a . e k2 2 .
kw E [wk ], l1 , l2 ∈ {0, 1, 2, 3, 4}.
v hti2 k Lv hti2 Int
m+n=0
(6.5)
low is defined in (3.4). The smooth cutoff function χ
Here the implicit constant is universal and EInt ∗
is defined in (6.4). Moreover, for any m ∈ N, the following Sobolev norm estimate holds
(I) (I)
kχ∗ |k|m ∂vl φk kL∞
v
ek kL2v ,
≤ C(m)kw l ∈ {0, 1, 2, 3, 4}. (6.6)

64
 Next, the following inviscid damping estimate holds for l ∈ {0, 1, 2},
∞
X n
(I) ϕ2 X (I)
a2m,n kχ∗ |k|m (∂v
b + ikt)n ∂vl φk k2L∞ . k(∂v + ikt)ℓ Rk k2L2v
v (|k|t)2n
m+n=0 ℓ=0
2 X n
ϕ (I) ϕ2
. k(∂v + ikt)ℓ w
ek k2L2v . low
EInt [wk ]. (6.7)
(|k|t)2n (|k|t)2n
ℓ=0

(I)
Here Rk is defined in (6.3). Furthermore, for finite Sobolev norm, the following inviscid damping
estimate holds
1000
X 1002
b m n (I) 1 X b (I)
Bm,n k|k| (∂v + ikt) φk kL2v . 2
Bm,n k|k|m (∂v + ikt)n wk kL2v . (6.8)
m+n=0
hti m+n=0

b) Exterior estimates: Under the assumption (2.22), we have the following estimate
∞
X 2
X ∞
X
(E) (E) −1/9 (γ) (α) (I)
a2m,n kJm,n
(a,b,c)
φk k2L2 . a2m,n kJm,n
(0)
ek k2L2 + e−ν
w ek k2L2
(EH + EH )kw
m+n=0 a+b+c=0 m+n=0
−1/9
 
(γ) (γ) (α) low
.e−ν Ek + (EH + EH )EInt;k .

(E)
Here, the left hand side contains the usual H 2 -based Gevrey bound of φk . Moreover, if a+b+c = 3,
∞
X X (E)
a2m,n kJm,n
(a,b,c)
φk k22
m+n=0 a+b+c=3
∞
X 1
X (E)
. a2m,n kJm,n
(a,b,c)
ek k22
w
m+n=0 a+b+c=0
!
X∞ X  
−ν −1/8 (0) (E) 2 (I) (ι) (ι)
+e a2m,n kJm,n w
ek k2 + ek k2Hv1
kw EH + DH
m+n=0 ι∈{α,γ}
 
−1/9
X  
.e−ν D (γ) + (E (γ) + EInt;k
low
)
(ι) (ι)
EH + DH  .
k k
ι∈{α,γ}

(E)
Here the left hand side contains the usual H 3 -based Gevrey bound of φk .

Remark 6.3. The above estimates in b) hold with “∞” replaced by “M ∈ N”.

Our strategy for proving the estimates (6.6), (6.7) will be to use elliptic regularity estimates
(E)
and a delicate inductive scheme in Gevrey spaces to get the φk bounds, and carry out detailed
(I)
analysis of the Green’s function to obtain the φk bounds.
Since the exterior estimate is the most delicate one, we detail our strategy:
Step # 1: First of all, by the assumption (2.22) and the technical Lemma 6.21, we have that
∞ X
X 1
2 (a,b,0) (γ) (α)
B0,n ϕ2n kJ0,n χ1 (vy2 − 1))k2L2 ≤C exp{−ν −1/8 }(EH + EH ),
(e (6.9)
n=0 a+b=0

65
 ∞ X
X
2 (a,b,0) (γ) (α)
B0,n ϕ2n kJ0,n χ1 (vy2 − 1))k2L2 ≤C exp{−ν −1/8 }(DH + DH ),
(e
n=0 a+b=2

where B0,n is defined in (2.9).

Step # 2: Through detailed analysis of the elliptic equation, we derive the following pointwise-
in-time estimate given that the right hand side of (6.9) is small enough,
∞
X 2
X ∞
X
(E) (E) (I)
a2m,n kJm,n
(a,b,c)
φk k2L2 . a2m,n kJm,n
(0)
ek k2L2 + exp{−ν −1/9 }kw
w ek k2L2 .(6.10)
m+n=0 a+b+c=0 m+n=0

Then we prove the following estimate

∞
X X (E)
a2m,n kJm,n
(a,b,c)
φk k22
m+n=0 a+b+c=3
∞
X 1
X (E)
. a2m,n kJm,n
(a,b,c)
ek k22
w
m+n=0 a+b+c=0
∞ X
2
! ∞
!
X (a,b,0)
X
2 (0) (E) 2 (I)
+ B0,n ϕ2n kJ0,n (e χ1 (vy2 − 1))k22 a2m,n kJm,n ek k2
w + ek k2H(6.11)
kw 1 .
k
n=0 a+b=0 m+n=0

(0) (E) (E)

We recall from (3.8b) that Jm,n w ek = χm+n |k|m q n Γnk w ek . The proof of these two claims (6.10),
(6.11) is the main technical part of this section. The proof of the claims are complicated thanks
(a,b,c) m+n
to the non-traditional definition of the operator Jm,n fk , which only assigns weights when
q
a + b + c ≤ n. Hence we distinguish between the small-n case and the large-n case separately. In
the first case, we mainly focus on deriving the Sobolev norm controls and keep track of various
boundary contributions arising when one implements integration by parts. This is the content of
Subsection 6.3.2. In the second case, boundary contributions are zero and we mainly focus on the
control over the combinatorial coefficients in the definition of the Gevrey norms. This is the content
of Subsection 6.3.3.
Step # 3: Finally, we use the technical Lemma 6.18 to translate the right hand side of (6.10)
(α) (γ) low .
and (6.11) to quantities E (γ) , D (γ) , EH , EH , EInt
Notation: In this section, the Ti1 i2 i3 will be used to denote various terms in the estimates and will
be local variables for the proof of each lemma. If the proof has several steps, the i1 will represent
the step number. Moreover, if we decompose the term Ti1 i2 into several sub-terms, they will be
denoted as Ti1 i2 i3 .
(I,·)
6.2 Bounds on Eell
We are interested here in the elliptic problem (3.2a), which we copy as follows:
(I) (I) (I) (I) (I)
(∂v2 − |k|2 )φk (v) = w ec1 (v)(∂v2 − ([vy ]∂v )2 )φk (v) =: Rk (v),
ek (v) + χ φk |v(t,±1) = 0.
(6.12)

Here [vy ] is the function vy (t, y) expressed in the v-coordinate and we drop the t-dependence. We
have explicit solution formula:
ˆ v(t,1)
(I) (I)
φk (t, v) = Gk (v, v ′ )Rk (t, v ′ )dv ′ , (6.13)
v(t,−1)

66
 
′ 1 sinh(k(v(t, 1) − v ′ )) sinh(k(−v(t, −1) + v)), v ≤ v′;
Gk (v, v ) := −
k sinh(k|v(t, 1) − v(t, −1)|) sinh(k(v(t, 1) − v)) sinh(k(−v(t, −1) + v ′ )), v ≥ v′.
cosh(k(|v − v ′ | − |v(t, 1) − v(t, −1)|)) cosh(k(v + v ′ − v(t, 1) − v(t, −1)))
=− +
2k sinh(k|v(t, 1) − v(t, −1)|) 2k sinh(k|v(t, 1) − v(t, −1)|)
′ ′
=:Hk (t, v − v ) + Sk (t, v + v ).
(I)
With the expression (6.13), we can estimate the higher derivatives of the solution φk . Thanks
ec1 (2.22b), we have the separation of support
to the definitions of the cutoff χ1 (2.4) and χ
1
χc1 (v(t, ·))}, support{χ1 (·)}) ≥
distancey (support{e > 0.
320
(I)
Since the support of the forcing Rk (6.12) is contained in support{e χc1 (v(t, ·))}, we expect that
(I)
φk is very smooth in terms of Gevrey index and radius of analyticity. Thanks to the Lebesgue
Dominated Convergence theorem, it can be justified that the ∂v derivatives of φ(I) has the following
(I)
expression given that the distance between v and support Rk is strictly positive,
ˆ v(t,1)
(I) (I)
∂vn φk (t, v) = ∂vn Gk (v, v ′ )Rk (t, v ′ )dv ′ , χc1 (·)}, v) > 0.
distance (support{e
v(t,−1)

Moreover, we would like to derive the inviscid damping estimate for the interior. To this end,
e k -vector field estimates of φ(I) .
we will derive finite Γ k

Lemma 6.4. Assume that (2.22a) holds and define

e k = ∂v + ikt.
Γ

If the threshold ǫ is smaller than a constant depending on a fixed M ∈ N, the following estimates
hold
M X
X M
X  X
M
(I)
e n φ k2 2 + e n R(I) k2 2 ≤C M, k[vy ]2 − 1k M +2 enw (I) 2
k∂vb |k|c Γ k k Lv kΓ k k Lv Hv (suppχ
ec ) kΓ k ek k(6.14)
L2v ;
1
n=0 b+c=2 n=0 n=0
M
X +2  X
M
(I) (I)
k∂vn φk k2L2v ≤C M, k[vy ]2 − 1kHvM +2 (suppχec ) k∂vn w 2
ek k(6.15)
L2v .
1
n=0 n=0

(I)
Here Rk is defined in (6.3) and kfk kHv2 is the standard Sobolev norm in the v-coordinate, i.e.,
XM
kfk k2H M := k∂vn fk k2L2v .
v
n=0

Remark 6.5. We remark that in the sequel, we only use the finite regularity estimate (e.g.,
(I) (I)
M = 1000) of the functions Rk (6.3) and φk in this lemma. Hence the explicit expression of the
coefficients in (6.14) and (6.15) are not needed.

Proof. We divide the proof in two steps.

(I)
Step # 1: Proof of (6.14). First of all, we study the structure of the right hand side Rk in
(6.12). For the sake of notation simplicity, we denote

z(t, v) := Z(t, y(v)), Z(t, y) = vy (t, y)2 − 1.

67
By applying the Γ en , n ∈ N, on both side of the equation (3.2b), we obtain that
k
  
2 2 e n (I) e n (I) e n c 2 2 (I) 1 2 (I)
(∂v − |k| )Γk φk (t, v) =Γk wek (t, v) + Γk χ e1 −([vy ] − 1)∂v φk + ∂v ([vy ] − 1)∂v φk
2
  
1
n (I)
e1 z ∂v φk − ∂v z ∂v φk
2 (I)
e e n c (I)
=Γk wek (t, v) − Γk χ . (6.16)
2
Next, we apply the induction argument to prove the estimate (6.14). For (6.16)n=0 , by a simple
variant of the elliptic regularity lemma 6.12 and the Sobolev estimate (2.22a), we have obtained
that
X X
χc1 z∂v2 φk kL2v + ke
χc1 ∂v z∂v φk kL2v . kw
(I) (I) (I) (I) (I) (I)
k∂vb |k|c φk kL2v .kw
ek kL2v + ke ek kL2v + ǫ k∂vb |k|c φk kL2v .
b+c=2 b+c=2

Hence if the ǫ is small enough compared to uniform constants, we have that

X (I) (I)
k∂vb |k|c φk kL2v ≤ Ckw
ek kL2v .
b+c=2

This concludes the starting level of the induction.

Assume that the following estimate holds for 1 ≤ n − 1 ≤ N :
ℓ
X ′
eℓ φ(I) kL2
k|k|2 Γ e ℓ φ(I) kL2 + k∂ 2 Γ
+ k|k|∂v Γ eℓ (I) eℓ w (I)
k k v k k v v k φk kL2v . kΓ k ek kL2v , ∀ℓ ≤ n − 1. (6.17)
ℓ′ =0

We would like to show that the same estimate holds with “n − 1” replaced by “n”. We test the
en φ(I) and apply integration by parts on v ∈ [v(t, −1), v(t, 1)]:
equation (6.16) with the Γ k k

v=v(t,1) ˆ v(t,1)
k∂v Γ e n φ(I) k2 2 =Re∂v Γ
e n φ(I) k2 2 + |k|2 kΓ e n φ(I) Γ
e n φ(I) − Re en w
Γ
(I) e n (I)
k k Lv k k Lv k k k k k ek Γk φk dv
v=v(t,−1) v(t,−1)
v(t,1)  
1 c
ˆ
e1 z ∂v φk − χ
2 (I)
e ∂v z ∂v φk
e n c (I) e n (I)
+ Re Γk χ Γk φk dv
v(t,−1) 2 1
=:T1 + T2 + T3 . (6.18)

Now we treat the first term on the right hand side of (6.18). We observe that since χ ec1 (v) is
identically zero in a neighborhood of the boundary, we have that the (3.2a) yields that
(I) (I) (I)
∂v2m φk (t, v(t, ±1)) = 0, ∂v2m+1 φk (t, v(t, ±1)) = |k|2m ∂v φk (t, v(t, ±1)), m ∈ N. (6.19)

Now we apply these relations to obtain

n   n  
! !
v=v(t,1) X X
e n φ(I) Γ
e n φ(I) n n ℓ
(I) (I)
Re∂v Γ k k = Re ∂vℓ+1 (ikt)n−ℓ φk∂ (ikt)n−ℓ φk
k k
v=v(t,−1) ℓ ℓ v v=v(t,±1)
ℓ=0 ℓ=0
n  
! n  
!
X n X n
=Re 1ℓ even ∂v (ikt) φk
ℓ+1 n−ℓ (I)
1ℓ odd ∂vℓ (ikt)n−ℓ φk(I) .
ℓ ℓ v=v(t,±1)
ℓ=0 ℓ=0

For every pairing in this expression, we have that

1ℓ1 even 1ℓ2 odd Re(ikt)n−ℓ1 ∂vℓ1 +1 φk(I) (ikt)n−ℓ2 ∂vℓ2 φk(I)
v=v(t,±1)

68
  
=1ℓ1 even 1ℓ2 odd |k|ℓ1 +ℓ2 −1 Re (ikt)2n−ℓ1 −ℓ2 (−1)n−ℓ2 ∂v φk ∂v φk
(I) (I)
= 0.
v=v(t,±1)

Therefore, the boundary term add up to zero in (6.18). Now the second term in (6.18) can be
estimated as follows:
enw (I) e n (I)
|T2 | ≤CkΓ k ek kL2v kΓk φk kL2v .

Since χ ec1 vanishes near the boundary, we can apply integration by parts to rewrite the last term
and estimate it as follows
X n  ˆ  
n 1 n−ℓ c
|T3 | = χ1 z)∂v Γk φk − ∂v (e
n−ℓ c
∂v (e 2 e ℓ (I)
χ1 ∂v z)∂v Γk φk
e ℓ (I) e n (I)
Γk φk dv
ℓ 2
ℓ=0
ˆ   ˆ  
e1 z∂v Γk φk ∂v Γk φk dv +
n (I)
χc1 z)∂v Γ
c e e n (I) e n φ(I) Γe n φ(I) dv
≤ χ ∂v (e k k k k

n−1
X n
X
χc1 zkWvn,∞ kΓ χc1 ∂v zkWvn,∞ kΓ
(I)
e n φ kL2 (I)
e ℓ φ kL2 + C(n)ke e n φ kL2 (I) (I)
e ℓ φ kL2 .
+ C(n)ke k k v
k∂v2 Γ k k v k k v
k∂v Γ k k v
ℓ=0 ℓ=0

Now we invoke the induction hypothesis (6.17) to obtain that

  n−1
X
|T3 | ≤C(n)kzkHv2 (supp(eχc1 )) k∂v Γk φk kL2v + kkΓk φk kL2v + C(n)kzkHvn+2 (supp(eχc )) kΓk φk kL2v
n (I) 2 n (I) 2 n (I)
e e e eℓ w (I)
kΓ k ek kL2v .
1
ℓ=0

Combining the estimates above and (6.18), we apply the assumption (2.22a) to obtain that
n
X
en φ(I) k2 2 + kΓ
k∂v Γ e n φ(I) kL2 kk2 Γ
e n φ(I) kL2 . eℓ w
kΓ
(I) e n (I)
k k Lv k k v k k v k ek kL2v kΓk φk kL2v . (6.20)
ℓ=0

Hence, we have that

n
X
e n φ(I) kL2
k|k|2 Γ eℓ w
kΓ
(I)
k k v
. k ek kL2v .
ℓ=0

Now we multiply the equation (6.20) by |k|2 and invoke the above relation to get
n
X
(I)
en φ kL2 . eℓ w (I)
k|k|∂v Γ k k v
kΓ k ek kL2v .
ℓ=0

e n φ(I) , and apply the estimates derived before to obtain

Finally, we test the equation (6.16) by ∂v2 Γ k k
that
 X
e n φ(I) k2 2 ≤C n, kzk n+2
k∂v2 Γ kΓe ℓ w(I) kL2 k∂ 2 Γ
e n (I) 2 e n (I) 2 e n (I)
k k Lv Hv (supp(e
χc )) 1
k k v v k φk kL2v + k∂v Γk φk kL2v k|k| Γk φk kL2v
ℓ≤n

+ Ckz e n φ(I) k2 2 .
kHv2 (supp(eχc1 )) k∂v2 Γ k k Lv

Hence, by the estimate (2.22a) and the smallness of ǫ, we have that

 X
e n φ(I) kL2 ≤ C n, kzk n+2
k∂v2 Γ e ℓ w(I) kL2 .
kΓ
k k v H v (supp(e
χ c )) k k 1 v
ℓ≤n

69
Hence, we observe that

X  X
n
e n φ(I) kL2 ≤ C n, kzk n+2
k∂vb |k|c Γ eℓ w
kΓ
(I)
k k v Hv (supp(e
χc )) k ek kL2v . 1
b+c=2 ℓ=0

(I)
This concludes the proof of (6.17) for n. To estimate the Rk in (3.2a), we can invoke the elliptic
P +2 n
n=0 k∂v zkH n (suppχc1 ) . ǫ. We omit further
(I)
estimate of φk and combine it with the assumption M
details for the sake of brevity.
Step # 2: Proof of (6.15). The proof is identical to the one in step # 1. One needs to replace
e k by ∂v . Thanks to the condition (6.19), the boundary contributions
Γ
v=v(t,1)
(I) (I)
Re∂vn+1 φk ∂vn φk = 0.
v=v(t,−1)

We omit further details. This completes the proof.

Now we can prove Proposition 6.2 a).

Proof of Proposition 6.2 a). We divide the proof into several steps.
Pn #
Step
n

1: Proof of (6.5), (6.6) (6.7): We apply the property that tϕ . 1, and the fact
n
ℓ=0 ℓ = 2 to estimate the term as follows:

∞
X
e n |k|m+l1 ∂vl2 φ k2L∞ (I)
a2m,n kχ∗ Γ
b k k v
m+n=0
∞ 2
X ˆ v(1)
(I)
= bm,n
B 2
ϕ2(1+n) χ∗ (∂v + ikt)n |k|m+l1 ∂vl2 Gk (v, v ′ )Rk (v ′ )dv ′
m+n=0 v(−1)
L∞
X∞ X n  2 ˆ v(1) 2
b 2 n 2(1+n′ ) ′ ′ (I)
≤ Bm,n ′ ϕ χ∗ (y(v)) |k|m+n−n +l1 ∂vn +l2 Gk (v, v ′ )Rk (v ′ )dv ′
m+n=0 ′
n v(−1)
n =0 L∞
v
∞ 2
X ′
ˆ v(1)
′ ′ (I)
≤ max bm,n
B 2
22n ϕ2(1+n ) χ∗ |k|m+n−n +l1 ∂vn +l2 Gk (v, v ′ )Rk (v ′ )dv ′
n′ {0,··· ,n} v(−1)
m+n=0 L∞
v
∞
!
X b2(m+n)s 22n
λ 2
≤ϕ 2 (I)
kRk k2L2v max sup 1
|v−v′ |≥c>0 |k|
m+n−n′ +l1 n′ +l2
∂v Gk (v, ·)
n′ ∈{0,··· ,n} ((m + n)!)2s v∈[v(t,−1),v(t,1)] L2v′
m+n=0
(I)
=: ϕ2 kRk k2L2v T1 . (6.21)

To estimate the last line, we recall the explicit form of the Green’s function (6.13). Without loss
of generality, we consider the v ≤ v ′ case . Hence,
∞
X b 2(m+n)s
(2λ) |k|2(m+n+l1 +l2 )
T1 1v≤v′ ≤ max
n′ ∈{0,··· ,n} (m + n)!2s k 2 (sinh(k|v(1) − v(−1)|))2
m+n=0

2
1|v−v′ |≥c sinh k(v(t, 1) − v′ ) sinh(n ) (k(−v(t, −1) + v))
′
× sup
v L2v′
∞
X b 2(m+n)s
(2λ) |k|2(m+n+l1 +l2 )
ˆ
′
. sup e2k(v(1)−v ) e2k(v−v(t,−1)) dv ′
(m + n)!2s k2 (sinh(k|v(1) − v(−1)|))2 v v′ −v≥c
m+n=0

70
 ∞
X b 2(m+n)s |k|2(m+n+l1 +l2 )
(2λ)
ˆ
′
. 2s 2
sup e−2k(v −v) dv ′
m+n=0
(m + n)! k v v′ −v≥c
∞
X b 2(m+n)s |k|2(m+n+l1 +l2 ) ∞
X b 2(m+n)s (m + n)!2
(2λ) (2λ)
. . .
(m + n)!2s e2kc (m + n)!2s (c/2)2m+2n
m+n=0 m+n=0

P∞ CM
Now we use the fact that M =0 M !2s−2 . 1 for any fixed bounded C < ∞ to obtain
∞
X X b 2M s ∞
X b 2M s
(4λ/c) −2M s (8λ/c)
T1 1v≤v′ . . M 2 b c, s).
≤ C(λ,
M !2s−2 M !2s−2
M =0 m+n=M M =0

Now we invoke the estimate (6.14) and bound the expression (6.21) as follows:
∞
X
b e n |k|m+l1 ∂vl2 φ(I) k2L∞ .ϕ2 kw
a2m,n kχ∗ Γ
(I)
ek k2L2v . (6.22)
k k y
m+n=0

This concludes the proof of (6.5).

The derivation for the estimate (6.6) is similar, we have that
ˆ v(1)
(I) (I) (I) (I)
kχ∗ |k|m ∂vl φk kL∞
v
≤C χ∗ |k|m ∂vl Gk (v, v ′ )Rk (v ′ )dv ′ ≤ C(m)kRk kL2v ≤ C(m)kw
ek kL2v .
v(−1)
L∞
v

To derive the estimate (6.7), we modify the previous computation as follows

∞
X
e n |k|m ∂ l φ(I) k2 ∞
a2m,n kχ∗ Γ
b k v k Ly
m+n=0
∞ 2
X ′
ˆ v(1)
′ ′ (I)
≤ max bm,n
B 2
22n ϕ2(1+n ) χ∗ |k|m+n−n ∂vn +l Gk (v, v ′ )Rk (v ′ )dv ′
n′ ∈{0,··· ,n} v(−1)
m+n=0 L∞
v
∞ 2
X ˆ v(1)
∂ n′ e −iktv′
= max b 2 22n ϕ2(1+n′ ) χ∗
B
′ ′
|k|m+n−n ∂vn +l Gk (v, v ′ ) v iktv′
e
(I)
Rk (v ′ )dv ′
m,n
n′ ∈{0,··· ,n} v(−1) (−ikt)n
m+n=0 L∞
v
n 
X  X
∞
ϕ2 n bm,n
2 ′
. max B 22n ϕ2n
|kt|2n ℓ n′ ∈{0,··· ,n}
ℓ=0 m+n=0
2
ˆ v(1)
′ ′ ′ −iktv′ iktv′ e ℓ (I) ′
× χ∗ |k|m+n−n ∂vn +l ∂vn−ℓ
′ Gk (v, v )e e Γk Rk (v )dv ′
v(−1)
L∞
v
n
ϕ2 X e ℓ (I) 2
. ek kL2v .
kΓk w
|kt|2n
ℓ=0

Here in the last line, we have invoked (6.14) and applied a similar argument as in derivation of
(6.22).
Step # 2: Proof of (6.8). We apply the following argument:
ˆ v(1)
e n φ(I)
|k|m Γ =
(I)
|k|m (∂v + ikt)n Gk (v, v ′ )Rk (v ′ )dv ′ .
k k
v(−1)

71
Now we invoke the structure of the Green’s function (6.13)

Gk (v, v ′ ) = Hk (v − v ′ ) + Sk (v + v ′ ).

Here Hk (v − v ′ ) encodes the jumps of derivatives and the Sk (v + v ′ ) satisfies the following bound
(k > 0)
∂vn cosh(k(v + v ′ − v(t, 1) − v(t, −1)))
|k|m ∂vn Sk (v + v ′ ) = |k|m
2k sinh(k|v(t, 1) − v(t, −1)|)
n ′ ′
o
.|k|m+n max ek(v+v −2v(t,1)) , e−k(v+v −2v(t,−1))
.|k|m+n e−c|k| . C(c, m + n), dist(v ′ , {v(t, 1), v(t, −1)}) ≥ c > 0. (6.23)

Hence we have that

ˆ v(1)
(I)
e n φ =(∂v + ikt)n−1 (I)
|k|m Γ k k |k|m (−∂v′ + ikt) Gk (v, v ′ )Rk (v ′ )dv ′
v(−1)
ˆ v(1)
n−1 (I)
+ (∂v + ikt) |k|m (∂v + ∂v′ ) Gk (v, v ′ )Rk (v ′ )dv ′ .
v(−1)

Since (∂v + ∂v′ )Hk (v − v ′ ) = 0 and Gk (v, v ′ = ±1) = 0, we have that

ˆ v(1)
|k|m Γe n φ(I) =(∂v + ikt)n−1 (I)
|k|m Gk (v, v ′ ) (∂v′ + ikt) Rk (v ′ )dv ′
k k
v(−1)
ˆ v(1)
n−1 (I)
+ (∂v + ikt) |k|m 2∂v Sk (v + v ′ )Rk (v ′ )dv ′
v(−1)
ˆ v(1)
(I)
=(∂v + ikt)n−2 |k|m Gk (v, v ′ ) (∂v′ + ikt)2 Rk (v ′ )dv ′
v(−1)
ˆ v(1)
(I)
+ (∂v + ikt)n−2 |k|m 2∂v Sk (v + v ′ ) (∂v′ + ikt) Rk (v ′ )dv ′
v(−1)
ˆ v(1)
(I)
+ (∂v + ikt)n−2 |k|m 2∂v2 Sk (v + v ′ )Rk (v ′ )dv ′
v(−1)
ˆ v(1)
(I)
+ (∂v + ikt)n−2 |k|m 2∂v Sk (v + v ′ )(∂v′ + ikt)Rk (v ′ )dv ′
v(−1)
ˆ v(1)
(I)
− (∂v + ikt)n−2 |k|m 2∂v Sk (v + v ′ )∂v′ Rk (v ′ )dv ′ .
v(−1)

(I)
Since Rk and all its higher derivatives vanish on the boundary, we implement the integration by
parts in the last term,
ˆ v(1)
m e n (I) (I)
|k| Γk φk =(∂v + ikt)n−2
|k|m Gk (v, v ′ ) (∂v′ + ikt)2 Rk (v ′ )dv ′
v(−1)
ˆ v(1)
n−2 (I)
+ 4(∂v + ikt) |k|m ∂v Sk (v + v ′ ) (∂v′ + ikt) Rk (v ′ )dv ′
v(−1)
ˆ v(1)
(I)
+ 4(∂v + ikt)n−2 |k|m ∂v2 Sk (v + v ′ )Rk (v ′ )dv ′ .
v(−1)

72
Hence after n iteration steps, we have that
ˆ v(1) n
X ˆ v(1)
′ ′
e n φ(I)
|k|m Γ = m e n R(I) (v ′ )dv ′ +
|k| Gk (v, v )Γ ′
Cm′ ,n′ e n′ R(I) (v ′ )dv ′ .
|k|m+m ∂vm Sk (v + v ′ )Γ
k k k k k k
v(−1) m′ +n′ =0 v(−1)

′
Now we follow the same strategy as in the proof of (6.7). By applying the fact that ∂v′ (eiktv fk (v ′ )) =
′ ′
eiktv (∂v′ + ikt)fk (v ′ ) = eiktv Γe k fk , we obtain that
′
v(1)
∂v′ 2 e−iktv iktv′ e n (I) ′ ′
ˆ
e n φ(I)
|k|m Γ = m
|k| Gk (v, v ) ′
e Γk Rk (v )dv
k k
v(−1) (−ikt)2
n
X ˆ v(1) ′
′ ′ ∂v′ 2 e−iktv iktv′ e n′ (I) ′ ′
+ Cm′ ,n′ |k|m+m ∂vm Sk (v + v ′ ) e Γk Rk (v )dv
′ ′ v(−1) (−ikt)2
m +n =0
ˆ v(1)
1 e n+2 R(I) (v ′ )dv ′
= |k|m Gk (v, v ′ )Γ k k
(−ikt)2 v(−1)
ˆ v(1)
2 e n+1 R(I) (v ′ )dv ′ + 1 e n R(I) (v ′ )
+ 2
|k|m ∂v Gk (v, v ′ )Γ k k 2
|k|m Γ k k
(−ikt) v(−1) (−ikt)
n+2
X ˆ v(1)
1 e ′ ′
e n′ R(I) (v ′ )dv ′ .
+ 2 2 Cm′ ,n′ |k|m+m ∂vm Sk (v + v ′ )Γ k k
k t ′ ′ v(−1)
m +n =0

Here we recall the Sk -estimate for v ′ supported away from the boundary (6.23). By taking the L2
norm and invoking (6.14), we have the following
1000
X 1002
X 1002
X
bm,n
2 e n φ(I) k2 2 . 1 b 2 ′ ′ k|k|m′ Γ
e n′ R(I) k2 2 . 1 b 2 ′ ′ k|k|m′ Γ
e n′ w(I) k2 2 .
B k|k|m Γ k k Ly B m ,n k k Lv B m ,n k k Lv
k 4 t4 k 4 t4
m+n=0 m′ +n′ =0 m′ +n′ =0

This concludes the proof of (6.8).

(ℓ)
6.3 Bounds on Jell
We recall the equation (3.2b)
(E) (E) 2 (I) (E)
∆k φk =w
ek e1 (∂v − ∂y2 )φk ,
+χ φk |y=±1 = 0.


Recalling the vector field Γk fk = vy−1 ∂y + ikt fk = (∂v + ikt) fk , the χm+n cutoff (2.4) and the
q n weight (2.6), then the equation for χm+n |k|m q n Γnk φk can be rewritten as follows:
(E)
∆k (χm+n |k|m q n Γnk φk )
(E) 2 (I) (E)
=χm+n |k|m q n Γnk w
ek + χm+n |k|m q n Γnk (e
χ1 (∂v − ∂y2 )φk ) + χm+n |k|m q n [∂yy , Γnk ]φk
(E) (E)
+ χm+n [∂yy , q n ](|k|m Γnk φk ) + [∂yy , χm+n ](|k|m q n Γnk φk )
(E)
=:χm+n |k|m q n Γnk w
ek (I)
+ Cm,n + T1;m,n + T2;m,n + T3;m,n . (6.24)
(I)
Here Cm,n denotes the contribution from the interior and {Tj;m,n }3j=1 represents the commutator
terms. We recall the simplified notations
(E) (E) (E) (E)
φm,n;k :=χm+n |k|m q n Γnk φk , em,n;k := χm+n |k|m q n Γnk w
ω ek .

73
We expand the expression
2 (I) 2 (I) (I) 2 (I) (I)
(∂v − ∂y2 )φk =∂v φk − vy ∂v (vy ∂v φk ) = (1 − vy2 )∂v φk + vyy ∂v φk
2 (I) 1 (I)
= − (vy2 − 1)∂v φk + ∂v (vy2 − 1)∂v φk . (6.25)
2
The interior contribution can be rewritten as follows
 
(I) m n n 2 (I) 1 (I)
Cm,n := χm+n |k| q Γk −e χ1 Z ∂v φk + χ e1 ∂v Z ∂v φk , Z := vy2 − 1. (6.26)
2
| {z }
2 2 (I)
=χ e1 (∂v − ∂y )φk

Here the commutator terms are the following

!
X
n−1 
n  n−ℓ−1 ′′ 2 n−ℓ ′′

(E)
T1;m,n =χm+n |k|m q n 2∂v v ∂v + ∂v v ∂v Γℓk φk ; (6.27a)
ℓ
ℓ=0
(E)
T2;m,n =χm+n |k|m (n(n − 1)q n−2 (q ′ )2 + nq n−1 q ′′ + 2nq n−1 q ′ ∂y )Γnk φk ;
(E)
T3;m,n =(2χ′m+n ∂y + χ′′m+n )|k|m q n Γnk φk .

We organize our analysis as follows: in Subsection 6.3.1, we derive estimate on the interior con-
(I)
tribution Cm,n ; in Subsection 6.3.2, we derive estimate for small n; in Subsection 6.3.3, we derive
estimate for large n; in Subsection 6.3.4, we conclude.
6.3.1 Interior Contributions
We present the estimate of the second term in (6.24).
(I)
Lemma 6.6. For any M ∈ N, the following estimate of the Cm,n (6.26) holds
M
X M X
X
ϕ2n+2 k1suppχe1 J0,n
(I) (a,b,0)
a2m,n kCm,n
(I) 2
ek k2L2
kL2y .kw 2
B0,n Zk2L2 . (6.28)
m+n=0 n=0 a+b≤1

Moreover,
M
X M X
X
ϕ2n+2 k1suppχe1 J0,n
(I) (a,b,0)
a2m,n k∂y Cm,n
(I) 2
ek k2H 1
kL2y .kw 2
B0,n Zk2L2 . (6.29)
m+n=0 n=0 a+b≤2

Here, we use the notation Z := (vy2 − 1). Moreover, the implicit constant does not depend on M
and we can take the limit as M → ∞.
Proof. We divide the proof into two steps.
Step # 1: Proof of (6.28).
(I) 2 (I)
Thanks to Proposition 6.2, we observe that ∂v φk , ∂v φk have very high regularity.
Now we are ready to estimate the expression (6.28). We apply the product rule
M
X 2 (I)
a2m,n kχm+n |k|m q n Γnk (e
χ1 (∂v − ∂y2 )φk )k2L2y
m+n=0
M   !2
X 2 (I) 1 (I)
= a2m,n m n n
χm+n |k| q Γk χ e1 Z∂v φk − χ e1 ∂v Z∂v φk
2 L2y
m+n=0

74
 M n     !2
X X n n−ℓ 2 (I) 1 n−ℓ (I)
. a2m,n χm+n q n ∂v (e
χ1 Z) |k|m Γℓk ∂v φk − ∂v χ1 ∂v Z) |k|m Γℓk ∂v φk
(e
ℓ 2 L2y
m+n=0 ℓ=0
M n  
!2
X X n n−ℓ 2 (I)
. a2m,n χm+n q n ∂v χ1 Z) |k|m Γℓk ∂v φk
(e
m+n=0
ℓ L2y
ℓ=0
M n  
!2
X X n n−ℓ (I)
+ a2m,n χm+n q n ∂v (e
χ1 ∂v Z) |k|m Γℓk ∂v φk =: T1 + T2 . (6.30)
m+n=0
ℓ L2y
ℓ=0

Recalling the definitions (2.9), (3.10), (3.3), we have the following relation
λ(m+n)s ϕ(n+1)  s
am,n ((m+n)!)s n−ℓ (n + m − (m + ℓ))!(ℓ + m)!
= b (ℓ+m)s ϕ(ℓ+1)
=ϕ
B0,n−ℓ b
am,ℓ λ(n−ℓ)s
× λ(t) (n + m)!
((n−ℓ)!)s ((ℓ+m)!)s
 −s  −s
n−ℓ n + m n−ℓ n
=ϕ ≤ϕ . (6.31)
m+ℓ ℓ
Now we apply the Hölder inequality to obtain that
!2
M
X X
n−1
n

am,n n−ℓ 2 (I)
T1 . B0,n−ℓ χn−ℓ q n−ℓ ∂v (e
χ1 Z) am,ℓ k|k|m Γℓk ∂v φk kL∞
b
ℓ B0,n−ℓ b
am,ℓ L2y v
m+n=0 ℓ=0
M n  1−s
!2
X X n n−ℓ 2 (I)
n−ℓ n−ℓ
. ϕ B0,n−ℓ χn−ℓ q ∂v (e
χ1 Z) am,ℓ k|k|m Γℓk ∂v φk kL∞
b v
.
ℓ L2y
m+n=0 ℓ=0

Now we have that by the relation (2.1), the combinatorial fact (6.88), and interchanging the order
of summation,
! n−1 !
M
X X n−2σ−2σ∗
n−1 X 2
2n−2ℓ 2 n−ℓ n−ℓ 2 m ℓ 2 (I) 2
T1 . ϕ B0,n−ℓ;0 χn−ℓ q ∂v χ1 Z) 2 b
(e am,ℓ k|k| Γk ∂v φk kL∞
ℓ L v
m+n=0 ℓ=0 ℓ=0
M M
X X −m n−1
X n−ℓ 2 2 (I)
. ϕ2n−2ℓ B0,n−ℓ
2
χn−ℓ q n−ℓ ∂v (e
χ1 Z) a2m,ℓ k|k|m Γℓk ∂v φk k2L∞
b v
L2
m=0 n=0 ℓ=0
XM M −m−1
X MX −m
n−ℓ 2 2 (I)
= ϕ2n−2ℓ B0,n−ℓ
2
χn−ℓ q n−ℓ ∂v (e
χ1 Z) a2m,ℓ k|k|m Γℓk ∂v φk k2L∞
b v
L2
m=0 ℓ=0 n=ℓ+1
M
! M
!
X n X 2 (I)
2
. B0,n χ1 Z)k2L2
ϕ2n kχn q n ∂v (e a2m,ℓ k|k|m Γℓk ∂v φk k2L∞
b v
.
n=0 m+ℓ=0

Thanks to the Proposition 6.2, we have that

M
!
X n (I)
2
T1 . B0,n χ1 Z)k2L2
ϕ2n+2 kχn q n ∂v (e ek k2L2v .
kw
n=0

A similar argument yields that

M
!
X n (I)
2
T2 . B0,n ϕ2n+2 kχn q n ∂v (e
χ1 ∂v Z)k2L2 ek k2L2v .
kw
n=0

75
Combining the bounds and (6.30) yields that
M
X M
X  
n n (I)
a2m,n kCm,n
(I) 2
kL2y . 2
B0,n χ1 Z)k2L2 + kχn q n ∂v (e
ϕ2n+2 kχn q n ∂v (e ek k2L2v
χ1 ∂v Z)k2L2 kw
m+n=0 n=0
!
M
X   1
X
n+1 ℓ
ϕ2 k1suppχe1 ∂v Zk2L2
(I) (0,0,0)
ek k2L2v
.kw 2
B0,n ϕ2n+2 kJ0,n Zk2L2 + kχn q n ∂v Zk2L2 + .
n=1 ℓ=0

Now we invoke Lemma 6.20 to obtain (6.28). This concludes Step # 1.

Step # 2: Proof of (6.29). We expand the expression on the left hand side as follows
M
X    2
2 (I)
a2m,n ∂y χm+n |k|m q n Γnk χe1 (∂v − ∂y2 )φk
L2y
m+n=0
M
X    2
2 (I)
. a2m,n χm+n |k|m q n ∂v Γnk χ
e1 (∂v − ∂y2 )φk
L2y
m+n=0
M
X   2
2 (I)
+ a2m,n ∂v (χm+n q n ) |k|m Γnk χe1 (∂v − ∂y2 )φk =: T3 + T4 . (6.32)
L2y
m+n=0

We first estimate the T3 , which can be further decomposed into two sub-terms (m + n 6= 0 and
m + n = 0) with the expression (6.25):
M
X    2
2 (I)
T3 ≤ a2m,n χm+n |k|m q n ∂v Γnk χ
e1 Z ∂v φk
L2y
m+n=0
M
X    2
(I)
+ a2m,n χm+n |k|m q n ∂v Γnk χ
e1 ∂v Z ∂v φk
L2y
m+n=0
M
X 2
X   2
l 3−l (I)
. a2m,n χm+n |k|m q n Γnk ∂v (e
χ1 Z) ∂v φk
L2y
m+n=1 l=0
2 X
X 3
ℓ1 ℓ2 (I)
+ 1m+n=0 ϕ 2
k1suppχe1 ∂v Zk2L2 k∂v φk k2L2 =: T31 + T32 . (6.33)
ℓ1 =0 ℓ2 =0

Thanks to the Sobolev estimate (6.15), we have that the second term is bounded
2
X
T32 . 1m+n=0 ϕ2 k1suppχe1 J0,0
2 (0,b,0) (I)
B0,0 Zk2L2 kwk k2H 1 . (6.34)
b=0

Next, we focus on the estimate of the first term T31 . We recall the property of the cutoff χ∗ (6.4)
that χ∗ ≡ 1 on the support of χm+n , m + n 6= 0, and apply the product rule to derive that
n  
!2
M
X 2
X X n   
n−ℓ+l 3−l (I)
T31 . a2m,n χm+n χ∗ q n−ℓ ∂v χ1 Z) |k|m q ℓ Γℓk ∂v φk
(e
ℓ L2y
m+n=1 l=0 ℓ=0
M 2 n  −σ−σ∗
!2
X X X n n−ℓ+l 3−l (I)
. B0,n−ℓ ϕn−ℓ χn−ℓ q n−ℓ ∂v (e
χ1 Z) bm,ℓ
a χ∗ |k|m q ℓ Γℓk ∂v φk .
ℓ L2y L∞
y
m+n=1 l=0 ℓ=0

76
Here in the last line, we have used the relation (6.31) and the Hölder inequality. Then we apply
(I)
the bound of φk (6.5), the estimate of Z = vy2 − 1 (6.113) , and the combinatorial fact (6.88) to
obtain that
2 M n  −2σ
! n 
X X X n X X 2 3−l (I) 2
ϕ2n−2ℓ 1suppeχ1 J0,n−ℓ Z 2 b
 2 (a,b,0) 
T31 . B0,n−ℓ a2m,ℓ χ∗ |k|m q ℓ Γℓk ∂v φk
ℓ Ly L∞
y
l=0 m+n=1 ℓ=0 ℓ=0 a+b≤2
 
X2 X M M X−m X n X 2 3−l (I) 2
ϕ2n−2ℓ 1suppeχ1 J0,n−ℓ Z 2  b
2 (a,b,0)
.  B0,n−ℓ a2m,ℓ χ∗ |k|m q ℓ Γℓk ∂v φk ∞
Ly Ly
l=0 m=0 n=0 ℓ=0 a+b≤2
 
X M M
2 X X −m M
X −m X
3−l (I) 2 2
. a2m,ℓ
b χ∗ |k|m q ℓ Γℓk ∂v φk  2
B0,n−ℓ ϕ2n−2ℓ 1suppeχ1 J0,n−ℓ
(a,b,0)
Z 
L∞ y L2y
l=0 m=0 ℓ=0 n=ℓ a+b≤2
M X
X
ϕ2n+2 k1suppχe1 J0,n
(I) (a,b,0)
ek k2L2
. kw 2
B0,n Zk2L2y .
n=0 a+b≤2

Combining this bound with (6.34) yields that the T3 term is consistent with (6.29).
To estimate the T4 -term in (6.32), we invoke the expansion (6.26), the bound (2.5a), the bound
(2.1) and the observation that the term with m + n = 0 is zero. Then we apply the argument as in
(6.33) to obtain that
M
X 1
X   2
T4 . 1n≥1 a2m,n (m + n)1+σ χ∗ χm+n−1 q n−1 |k|m (∂v + ikt)Γkn−1 ∂v l Z ∂v 2−l φk(I)
L2y
m+n=1 l=0
M X
X 1   2
l 2−l (I)
+ a2m,0 m1+σ χ∗ χ
e1 χm−1 |k|m ∂v Z ∂v φk
L2y
m=1 l=0
2 M n  
X X X λ2s a2m,n−1 ϕ2 n − 1 3−l (I) 2
. 1n≥1 1suppeχ1 χn−1−ℓ q n−1−ℓ ∂v n−1−ℓ+l Z χ∗ |k|m Γℓk ∂v φk
(m + n)2σ∗ ℓ L2 L∞
l=0 m+n=1 ℓ=0
1 M n  
X X X λ2s a2m,n−1 ϕ2 t2 n − 1 2−l (I) 2
+ 1suppeχ1 χn−1−ℓq n−1−ℓ ∂v n−1−ℓ+l Z χ∗ |k|m Γℓk |k|∂v φk
(m + n)2σ∗ ℓ L2 L∞
l=0 m+n=1 ℓ=0
1
X M
X 2−l
k1
2s (0,b,0) (I)
+λ e1 J0,0
suppχ Zk2L2 a2m−1,0 m−2σ∗ kχ∗ |k|∂v |k|m−1 φk k2L∞
y
.
b=0 m=1

Now we invoke the bound (6.5), the combinatorial fact (6.88), the estimates (6.112), (6.113) to
obtain that
2
X M
X n
X λ2s n−1−ℓ+l
T4 . 2σ
2
Bn−1−ℓ ϕ2n−2−2ℓ 1suppχe1 χn−1−ℓ q n−1−ℓ ∂v Z
(m + n) ∗ L2
l=0 m+n=1 ℓ=0
 
3−l (I) 2 2−l (I) 2
×b 2
am,ℓ χ∗ |k|m Γℓk ∂v φk ∞
+ χ∗ |k|m Γℓk |k|∂v φk
L L∞
1
X
ϕ2 k1suppχe1 J0,0
(0,b,0) (I)
+ λ2s Zk2L2 kw
ek k2L2v
b=0
M X
X
ϕ2n+2 k1suppχe1 J0,n
(I) (a,b,0)
ek k2L2v
.kw 2
B0,n Zk2L2 .
n=0 a+b≤2

77
6.3.2 Lower regularity bounds
In this section, we derive estimates with small-n level. To simplify the notation, we introduce the
norm/semi-norm
X X
kfk k2Ḣ j := k∂yb |k|c fk k2L2y , kfk k2H j := kfk k2 j′ .
k k Ḣk
b+c=j j ′ ≤j

The first main lemma is the following.

Lemma 6.7. Assume the bootstrap assumption. The following estimates hold for n ≤ 1 and
arbitrary M ∈ N,
M
X X M
X  
e˚
(a,b,c) (E) (I) 2
a2m,0 kJm,0 φk k2L2 . a2m,0 kw (E) 2
m,0;k kL2 + kCm,0 kL2y , (6.35a)
m=0 a+b+c≤2 m=0
M
X X M
X  
e˚
(a,b,c) (E) (I) 2 (I)
a2m,0 kJm,0 φk k2L2 . a2m,0 kw (E) 2
m,0;k kḢ 1 + kCm,0 kH 1 + |vyy | y=±1
2
ek k(6.35b)
kw L2v ,
k k
m=0 a+b+c=3 m=0
M
X −1 X M
X  
(a,b,c) (E)
a2m,1 kJm,1 φk k2L2 . 1n∈{0,1} a2m,n kJm,n
(0) (E) 2 (I) 2
ek kL2 + kCm,n
w kL2y , (6.35c)
m=0 a+b+c≤2 m+n=0
M
X −1 X M
X  
(a,b,c) (E)
a2m,1 kJm,1 φk k2L2 . 1n∈{0,1} a2m,n kJm,n
(0) (E) 2 (I) 2
ek kH 1 + kCm,n
w
k
(I)
ek k2L2v .
kH 1 + |vyy | y=±1 kw
k
m=0 a+b+c=3 m+n=0
(6.35d)

Here the implicit constant is independent of M and S1n -set is defined in (3.7).
Proof. Step # 0: Preliminaries and 0-th level estimates. First of all, we observe the following
equivalence relation for k 6= 0, ℓ ∈ {2, 3},
X X
(0,b,c)
kJm,n fk kL2 ≈ k∂yb |k|c (χm+n |k|m q n Γnk fk )kL2 . (6.36)
b+c=ℓ b+c=ℓ

Here the implicit constant depends on kvy kL∞ W ℓ−1,∞ and kvy−1 kL∞ t,y
.
t y
Thanks to the elliptic estimate (6.85), the relations (6.36)ℓ=2 , (6.26) and the bound (6.28), we
have that
X (0,b,c) (E) (E) (E) (I)
kJ0,0 φk kL2y . kφk kḢ 2 . kw ek kL2y + C0,0 2 . (6.37)
k Ly
b+c=2

This is consistent with (6.35a)m=0 .

Next we prove the m = 0 case of the estimate (6.35b). From the relations
(E) (E) (E) 2 (I) 2 (I) 2 (I)
∂yy φk − |k|2 φk =w
ek e1 (∂v − ∂y2 )φk ,
+χ (∂v − |k|2 )φ(I) = w ec1 (∂v − ∂y2 )φk ,
ek + χ
(E) (I) 2 (I)
e(E) (y = ±1) = φk (v = v(±1)) = ∂v φk (v = v(±1)) = 0,
φk (y = ±1) =w

one obtain that

 
(E) (I) 2 (I) (I) (I)
∂yy φk = −∂yy φk = −vy2 ∂v φk − vyy ∂v φk = −vyy ∂v φk .
y=±1 y=±1 y=±1 y=±1

78
 (I)
Thanks to the solution expression (6.13) and the fact that v(±1) and the support of Rk have order
1 distance, we have that by (6.6)

(I) C (I)
|∂v φk (v = v(±1))| ≤ e k 2.
kw (6.38)
|k|2 k Lv

Hence the following relations hold

(E) (E) 2 (I) (E)
∆k |k|φk =|k|w
ek χ1 (∂v − ∂y2 )φk ,
+ |k|e |k|φk y=±1
= 0;
(E) (E) 2 (I) (E) (I)
∆k ∂y φk =∂y w
ek χ1 (∂v − ∂y2 )φk ),
+ ∂y (e ∂yy φk y=±1 = −vyy ∂v φk y=±1
.

(I)
Thanks to the elliptic bounds (6.84), (6.86), the definition of C0,0 (6.26), and the estimate (6.15)
the following estimates hold

(E) (E) (E) (E) (I) (I)

ek kL2 + kC0,0 kḢ 1 + |k|1/2 |vyy ∂v φk |
ek kL2 + k∂y w
k|k|φk kḢ 2 + k∂y φk kḢ 2 . k|k|w
k k k
v=v(±1)
(E) (I) (I)
ek kḢ 1
.kw + kC0,0 kḢ 1 + |vyy | y=±1 kw
ek kL2v .
k k

Combining it with (6.36) yields that

X (0,b,c) (E) (E) (I) (I)
kJ0,0 φk kL2 . kwek kḢ 1 + kC0,0 kḢ 1 + |vyy | y=±1
kw
ek kL2v .
k k
b+c=3

The implicit constant depends on kvy−1 kL∞ , kvy kW 2,∞ . This is consistent with (6.35b)m=0 . This
concludes the proof at the zeroth level.
Step # 1: Proof of (6.35a).
   
(E) (E) (E) (I) (E)
∆k χm |k|m φk ek + [∂yy , χm ] |k|m φk
=χm |k|m w + Cm,0 , χm |k|m φk (y = ±1) = 0.

Thanks to the elliptic estimate (6.84), we have that

X (0,b,c) (E) (E) (E) (I)
a2m,0 kJm,0 φk k22 .a2m,0 kχm |k|m wek k2L2 + a2m,0 k[∂yy , χm ](|k|m φk )k2L2 + a2m,0 kCm,0 k2L2
b+c=2
=:T11 + T12 + T13 . (6.39)

The T11 and T13 are already consistent with (6.35a), so we focus on the T12 -term. If m = 0, the
commutator is vacuous. If m = 1, the term can be bounded by applying (6.37)

(E) (E) (E) (I) 2

ek k2L2y + C0,0
T12 . a21,0 k∂y |k|φk k2L2y + a21,0 k|k|φk k2L2y . kw .
L2y

Finally, if m ≥ 2, we estimate the T12 term with (2.1) and the equivalence relation (6.36), as follows:
(E) (E)
T12 .a2m,0 kχ′′m |k|2 (χm−2 |k|m−2 φk )k2L2 + a2m,0 kχ′m |k|∂y (χm−1 |k|m−1 φk )k2L2
(E) (E)
.λ4s a2m−2,0 k|k|2 (χm−2 |k|m−2 φk )k2L2 + λ2s a2m−1,0 k|k|∂y (χm−1 |k|m−1 φk )k2L2
m−1
X X
2s (0,b,c) (E)
.λ a2m′ ,0 kJm′ ,0 φk k2L2 .
m′ =m−2 b+c=2

79
Now we sum from m = 0 to M in (6.39) to obtain that
M X
X (0,b,c) (E)
a2m,0 kJm,0 φk k2L2y
m=0 b+c=2
M
X M
X M
X −1 X
(E) (I) 2 (0,b,c) (E)
≤C a2m,n kw
em,0 k2L2y +C a2m,0 Cm,0 2 + Cλ 2s
a2m,0 kJm,0 φk k2L2y .
Ly
m=0 m=0 m=0 b+c=2

By taking the λ > 0 small enough compared to the constant appeared in the expression, we absorb
the last term by the left hand side and obtain (6.35a).
Step # 2: Proof of (6.35b). We consider the m ≥ 1 case. Since (a, b, c) ∈ S03 , the only non-
vacuous terms on the left hand side come from k∂yb |k|c φ˚
(E)
m,0;k kL2 , b + c = 3. First, we observe
that the estimate (6.35a) yields that
M X
X M
X M
X
(E) 2 (E) 2 (I) 2
a2m,0 ∂yb |k|c+1 φk . a2m,0 |k|w
ek + a2m,0 |k|Cm,0 . (6.40)
L2 L2 L2y
m=0 b+c=2 m=0 m=0

Hence the only unknown component is for ∂y3 φ˚ (E)

m,0;k . To obtain this, we consider the term
∂y (φ˚
(E) m (E)
m,0;k ) = ∂y (χm |k| φk ). We observe that it solves the following equation with Neumann
boundary condition,
 
∆k ∂y φ˚ ˚ m (E)
+ ∂y Cm,0 , ∂yy φ˚
(E) (E) (I) (E) m (I)
m,0;k = ∂y w
e m,0;k + ∂ [∂
y yy , χ m ] |k| φk m,0;k y=±1 = −|k| vyy ∂v φk v=v(±1)
.

The main term to estimate is the second term on the right hand side, we rewrite it in the following
fashion,
     
(E) m (E) m (E) (E)
∂y [∂yy , χm ] |k|m φk = χ′′′ ′′
m |k| φk + 2χm ∂y |k| φk + 2χ′m ∂yy |k|m φk

 ′′′ 3 ˚(E) ′′ 2 ˚
(E) ′ ˚
(E)
 χm |k| φ m−3,0;k + 2χm |k| ∂y φ m−2,0;k + 2χm |k|∂yyφ m−1,0;k
  , m ≥ 3;
′′′ 2 (E) ′′ 2 (E) ′ (E)
= χ2 |k| φk + 2χ2 |k| ∂y φk + 2χ2 |k|∂yy χ1 |k|φk , m = 2;


 ′′′
χ |k|φ
(E) ′′
+ 2χ |k|∂ φ
(E) ′ (E)
+ 2χ |k|∂ φ , m = 1.
1 k 1 y k 1 yy k

Here φ˚(E) m (E)

m,0;k = χm |k| φk . Thanks to the elliptic estimate (6.86), (6.48)m=0 , (6.49)m=0 , (6.5),
and (6.38), we have that for m = 1,

a1,0 k∂y φ˚ e˚
(E) (E) (E) (E)
1,0;k kḢ 2 .a1,0 kw
k
1,0;k kḢ 1 + a1,0 k|k|∂yy φk kL2 + a1,0 k|k|∂y φk kL2
k

(I) (I)
+ a1,0 k∂y C1,0 kL2 + a1,0 |k|1/2 ||k|vyy ∂v φk |
v=v(±1)
1
X 1
X
e˚(E) (I) (I)
. am,0 kw m,0;k kḢ 1 + am,0 kCm,0 kH 1 + |vyy | y=±1
ke
ωk kL2v .
k k
m=0 m=0

This is consistent with (6.35b). For the m = 2 case, we estimate it using the bounds (6.5), (6.38),
and (6.40),
1
X X
a2,0 k∂y φ˚
(E) e˚
2,0;k kḢ 2 .a2,0 kw
(E)
2,0;k kḢ 1 + am,0 k∂yb |k|c+1 φ˚
(E)
m,0;k kL2
k k
m=0 b+c=2

80
 (I) (I)
+ a2,0 k∂y C2,0 kL2 + a2,0 |k|1/2 |k|2 vyy ∂v φk
v=v(±1)
2
X 2
X 2
X  
e˚
(I) (I)
. am,0 kw (E)
m,0;k kḢ 1 + am,0 kCm,0 kH 1 + |vyy | y=±1
am,0 χ∗ ∂v |k|m+1/2 φk
k k ∞
m=0 m=0 m=0
X2 X2
e˚(E) (I) (I)
. am,0 kw m,0;k kḢ 1 + am,0 kCm,0 kH 1 + |vyy | y=±1
ω
ek .
k k L2v
m=0 m=0

Here χ∗ is defined in (6.4). Finally, we consider the case where m ≥ 3. In this case, the Gevrey
coefficients starts to play major roles. We recall the fundamental relations discussed in the previous
sections (4.29), and the estimates of the cutoff χm+n (2.5a),

am,0 k∂y φ˚
(E)
m,0;k kḢ 2
k

e˚
.am,0 kw (E)
m,0;k kḢ 1 + am,0 m
3+3σ
k|k|3 φ˚
(E)
m−3,0;k kL2 + am,0 m
2+2σ
k|k|2 ∂y φ˚
(E)
m−2,0;k kL2
k

+ am,0 m1+σ k|k|∂y2 φ˚

(I) 1/2 (I)
(E)
m−1,0;k kL2 + am,0 k∂y Cm,0 kL2 + am,0 |k| |k|m vyy ∂v φk
v=v(±1)

e˚
.am,0 kw (E) 3s
m,0;k kḢ 1 + λ am−3,0 m
−3σ∗
k|k|3 φ˚
(E) 2s
m−3,0;k kL2 + λ am−2,0 m
−2σ∗
k|k|2 ∂y φ˚
(E)
m−2,0;k kL2
k

+ λs am−1,0 m−σ∗ k|k|∂y2 φ˚

(I) 1/2 (I)
(E)
m−1,0;k kL2 + am,0 k∂y Cm,0 kL2 + am,0 |k| |k|m vyy ∂v φk
v=v(±1)
m−1
X X
e˚ k∂yb |k|c+1 φ˚
(E) s (E) ′ (I) (I)
.am,0 kw m,0;k kḢ 1 + λ m ,0;k kL2 + am,0 kCm,0 kH 1 + |vyy | y=±1
ωk kL2v .
ke
k k
m′ =m−3 b+c=2

Now we sum over the estimates we obtained so far and invoke the bound (6.40) to obtain
M X
X M
X M
X
a2m,0 k∂yb+1 |k|c φ˚ 2 (I) (I)
(E)
m,0;k kL2 . a2m,0 kẘm,0;k k2Ḣ 1 + a2m,0 kCm,0 k2Ḣ 1 + |vyy | y=±1
ωk kL2v .
ke
k k
m=0 b+c=2 m=0 m=0

Combining this estimates and (6.40) yields (6.35b).

Step # 3: Proof of (6.35c). Here we highlight that on the left hand sides of (6.35c), there
(1,0,0) (E)
is an extra term Jm,1 φk to be estimated. Hence, we further decompose the argument in two
substeps:
(0,b,c) (E)
Step # 3a: In this substep, we only consider the Sobolev component kJm,1 φk k2L2 on the left
hand side of (6.35c). First of all, we have the commutator relation

[A, BCD]f =ABCDf − BCDAf

=ABCDf − BACDf + BACDf − BCADf + BCADf − BCDAf
=[A, B](CDf ) + B[A, C]Df + BC[A, D]f. (6.41)

By setting A = ∂yy , B = χm+1 , C = q, D = Γk , we can derive the following equation for φ˚

(E)
m,1;k :

∆k φ˚
(E) m (E)
m,1;k =∆k (χm+1 |k| qΓk φk )
(E) (E) 2 (I)
=χm+1 |k|m qΓk w
ek + [∂yy , χm+1 qΓk ]|k|m φk + χm+1 |k|m qΓk (e
χ1 (∂v − ∂y2 )φk )
(E) (E) (E)
=χm+1 |k|m qΓk w
ek + [∂yy , χm+1 ](|k|m qΓk φk ) + χm+1 [∂yy , q](|k|m Γk φk )

81
 (E) (I)
+ χm+1 q[∂yy , Γk ]|k|m φk + Cm,1 ,
φ˚
(E) m (E)
m,1;k (y = ±1) =χm+1 |k| qΓk φk (y = ±1) = 0.

Thanks to the equivalence (6.36) and the elliptic estimate (6.84), we have that
X X
a2m,1 k∂yb |k|c φ˚
(0,b,c) (E)
a2m,1 kJm,1 φk k22 . (E) 2
m,1;k kL2
b+c≤2 b+c=2
(E) (E) (E)
.a2m,1 kχm+1 |k|m qΓk w
ek k2L2 + a2m,1 k[∂yy , χm+1 ](|k|m qΓk φk )k2L2 + a2m,1 kχm+1 [∂yy , q](|k|m Γk φk )kL2
(E) (I)
+ a2m,1 kχm+1 q[∂yy , Γk ]|k|m φk k2L2 + a2m,1 kCm,1 k2L2
5
X
=: T3j . (6.42)
j=1

Since the T31 , T35 terms are consistent with (6.35c), we focus on the remaining terms and start
from T32 . Thanks to a similar argument as in the estimate of T12 -term in (6.39), we distinguish
between m = 0 and m ≥ 1 cases. If m = 0, we apply the facts that am,n = Bm,n ϕn+1 (2.12a) and
tϕ ≤ C (2.11),
2 (E) (E)
T32 .B0,1 ϕ4 k∂y (q(∂v + ikt)φk )k2L2y + B0,12
ϕ4 k(∂v + ikt)φk k2L2y
X (E) (E) (I) 2
2
. B0,0 ϕ2 k∂yb |k|φk k2L2 . kw
ek k2L2y + C0,0 2 .
Ly
b+c=2

Here in the last line, we have invoked (6.37). On the other hand, if m ≥ 1, we estimate the T32
term with (2.1), the χm+n -estimate (2.5a), the relation (4.29) and the equivalence relation (6.36),
as follows:
(E) (E)
T32 .a2m,1 kχ′′m+1 |k|q(∂v + ikt)(χm−1 |k|m−1 φk )k2L2 + a2m,1 kχ′m+1 ∂y (q(∂v + ikt)(χm |k|m φk ))k2L2
X (E)
X (E)
.λ4s a2m−1,0 (tϕ)2 k∂yb |k|c (χm−1 |k|m−1 φk )k2L2 + λ2s a2m,0 (tϕ)2 k∂yb |k|c (χm |k|m φk )k2L2
b+c=2 b+c=2
m
X X (0,b,c) (E)
.λ2s a2m′ ,0 kJm′ ,0 φk k2L2 .
m′ =m−1 b+c=2

The T33 term in (6.42) can be estimated as follows

(E) (0,2,0) (E) 2 (0,1,1) (E) 2
T33 .a2m,1 kχm+1 ∂v (∂v + ikt)(χm |k|m φk )k2L2 . λ2s a2m,0 ϕ2 (kJm,0 φk kL2 + t2 kJm,0 φk kL2 )
X (0,b,c) (E)
.λ2s a2m,0 kJm,0 φk k2L2 .
b+c=2

Thanks to the commutator relation (4.7b), we have that the T34 term can be estimated as follows:
 
2 (E) (E)
T34 .a2m,1 kχm+1 qvyy ∂v |k|m φk k2L2 + kχm+1 qvy−1 vyyy ∂v |k|m φk k2L2
 
2 (E) (E)
.λ2s a2m,0 (m + 1)−2−2σ−2σ∗ k∂v (χm |k|m φk )k2L2 + k∂v (χm |k|m φk )k2L2
 
(0,2,0) (E) (0,1,1) (E)
.λ2s a2m,0 kJm,0 φk k2L2 + kJm,0 φk k2L2 .

Now we sum from m = 0 to M − 1 and invoke (6.35a) to obtain (6.35a).

82
 (1,0,0) (E)
Step # 3b: In this substep, we consider the estimate of the quantity Jm,1 φk . We observe
that
(1,0,0) (E) m+1 (E) (E)
Jm,1 φk = χm+1 |k|m q(∂v + ikt)φk = (m + 1)χm+1 (∂v + ikt)(χm |k|m φk ).
q

Hence the estimate is a direct consequence of (6.35a),

M
X −1 M
X −1 X
(1,0,0) (E) 2
k∂v |k|c φ˚
b
a2m,1 Jm,1 φk . a2m,1 (m + 1)2 hti2 (E) 2
m,0;k kL2
L2
m=0 m=0 b+c=1
M
X −1 X M
X −1 M
X −1
k∂v |k|c φ˚ e˚
b (I)
.λ2s a2m,0 (E) 2
m,0;k kL2 . a2m,0 kw (E) 2
m,0;k kL2 + a2m,0 kCm,0 k2L2y .
m=0 b+c=1 m=0 m=0

This concludes the proof of (6.35c).

Step # 4: Proof of (6.35d). Since a + b + c = 3 ≥ 1 in this case, we only needs to consider
classical derivative estimates.
From (6.35c), we have that
M
X −1 X M
X M
X
b (0) (E)
a2m,1 k∂v |k|c+1 Jm,1 φk k2L2 . 1 2 (0) (E) 2
ek kL2
n∈{0,1} am,n k|k|Jm,n w + 1n∈{0,1} a2m,n kCm,n
(I) 2
kH 1 .
k
m=0 b+c=2 m+n=0 m+n=0

We observe that the ∂y (φ˚

(E) m (E)
m,1;k ) = ∂y (χm+1 |k| q(∂v + ikt)φk ) solves the following equation:

∆k ∂y φ˚ e˚ m (E)
(E) (E) (I)
m,1;k =∂y w m,1;k + ∂y [∂yy , χm+1 qΓk ](χm |k| φk ) + ∂y Cm,1 .

Furthermore, we have that

∂y φ˚
(E)
(E)
m,1;k y=±1
=vy−1 ∂y |k|m φk y=±1
.

Hence the elliptic estimate (6.87) yields that

X
a2m,1 k∂yb+1 |k|c φ˚
(E) 2 2
e˚
m,1;k kL2 .am,1 k∂y w
(E) 2 2 m (E) 2
m,1;k kL2 + am,1 k∂y [∂yy , χm+1 qΓ](χm |k| φk )kL2
b+c=2
(I) (E)
+ a2m,1 k∂y Cm,1 k2L2 + a2m,1 |k|3 |∂v (|k|m φk )(y = ±1)|2
4
X
=: T4j . (6.43)
j=1

Since the T41 and T43 are consistent with (6.35d), we focus on T42 and T44 . We distinguish between
the m = 0, m = 1 and the m ≥ 2 case.
Step # 4a: the m = 0 case.
For the T42 term in (6.43), we explicitly write it out as follows,
  2
(E) (E) (E) (E)
T42 .a20,1 ∂y ∂yy (χ1 qvy−1 )∂y φk + 2∂y (χ1 qvy−1 )∂yy φk + ∂yy (qχ1 )iktφk + 2∂y (qχ1 )ikt∂y φk
L2
X (E) (E) (I) (I)
.a20,1 hti2 k|k|b ∂yc φk k2L2 . a20,0 kw ek k2L2v + a20,0 kC0,0 k2H 1 .
ek k2Ḣ 1 + |vyy |2 y=±1 kw
k k
b+c=3

83
This is consistent with (6.35d). To estimate the T44 term, we invoke the Gagliardo-Nirenberg
interpolation inequality to obtain that
(E) (E) (E) (E)
|∂y φk (y = ±1)|2 . k∂y φk kL2 (k∂yy φk kL2 + k∂y φk kL2 ). (6.44)

Hence, we invoke (6.40) to obtain that

(E) (E) (E)
T44 .a20,1 |k|3 k∂y φk k2 (k∂yy φk k2 + k∂y φk k2 )
(E) (E) (E)
.a20,1 k|k|2 ∂y φk k2 (k|k|∂yy φk k2 + k|k|∂y φk k2 )
(E) (I) (I)
.a20,0 kw
ek k2Ḣ 1 + |vyy | y=±1
ek k2L2v + a20,0 kC0,0 k2H 1 .
kw
k

This is consistent with (6.35d). Combining the estimates above, we conclude Step # 4a.
Step # 4b: the m = 1 case. The treatment is similar to Step # 4a and we omit the details.
Step # 4c: the m ≥ 2 case.For the T42 term, we apply the commutator relation (6.41) to
decompose it as follows
(E)
T42 =a2m,1 k∂y [∂yy , χm+1 qΓ](χm |k|m φk )k2L2

(E) (E)
2
.am,1 k∂y [∂yy , χm+1 ](χm |k|m qΓk φk )k2L2 + k∂y (χm+1 [∂yy , q](χm |k|m Γk φk ))kL2
 3
X
(E)
+ k∂y (χm+1 q[∂yy , Γk ](χm |k|m φk )k2L2 =: T42j . (6.45)
j=1

(4)
To estimate the T21 -term in (6.45), we invoke the relation (2.1) and the estimate (2.5a), together
(a,b,c) (E)
with the definition of Jm,n φk (3.6b) as follows
(E) (E)
T421 .a2m,1 (m + 1)6+6σ kχm |k|m qΓk φk k2L2 + a2m,1 (m + 1)4+4σ k∂y (χm |k|m qΓk φk )k2L2
(E)
+ a2m,1 (m + 1)2+2σ k∂yy (χm |k|m qΓk φk )k2L2
m+1 2 (E)
 2
.a2m−2,1 λ4s (m + 1)−4σ |k| χm−2 |k|m−2 qΓk φk
q L2
2
m+1 (E)
+ a2m−1,1 λ2s (m + 1)−4σ ∂v |k|(χm−1 |k|m−1 qΓk φk )
q L2
(E)
+ a2m−1,1 λ2s (m −2σ
+ 1) k|k|∂yy (χm |k| m−1
qΓk φk )k2L2
(1,0,2) (E) 2 (1,1,1) (E) 2 (0,2,1) (E)
.a2m−2,1 λ4s Jm−2,1 φk + a2m−1,1 λ2s Jm−1,1 φk + a2m−1,1 λ2s kJm−1,1 φk k2L2 .
L2 L2

Next we recall that on the support of χm+1 , χm = χm−1 ≡ 1 and estimate the T422 -term as follows
 
(E) (E)
T422 .a2m,1 k∂y (χm+1 q ′′ (χm |k|m Γk φk ))k2L2 + k∂y (χm+1 q ′ ∂y (χm |k|m Γk φk ))k2L2

(E) (E)
.am,1 k∂y (χm+1 q ′′ )(χm |k|m Γk φk )k2L2 + kχm+1 q ′′ ∂y (χm |k|m Γk φk )k2L2
2


′ m (E) 2 ′ 2 m (E) 2
+ k∂y (χm+1 q )∂y (χm |k| Γk φk )kL2 + kχm+1 q ∂y (χm |k| Γk φk )kL2

(E) (E)
.a2m,1 (m + 1)2+2σ kΓk (χm |k|m φk )k2L2 + k∂v Γk (χm |k|m φk )k2L2

84
 
(E) (E)
+ (m + 1)2+2σ k∂v Γk (χm |k|m φk )k2L2 + k∂y2 Γk (χm |k|m φk )k2L2
 X X 
2 2s −2σ∗ (0,b,c) (E) 2 (0,b,c) (E) 2
.am,0 λ (m + 1) kJm,0 φk kL2 + kJm,0 φk kL2 .
b+c=2 b+c=3

An application of the commutator relation (4.7b) and the estimates yields that the T423 term is
bounded as follows
(E) (E)

T423 .a2m,1 k∂y (χm+1 q∂yy (vy−1 )(χm |k|m φk ))k2L2 + k∂y (χm+1 qvy ∂y vy−1 ∂v (χm |k|m φk ))k2L2
(E) (E)
.a2m,1 k∂y (χm+1 q∂yy (vy−1 )) χm |k|m φk k2L2 + kχm+1 q∂yy (vy−1 )∂y (χm |k|m φk )k2L2
(E) 2 (E)

+ k∂y (χm+1 qvy ∂y vy−1 ) ∂v (χm |k|m φk )k2L2 + kχm+1 qvy2 ∂y vy−1 ∂v (χm |k|m φk )k2L2
(E) (E)
.a2m,1 (m + 1)2+2σ kχm |k|m φk k2L2 + k∂v (χm |k|m φk )k2L2
(E) 2 (E)

+ (m + 1)2+2σ k∂v (χm |k|m φk )k2L2 + k∂v (χm |k|m φk )k2L2
X (0,b,c) (E)
.a2m,0 (m + 1)−2σ∗ λ2s kJm,0 φk k2L2 .
b+c=2

Combining the estimates of T421 , T422 , T423 together with the decomposition (6.45), we obtain that
(1,0,2) (E)
X (0,b,c) (E)
X (0,b,c) (E) 2
T42 . λ2s a2m−2,1 kJm−2,1 φk k2L2y + λ2s a2m−1,1 kJm−1,1 φk k22 + λ2s a2m,0 kJm,0 φk (6.46)
k2 .
b+c=3 b+c=3

Next we observe that the T43 -term is consistent with the final bound. We focus on the T44 -term.
By (6.44),
(E) (E) (E) (E)
|∂v (|k|m φk )(y = ±1)|2 . k∂y (χm |k|m φk )kL2 (k∂yy (χm |k|m φk )kL2 + k∂y (χm |k|m φk )kL2 ).

Hence,
(E) (E) (E)
T44 .a2m,1 k|k|2 ∂y (χm |k|m φk )kL2 (k|k|∂yy (χm |k|m φk )kL2 + k|k|∂y (χm |k|m φk )kL2 )
X (0,b,c) (E)
.λ2s a2m,0 kJm,0 φk k2L2 . (6.47)
b+c≤3

Combining the estimates (6.46), (6.47) we developed thus far and the decomposition (6.43), sum-
ming up in m and taking λ to be small yields the result (6.35d).

We also need the following lemma.

Lemma 6.8. Assume the bootstrap assumptions. Consider the solution to (3.2b). The following
estimates hold,

X 2
X  
(a,b,c) (E) (0) (E) 2
a20,2 kJ0,2 φk k2L2y . a2m,n kJm,n ek kL2y + kCm,n
w (I) 2
kL2y ; (6.48)
a+b+c≤2 m+n=0

X 2
X  
(a,b,c) (E) (0) (E) 2 (I)
a20,2 kJ0,2 φk k2L2y . a2m,n kJm,n ek kH 1 + kCm,n
w (I) 2
kH 1 + |vyy | y=±1
ek k2L2v
kw
k k
a+b+c=3 m+n=0

85
 !
2 X
X 2  
e˚
n (0,b,0) (I) 2
+ ϕ kJ0,n (vy2 − 1)kL2 a20,0 kw (E) k2
0,0;k L2 + kC 0,0 L2y .
k
n=0 b=1
(6.49)

(I) −1
Here the Cm,n are defined in (6.26). Here the implicit constant depends on the norms kvy kL∞ 3 , kvy kL∞ .
t Hy t,y

Proof. We organize the proof in steps.

Step # 1: Proof of (6.48). Step # 1a: a 6= 0 case. This case can only happen with n = 2
(a,b,c) (2,0,0) (E) (1,1,0) (E) (1,0,1) (E)
thanks to the definition of Jm,n . There are three sub-cases to consider, i.e., J0,2 φk , J0,2 φk , J0,2 φk .
(2,0,0) (E)
We observe that J0,2 φk is fairly direct to estimate with (6.48) and a0,2 . ϕ3 , i.e.,
 
(2,0,0) (E) (E) (E) (E) (E) (E)
a0,2 kJ0,2 φk k2 . ϕ3 kχ2 Γ2k φk k2 . ϕ3 k∂yy φk k2 + k∂y φk k2 + tk|k|∂y φk k2 + t2 k|k|2 φk k2
(E) (I)
ek kL2y + a0,0 kC0,0 kL2 .
.a0,0 kw (6.50)

Here we first apply the bound kvy−1 k∞ + kvyy k∞ ≤ C and then the bound tϕ . 1 (2.11).
(1,1,0) (E)
For the J0,2 φk term, we estimate it as follows,

(1,1,0) (E) (E)

a0,2 kJ0,2 φk k2 . ϕ3 kq −1 ∂v (χ2 q 2 Γ2k φk )k2
 
(E) (E) (E)
.ϕ3 kΓ2k φk χ2 k2 + kqΓ2k φk χ′2 k2 + kq∂v (Γ2k φk )χ2 k2 =: T11 + T12 + T13 . (6.51)

The T11 , T12 terms can be estimated as in (6.50). By invoking tϕ . 1, we end up with the result
(E) (I)
ek kL2y + kC0,0 kL2 ).
T11 + T12 . a0,0 (kw

Hence we focus on the last term T13 in (6.51). By applying (6.35a)M =0 , (6.35c)M =1 and the
commutator estimates (6.96), (6.95a), we have that
(E) (E)
T13 .ϕ3 kχ2 [q, ∂v Γk ](Γk φk )k2 + ϕ3 kχ2 ∂v Γk (χ1 qΓk φk )k2
   
.a0,0 ϕ2 k[q, ∂v ]Γk φk k2 + |k|tk[q, ∂v ]Γk φk k2 + ϕa0,1 k∂v φ˚ ˚
2 (E) (E) 2 (E) (E)
k
0,1;k 2 + tkk∂v φ k
0,1;k 2
2
X X  
2 2−ℓ (E) (E) (0) (E) (I)
.a0,0 ϕ k∂v Γk φk k2 + a0,0 ϕk|k|Γk φk k2 + am,n kJm,n ek k2 + kCm,n
w kL2y
ℓ=1 m+n≤1
X  
(0) (E) (I)
. am,n kJm,n ek k2 + kCm,n
w kL2y .
m+n≤1

To conclude, we have that

X  
(1,1,0) (E) (0) (E) (I)
a0,2 kJ0,2 φk kL2y . am,n kJm,n ek k2 + kCm,n
w kL2y .
m+n≤1

(1,0,1) (E)
We estimate the J0,2 φk -term in a similar fashion. We observe that

(1,0,1) (E) (E)

a0,2 kJ0,2 φk kL2y .ϕ3 kχ2 q|k|Γ2k φk kL2y
 
(E) (E)
.ϕ3 k|k|Γk (χ1 qΓk φk )kL2y + k|k|[q, ∂v ](χ1 Γk φk )kL2y

86
 .ϕ3 k|k|∂v φ˚ 2 ˚
(E) 2 (E) 2 (E) 2 (E)
0,1;k kL2y + ϕ k|k| φ 0,1;k kL2y + ϕ k|k|∂v φk kL2y + ϕk|k| φk kL2y
X
e˚(E) (E) (I)
.a0,1 kw ek kL2y +
0,1;k kL2y + a0,0 kw am,n kCm,n kL2y .
m+n≤1

This is consistent with (6.48) and concludes the treatment.

(0,b,c) (E)
Step # 1b: a = 0 case. Thanks to the definition, it is enough to consider the terms J0,2 φk , b+
c = 2. Recall the equation (6.24)m=0,n=2} with the Dirichlet boundary condition

(E)
χ2 q 2 Γ2k φk y=±1
= 0.

Thanks to the relation (6.36), it is enough to estimate kφ˚

(E)
0,2;k kḢk2 . The following estimate is a
direct consequence of Lemma 6.12,
3
X 5
X
e˚
(E) (I)
a0,2 kχ2 q 2 Γ2k φk kḢ 2 .ϕ3 kw (E) 3
0,2;k kL2 + ϕ kC0,2 kL2 + ϕ3 kTj,2 kL2 =: T1ℓ . (6.52)
k
j=1 ℓ=1

Here the T-terms are defined in (6.27). The T11 , T12 terms are consistent with the estimate (6.48).
Hence we focus on the rest of the terms. To begin with, we expand the T13 term with (6.27a) to
obtain the following estimate
1 
X 
2−ℓ 2 (E) 3−ℓ (E)
T13 .ϕ3 kχ2 q 2 ∂v (vy2 − 1) ∂v Γℓ φk kL2 + kχ2 q 2 ∂v (vy2 − 1) ∂v Γℓk φk kL2 .
ℓ=0

Application of the estimates tϕ ≤ C (2.11), kvy−1 kL∞ + kvy kW 2,∞ + kvy kH 3 ≤ C, (6.95a) and (6.96)
yields that
1 
X 
3 2−ℓ 2−ℓ 2 2 ℓ ℓ (E) 2 3−ℓ 2 ℓ (E)
T13 .ϕ kχ2 q ∂v (vy − 1) ∂v (q Γ φk )kL2 + kχ2 q ∂v (vy − 1) ∂v Γk φk kL2
ℓ=0
X1 
2−ℓ 2 (E) 2−ℓ 2 (E)
.ϕ3 kχ2 q 2−ℓ ∂v (vy2 − 1) ∂v (q ℓ Γℓ φk )kL2 + kχ2 q 2−ℓ ∂v (vy2 − 1) [∂v , q ℓ ]Γℓ φk kL2
ℓ=0
3−ℓ 3−ℓ

(E) (E)
+kχ2 q 2−ℓ ∂v (vy2 − 1) ∂v (q ℓ Γℓk φk )kL2 + kχ2 q 2−ℓ ∂v (vy2 − 1) [∂v , q ℓ ]Γℓk φk kL2
1
X 1
X
3−ℓ 2−ℓ 2 (E) 2−ℓ ℓ (E)
.ϕ k∂v (vy2 − 1)k L∞ ϕ ℓ
k∂v (χℓ q ℓ Γℓ φk )kL2 +ϕ 2
k∂v (vy2 − 1)kL∞ ϕk∂v Γφk kL2
ℓ=0 ℓ=0
2 (E) 2 (E)
+ ϕ k∂v (vy2
2
− 1)kL∞ ϕk∂v (χ1 qΓk φk )kL2 +ϕ 2
k∂v (vy2 − 1)kL∞ ϕkΓk φk kL2
3 (E)
+ ϕ3 k∂v (vy2 − 1)kL2 k∂v φk kL∞ .

Now we apply the Gagliardo-Nirenberg interpolation, the bound (6.35a) to obtain

1
X
(E) (I)
T13 . am,n (kw
em,n kL2y + kCm,n kL2y ).
m+n=0

For the T14 and T15 terms in (6.52), we implement similar arguments to get the result,
(E)
T14 =ϕ3 kT2,2 k2 = ϕ3 kχ2 (2(q ′ )2 + 2qq ′′ + 4qq ′ ∂y )Γ2k φk kL2

87
 (E) (E)
.ϕ3 χ2 Γ2k φk + ϕ3 χ2 qΓ2k φk
2 2
3 (E) 3 (E)
+ϕ ∂v Γk (χ1 qΓk φk ) +ϕ [q, ∂v Γk ](χ1 Γk φk )
2 2
1
X X
e˚
(E) (E)
.ϕ2 kχ1 qΓk φk kḢ 2 + ϕkφk kḢ 2 . a0,ℓ kw (E)
0,ℓ;k kL2y +
(I)
am,n kCm,n kL2y ;
k k
n=0 m+n≤1
3 3 (E)
T15 =ϕ kT3,2 k2 = ϕ k(2χ′2 ∂y + χ′′2 )q 2 Γ2k φk k2
(E)
.ϕ3 kχ′2 ∂y (qΓk (χ1 qΓk φk ))k2 + ϕ3 kχ′2 ∂y (q[q, Γk ](χ1 Γk φk ))k2
(E)
+ ϕ3 kχ′′2 (qΓk (χ1 qΓk φk ))k2 + ϕ3 kχ′′2 ([q, Γk ](χ1 qΓk φk ))k2
1
X X
e˚
(E)
.ϕ2 kχ1 qΓk φk kḢ 2 . a0,ℓ kw (E)
0,ℓ;k kL2y +
(I)
am,n kCm,n kL2y .
k
ℓ=0 m+n≤1

Combining the estimates above yields the estimate

(E)
X X
a20,2 kχ2 q 2 Γ2k φk k2Ḣ 2 . a2m,n kw (E) 2
em,n kL2y + (I) 2
a2m,n kCm,n kL2y .
k
m+n≤2 m+n≤2

This concludes the proof of (6.48).

(0,b,c) (E)
Step # 2: Proof of the estimate (6.49). We derive the estimate for the J0,2 φk terms
with b + c = 3.
First of all, thanks to (6.48), we have that

X 2
X  
(0,b,c+1) (E) 2 (0) (E) 2
a20,2 kJ0,2 φk kL2y . a2m,n k|k|Jm,n w (I) 2
ek kL2y + k|k|Cm,n kL2y . (6.53)
b+c=2 m+n=0

To obtain the remaining components of J0,2 φk , we focus on the quantity ∂y (φ˚

(3) (E) (E)
0,2;k ). Since there
is q 2 factors in the expression, the boundary condition ∂ (φ˚
(E) )(y = ±1) = 0 holds, and by the
y 0,2;k
estimate (6.84) and the equation (6.24), we obtained the following:

a0,2 k∂y (φ˚ e˚

(E) 3 (E) 3 (I) 3 3 3
0,2;k )kḢ 2 .ϕ k∂y w 0,2;k kL2 + ϕ k∂y C0,2 kL2 + ϕ k∂y T1,2 kL2 + ϕ k∂y T2,2 kL2 + ϕ k∂y T3,2 kL2
k
5
X
=: T2j . (6.54)
j=1

The T21 , T22 terms are consistent with (6.49). We estimate the T23 -term by applying (6.35) as
follows,
1  
! !
X 2 
3 2 1−ℓ 2 2−ℓ ℓ (E)
T23 . ϕ vy ∂v χ2 q 2∂v vyy ∂v + ∂v vyy ∂v Γk φk
ℓ
ℓ=0 2 L
1
X   1
X  
1−ℓ 2 (E) 2−ℓ (E)
. ϕ3 ∂v χ2 q 2 ∂v vyy ∂v (χℓ Γℓk φk ) + ϕ3 ∂v χ2 q 2 ∂v vyy ∂v (χℓ Γℓk φk )
L2 L2
ℓ=0 ℓ=0
  1
X  
3 2(E) 3 1−ℓ 2 (E)
.ϕ ∂v χ2 vyy q[q, ∂v ]Γk φk +ϕ ∂v χ2 ∂v vyy q 2−ℓ ∂v (χℓ q ℓ Γℓk φk )
L2 L2
ℓ=0

88
   1
X  
(E) 2−ℓ (E)
+ ϕ3 ∂v χ2 ∂v vyy q[q, ∂v ]Γk φk + ϕ3 ∂v χ2 ∂v vyy q 2−ℓ ∂v (χℓ q ℓ Γℓk φk )
L2 L2
ℓ=0
4
X
=: T23j . (6.55)
j=1

Here the last term with ℓ = 0 costs the most regularity on vy . We recall the definition Z = vy2 − 1,
and estimate it as follows:
 
3 3 2 (E)
T234 1ℓ=0 . ϕ3 ∂v χ2 q 2 ∂v Z ∂v φk
(E)
2
+ ϕ3 ∂v Z χ2 q 2 ∂v φk 2
.
L L

Now rewrite the first term:

 3
 2
 2
  2

∂v χ2 q 2 ∂v Z = ∂v χ2 q 2 ∂v Z + ∂v [χ2 q 2 , ∂v ]∂v Z
 
(0,2,0) 2 2
= J0,2 Z − ∂v ∂v χ2 q 2 ∂v Z + 2χ2 ∂v q q∂v Z
(0,2,0)



= J0,2 Z − ∂v (∂v χ2 q + 2χ2 ∂v q) ∂v χ1 q∂v Z + [∂v , q]∂v Z
(0,2,0) (0,2,0) (0,1,0)
= J0,2 Z − (∂v χ2 q + 2χ2 ∂v q)J0,1 Z − ∂v (∂v χ2 q + 2χ2 ∂v q)J0,1 Z
(0,1,0) (0,2,0)
− ∂v (∂v χ2 q∂v q + 2χ2 (∂v q)2 )J0,0 Z − (∂v χ2 q∂v q + 2χ2 (∂v q)2 )J0,0 Z.

Hence the term T234 1ℓ=0 can be estimated through Gagliardo Nirenberg interpolation and the
estimate (6.35a),
2 X2
! !
X X
T234 1ℓ=0 . n (0,b,0) (0,b,c) (E) (0,2,0) (E)
ϕ kJ0,n ZkL2 a0,0 kJ0,0 φk kL2 + a0,0 ϕ2 kJ0,0 φk kL2
n=0 b=1 b+c=2
!
2 X
X 2  
e˚
(0,b,0) (I) 2
. ϕn kJ0,n ZkL2 + 1 a20,0 kw (E) k2
0,0;k L2 + kC 0,0 L2y .
k
n=0 b=1

Here the implicit constant depends on kvy−1 kLt,y ∞ , kvy kL∞ 3 . Other terms in (6.55) can be esti-
t Hy
mated as follows
3
X 2
X  
T23j + 1ℓ=0 T234 . (0) (E)
am,n kJm,n (I)
ek kH 1 + kCm,n
w kH 1 + |vyy | y=±1
ω (I) kL2v .
ke
k k
j=1 m+n=0

Combining the decomposition (6.55) and the estimate we obtained, we have that T23 has a bound
which is consistent with (6.49) The estimate of the T24 -term in (6.54) is as follows,
(E)
T24 .ϕ3 k∂v (χ2 (2(q ′ )2 + 2qq ′′ + 4qq ′ ∂y )Γ2k φk )kL2
(E) (E) (E)
.ϕ3 k∂v Γ2k φk kL2 + ϕ3 kχ2 q∂v Γ2k φk kL2 + ϕ3 k∂v (qq ′′ )Γ2k φk kL2
(E) (E)
+ ϕ3 kχ2 vy qq ′ ∂v Γ2k φk kL2 + ϕ3 k∂v (vy qq ′ )Γ2k φk kL2
X (E) (E) (E)
.ϕ3 |kt|c k∂y1+b φk kL2 + ϕ3 kχ2 ∂v Γk (χ1 qΓk φk )kL2 + ϕ3 kχ2 [q, ∂v Γk ](Γk φk )kL2
b+c≤2
X X
(|k|t)c k∂yb φ˚
3 (E)
.ϕ |kt|c k∂y1+b φk kL2 + ϕ3 (E)
0,1;k kL2
b+c≤2 b+c≤1

89
 2
X
e˚
(E) (E) (I) (I)
ek kH 1 + a0,1 kw
.a0,0 kw 0,1;k kH 1 + |vyy | y=±1
ωk kL2v +
ke am,n kCm,n kH 1 .
k k k
m+n=0

The estimate of the T25 -term in (6.54) is as follows,

(E)
T25 .ϕ3 k∂y (2χ′2 ∂y + χ′′2 )q 2 Γ2k φk kL2
(E) (E)
.ϕ3 k∂v (χ′2 vy ∂v (q 2 Γk (χ1 Γk φk ))kL2 + ϕ3 k∂v (χ′′2 q 2 Γk (χ1 Γk φk ))kL2
(E) (E)
.ϕ3 k∂v (χ′2 vy ∂v Γk (χ1 qΓk φk ))kL2 + ϕ3 k∂v (χ′2 vy ∂v (q[q, ∂v ](χ1 Γk φk ))kL2
(E) (E)
+ ϕ3 k∂v (χ′′2 qΓk (χ1 qΓk φk ))kL2 + ϕ3 k∂v (χ′′2 q[q, ∂v ](χ1 Γk φk ))kL2
2
X
e˚(E) (I) (I)
.a0,1 kw 0,1;k kH 1 + |vyy | y=±1
ke
ωk kL2v + am,n kCm,n kH 1 .
k k
m+n=0

Combining the estimates above and the decomposition (6.54), we have that
2
X  
a0,2 k∂y φ˚
(E) (E) (I) (I)
0,2;k kH 2 . k
am,n k∂y w
em,n kL2y + kCm,n kH 1 + |vyy |
k y=±1
ωk kL2v
ke
m+n=0
!
2 X
X 2  
e˚
(0,b,0) (I) 2
+ ϕn kJ0,n ZkL2 a20,0 kw (E) 2
0,0;k kL2 + kC0,0 kL2y .
n=0 b=1

Combining this and (6.53) yields (6.49).

6.3.3 Higher Regularity Bounds in Gevrey Spaces

Lemma 6.9. Let M be an arbitrary natural number greater than 3. The following estimate holds
M
X X
1n≥2 a2m,n (0,b,c)
kJm,n φk k2L2
(E)

m+n=3 b+c=2
M
X M
X M
X −1 X
(E) (E)
. a2m,n kJm,n
(0)
ek k2L2 +
w a2m,n kCm,n kL2 + λ2s
(I) 2
a2m,n (a,b,c)
kJm,n φk k2L2
m+n=0 m+n=0 m+n=0 a+b+c=2
M
! M
!
X (≤1) 2 X (E) 2
2
+ B0,n 1
ϕ2n1 J0,n1 Z a2m,n2 Jm,n
(≤2)
φ
2 k
. (6.56)
L2 L2
n1 =0 m+n2 =0

(≤j)
Here am,n is defined in (2.12a), Bm,n is defined in (2.9), Jm,n fk is defined in (3.8b), and Z is the
short hand notation for Z = vy2 − 1. Here the implicit constant is independent of M and hence we
can take the limit as M approaches ∞.

Proof. Thanks to the equation (6.24) and the maximal regularity estimate (6.85), we have that,
M
X X
1n≥2 a2m,n kJm,n
(0,b,c)
φk k2L2
(E)

m+n=3 b+c=2
M
X  
. 1n≥2 a2m,n kJm,n
(0) (E) 2
w (I) 2
ek kL2 + kCm,n kL2 + kT1;m,n k2L2 + kT2;m,n k2L2 + kT3;m,n k2L2 (6.57)
.
m+n=3

90
The first two terms on the right hand side are consistent with the bound (6.56). Hence, we focus
on estimating the Ti;m,n -terms.
Step # 1: We start by considering the T1;m,n -term,
!2
X
n−1
n

n−ℓ 2

(E)

a2m,n kT1;m,n k2L2 . am,n q n−ℓ
∂v Z χm+n ∂v q ℓ Γℓk |k|m φk
ℓ 2
ℓ=0
!2
X
n−1
n

n−ℓ
h
2
i
(E)

n−ℓ ℓ
+ am,n q ∂v Z χm+n ∂v , q Γℓk |k|m φk
ℓ 2
ℓ=0
!2
X
n−1
n

n−ℓ+1

(E)

n−ℓ
+ am,n q ∂v Z χm+n ∂v q ℓ Γℓk |k|m φk
ℓ 2
ℓ=0
!2
X
n−1
n

n−ℓ+1
h i
(E)
 4
X
n−ℓ ℓ
+ am,n q ∂v Z χm+n ∂v , q Γℓk |k|m φk =: (T1;i )2 .
ℓ 2
ℓ=0 i=1
(6.58)

Before diving into details, we make the following observation concerning the combinatoric num-
bers. Recalling the definitions (2.9), (2.12a), we have the following relation

λ(m+n)s ϕ(m+n)  s
am,n ((m+n)!)s n−ℓ (m + n − (m + ℓ))!(m + ℓ)!
= =ϕ
B0,n−ℓ am,ℓ λ(n−ℓ)s λ(t)(m+ℓ)s ϕ(m+ℓ) (m + n)!
((n−ℓ)!)s × ((m+ℓ)!) s
 −s  −s
n−ℓ m + n n−ℓ n
=ϕ ≤ϕ . (6.59)
m+ℓ ℓ

Now we apply the Hölder inequality and the estimates of vy−1 , vyy to obtain that

X
n−1
n

am,n n−ℓ (0,2,0) (E)
T1;1 . B0,n−ℓ χn−ℓ q n−ℓ ∂v Z ∞ am,ℓ kJm,ℓ φk kL2
ℓ B0,n−ℓ am,ℓ L
ℓ=0
n−1  
X m + n 1−s n−ℓ (0,2,0) (E)
. ϕn−ℓ B0,n−ℓ χn−ℓ q n−ℓ ∂v Z 1 am,ℓ kJm,ℓ φk kL2 .
m+ℓ Ḣ
ℓ=0

Now we have that by the relation (2.1) and the combinatorial fact (6.88),
! n−1 
X n−2σ−2σ∗
n−1 XX (0,b,0) 2 (0,2,0) (E)
(T1;1 )2 .  ϕ2n−2ℓ B0,n−ℓ
2
J0,n−ℓ Z 2 a2m,ℓ kJm,ℓ φk k2L2 
ℓ L
ℓ=0 ℓ=0 b≤1
n−1
XX (0,b,0) 2 (0,2,0) (E) 2
. ϕ2n−2ℓ B0,n−ℓ
2
J0,n−ℓ Z a2m,ℓ kJm,ℓ φk kL2 . (6.60)
L2
ℓ=0 b≤1

For the second term T1;2 , we invoke the commutator relation (6.100a) to obtain

n−1  
!2
X n h i 
2 n−ℓ n−ℓ 2 ℓ (E)
(T1;2 ) . am,n χn−ℓ q ∂v Z χm+n ∂v , q χm+ℓ Γℓk |k|m φk
ℓ L2
ℓ=1

91
  2
n−1
X 1  
X 2
X X
n (0,b1 ,0) (a ,b ,0) (E)
. am,n J0,n−ℓ Z 2 2
kJm,(ℓ−i φ kL2  .
2 )+ k
ℓ L2
ℓ=1 b1 =0 i2 =0 a2 +b2 ≤2

Now we are in a situation which is almost identical to the term T1;1 , so we apply the same argument
as before to obtain that
n−1
X X 2
X X
(0,b ,0) 2 (a ,b ,0) (E)
(T1;2 )2 . 2
B0,n−ℓ ϕ2n−2ℓ J0,n−ℓ
1
Z a2m,(ℓ−i2 )+ kJm,(ℓ−i
2 2
φ k2L2 .(6.61)
2 )+ k
L2
ℓ=1 b1 ≤1 i2 =0 a2 +b2 ≤2

The T1;3 -term can be estimated as follows with the help of (6.59) and the bound (6.112),

X
n−1
n

n−ℓ+1

(E)

T1;3 . am,n χm+n q n−ℓ ∂v Z ∂v χm+ℓ q ℓ Γℓk |k|m φk
ℓ L2 L∞
ℓ=0
X
n−1
m+n
−σ−σ∗
n−ℓ+1 X (0,b ,0) (E)
. B0,n−ℓ ϕn−ℓ χn−ℓ q n−ℓ ∂v Z am,ℓ kJm,ℓ 2 φk kL2
m+ℓ L2
ℓ=0 b2 ≤2

X
n−1  1
m + n −σ X (≤1) (0,2,0) (E)
. B0,(n−ℓ−i1 )+ ϕ(n−ℓ−i1 )+ J0,(n−ℓ−i1 )+ Z am,ℓ kJm,ℓ φk kL2 .
m+ℓ L2 (suppχ
e1 )
ℓ=0 i1 =0

By Lemma 6.20, the T1;3 term can be estimated as follows:

n−1
X 1
X
2 (≤1) (0,2,0) (E)
(T1;3 ) . B0,(n−ℓ−i1 )+ ϕ(n−ℓ−i1 )+ J0,(n−ℓ−i1 )+ Z am,ℓ kJm,ℓ φk kL2 .(6.62)
L2 (suppχ
e1 )
ℓ=0 i1 =0

For the T1;4 -term in (6.58), we estimate it using similar method as before,

X
n−1 
m + n −σ n−ℓ+1 (E)
T1;4 . B0,n−ℓ ϕn−ℓ χn−ℓ q n−ℓ ∂v Z 2 am,ℓ [∂v , q ℓ ](χm+ℓ |k|m Γℓk φk ) ∞
m+ℓ L L
ℓ=1
X m + n−σ
n−1 X1
n−ℓ n−ℓ n−ℓ+1 b2 (E)
. B0,n−ℓ ϕ χn−ℓ q ∂v Z 2 am,ℓ ∂v [∂v , q ℓ ](χm+ℓ |k|m Γℓk φk ) .
m+ℓ L L2
ℓ=1 b2 =0

Application of the bounds (6.100b), and (6.112) yields that

n−1
X 2
X 2
X
(≤1) 2 (≤2) (E)
(T1;4 )2 . 2
B0,(n−ℓ−i 1 )+
ϕ2(n−ℓ−i1 )+ J0,(n−ℓ−i1 )+ Z a2m,(ℓ−i2 )+ kJm,(ℓ−i2 )+ φk k(6.63)
2
L2 .
L2
ℓ=1 i1 =0 i2 =0

Combining the decomposition (6.57), (6.58) and the estimates (6.60), (6.61), (6.62), (6.63) yields
that
M
X
1n≥2 a2m,nkT1;m,n k2L2
m+n=3
2
2 X M M −m n−1
!
X X X X (≤1) 2 (≤2) (E)
2
. B0,(n−ℓ−i 1 )+
ϕ2(n−ℓ−i1 )+ J0,(n−ℓ−i1 )+ Z 2 a2m,(ℓ−i2 )+ kJm,(ℓ−i2 )+ φk k2L2
L
i1 =0 i2 =0 m=0 n=0 ℓ=0
2 X
2 M M −m M −m
!
X X X (≤2) (E)
X (≤1) 2
. a2m,(ℓ−i2 )+ kJm,(ℓ−i2 )+ φk k2L2 2
B0,(n−ℓ−i 1 )+
ϕ2(n−ℓ−i1 )+ J0,(n−ℓ−i1 )+ Z
L2
i1 =0 i2 =0 m=0 ℓ=0 n=ℓ

92
 M
! M
!
X (≤1) 2 X 2
2 (≤2) (E)
. B0,n 1
ϕ2n1 J0,n1 Z 2 a2m,n2 Jm,n φ
2 k
. (6.64)
L L2
n1 =0 m+n2 =0

This is consistent with (6.56) and concludes Step # 1.

(E)
Step # 2: We can estimate the T2;m,n (= χm+n (n(n−1)q n−2 (q ′ )2 +nq n−1 q ′′ +2nq n−1 q ′ ∂y )|k|m Γnk φk )
in (6.24) as follows

n2 (E)
a2m,n kT2;m,n k2L2 . am,n χm+n |k|m q n−1 Γk (χm+n−1 Γkn−1 φk )
q L2
(E)
+ am,n n χm+n q n−1 Γk (χm+n−1 |k|m Γkn−1 φk )
L2
2
(E)
+ am,n n χm+n q n−1
(∂v Γk )(χm+n−1 |k|m Γkn−1 φk )
L2

am,n−1 m+n (E)
.λ2s σ+σ
ϕ Γk (χm+n−1 |k|m q n−1 Γkn−1 φk )
(m + n) ∗ q L2
am,n−1 (m + n)2 (E)
+ σ+σ
(χm+n−1 |k|m q n−1 Γkn−1 φk )
(m + n) ∗ q2 L2
am,n−1 (E)
+ ϕ χm+n Γk (χm+n−1 |k|m q n−1 Γkn−1 φk ) 2
(m + n)σ+σ∗ L
am,n−1 n−1 m n−1 (E)
+ ϕ χm+n [q , Γk ](χm+n−1 |k| Γk φk ) 2
(m + n)σ+σ∗ L
am,n−1 (E)
+ ϕ χm+n [q n−1 , ∂v Γk ](χm+n−1 |k|m Γkn−1 φk ) 2
(m + n)σ+σ∗ L
2
am,n−1 (E)
+ ϕ χm+n (∂v Γk )(χm+n−1 |k|m q n−1 Γkn−1 φk )
(m + n)σ+σ∗ L2
λ2s a2m,n−1 X (a,b,c) (E)
. 2σ+2σ
kJm,n−1 φk k2L2 .
(m + n) ∗
a+b+c=2

P
Hence the sum M m+n=3 1n≥2 am,n kT2;m,n kL2 is bounded by the right hand side of (6.56). Hence,
2 2

the Step # 2 is concluded.

Step # 3: The estimate of the T3;m,n term is similar to the estimate before. Here we recall the
properties of the cut-off function

k∂y χm+n k∞ ≤ C(m + n)1+σ , k∂yy χm+n k∞ ≤ C(m + n)2(1+σ) .

Similar to the estimate of T1;m,n , the main problem in estimating the χm+n commutator terms
that results in the loss of (m + n)2 powers. As a result, we have to open up two Γk -derivatives to
extract regularity information from the m + n − 2 level. We estimate the T3;m,n as follows
(E)
a2m,n kT3;m,n k2L2 ≤a2m,n k(2χ′m+n ∂y + χ′′m+n )q n Γnk |k|m φk k2L2

 2
(E)
.a2m,n (m + n)2+2σ ∂v q n ∂v + ikt (χm+n−1 |k|m Γkn−1 φk )
2

2 (E) 2
+ a2m,n (m + n)4(1+σ) q n ∂v + ikt (χm+n−2 |k|m Γkn−2 φk )
2
=:T3;1 + T3;2 . (6.65)

93
 (E) (E)
We first treat the term T3;1 . We recall the notation φm,n = χm+n |k|m q n Γnk φk . Applying the
combinatoric bound (4.29) and the commutator relations (6.100), yields that
λ2s a2m,n−1 2  
n m n−1 (E)
 
n m n−1 (E)
 2
T3;1 . ϕ ∂ v q ∂ (χ
v m+n−1 |k| Γ k φk ) + |k|t ∂v q χ m+n−1 |k| Γ k φk
(m + n)2σ⋆ 2 2
2s 2    
λ am,n−1 (E)
. ∂v q∂v χm+n−1 |k|m q n−1 Γkn−1 φk
(m + n)2σ⋆ 2
  
(E)
+ ∂v q[∂v , q n−1 ] χm+n−1 |k|m Γkn−1 φk
2
    2
m n−1 n−1 (E) m n−1 n−1 (E)
+ q|k|∂v χm+n−1 |k| q Γk φk + |k| χm+n−1 |k| q Γk φk
2 2

λ2s a2m,n−1 2 2 2
∂v φ˚ + ∂v φ˚ + (n − 1)∂v φ˚
2 (E) (E) (E)
. m,n−1;k m,n−1;k m,n−1;k
(m + n)2σ⋆ 2 2 2
2 2
n−1 ˚ 2 m+n−1
+ φ(E) m,n−1;k + |k|∂v φ˚
(E)
m,n−1;k + |k|φ˚
(E)
m,n−1;k
q 2 2 q 2
λ2s a2m,n−1 X (a,b,c) (E) 2
. Jm,n−1 φk .
(m + n)2σ⋆ 2
a+b+c=2

Next we treat the term T3;2 in (6.65). We further decompose the term as follows,
λ4s a2m,n−2 4 n  2 2 2

n−2 m (E)
2
T3;2 . 4σ
ϕ q ∂ v + 2ikt∂ v − k t (χ m+n−2 Γ k |k| φk )
(m + n) ⋆ 2
4s 2 
λ am,n−2 n 2 m n−2 (E)
2 (E) 2
. 4σ
q ∂v (χ m+n−2 |k| Γ k φk ) + |k|q n ∂v (χm+n−2 |k|m Γkn−2 φk )
(m + n) ⋆ 2 2

2
n 2 n−2 m (E)
+ χm+n−2 q |k| Γk |k| φk
2

=:T3;21 + T3;22 + T3;23 .

For the T3;21 term, we recall the commutator relations (6.100), (6.101a) and estimate it as follows:
 h i 
λ4s a2m,n−2 2 2 n−2 n−2 m (E)
2
2 2 n−2 n−2 m (E)
2
T3;21 . q ∂v (χm+n−2 q Γk |k| φk ) + q ∂v , q (χm+n−2 Γk |k| φk )
(m + n)4σ⋆ 2 2
4s 2
λ am,n−2 X (a,b,0) (E)
. 4σ
kJm,n−2 φk k22 .
(m + n) ⋆
a+b=2

This is consistent with the result. Next we estimate the T3;22 -term, we recall the commutator
relations (6.100), (6.101a) to estimate the term as follows
 
λ4s a2m,n−2 2 ˚ 2
2 n−2 m n−2 (E)
2
T3;22 . (E)
|k|q ∂v φ m,n−2;k + |k|q [∂v , q ](χm+n−2 |k| Γk φk )
(m + n)4σ⋆ 2 2
4s 2
λ am,n−2 X (a,b,c) (E)
. 4σ
kJm,n−2 φk k22 .
(m + n) ⋆
a+b+c=2

Finally, we estimate the T3;23 term,

λ4s a2m,n−2 (0,0,2) (E)
T3;23 . kJ φ k2 .
(m + n)4σ⋆ m,n−2 k

94
 P
Hence the sum M m+n=3 1n≥2 am,n kT3;m,n kL2 is bounded by the right hand side of (6.56). Hence,
2 2

the Step # 3 is concluded. Combining the estimates yields the result (6.56).

P 
(a,b,c) (E)
To derive the Sobolev H 1 -based Gevrey estimates a+b+c=3 Jm,n φk , we plan to invoke the
elliptic maximal regularity type estimate and the estimates of the terms ∂y T1;m,n , ∂y T2;m,n , ∂y T3;m,n .
We summarize the estimates of the ∂y T-terms in the following lemma.

Lemma 6.10. Assume that σ∗ ≥ 8σ, m + n ≥ 3, n ≥ 2. Then the following estimates hold
M 1 M
! M
!
X X X 2 X 2
1n≥2 a2m,n k∂y T1;m,n k2L2 . 2
B0,n 1
ϕ2n1
J
(≤1+i)
0,n1 Z 2
(≤3−i) (E)
a2m,n2 Jm,n 2
φk ;
L L2
m+n=3 i=0 n1 =0 m+n2 =0
(6.66a)

M
X M
X −1
(E) 2
1n≥2 a2m,n k∂y T2;m,nk2L2 .λ2s a2m,n Jm,n
(≤3)
φk ; (6.66b)
L2
m+n=3 m+n=2
M
X M
X −1
(E) 2
1n≥2 a2m,n k∂y T3;m,nk2L2 .λ2s a2m,n Jm,n
(≤3)
φk . (6.66c)
L2
m+n=3 m+n=1

(≤ζ) (≤ζ) (a,b,c) (a,b,0)

Here, we use the notation Jm,n fk (or J0,n f0 ) to denote the vectors {Jm,n fk }a+b+c≤ζ (or {J0,n f0 }a+b≤ζ ),
and measure them with the classical L2 -norm of vectors:
2 X 2 (≤ζ) 2 X (a,b,0) 2
(≤ζ) (a,b,c)
Jm,n fk := Jm,n fk , J0,n f0 := J0,n f0 .
L2 L 2 L2 L2
a+b+c≤ζ a+b≤ζ

Proof. We divide the proof into three steps.

Step # 1: Proof of (6.66a). By recalling the definition (6.27a) and Z = vy2 − 1, we have that

1   !2
n−1
XX n  
n−ℓ+i 2−i (E)
a2m,n k∂y T1;m,n k2L2 . a2m,n ∂y (χm+n q n )∂v Z ∂v Γℓk |k|m φk
ℓ L2
ℓ=0 i=0
2  
!2
n−1 X
X n n−ℓ+i 3−i
  2
X
(E)
+a2m,n χm+n q ∂v n
Z ∂v Γℓk |k|m φk =: T1j2 .
(6.67)
ℓ L2
ℓ=0 i=0 j=1

We decompose the proof into substeps. In each step, we estimate one term in (6.67).
Step # 1a: Estimation of the T11 2 term. We start by considering the T1 -term in (6.67). To
this end, we invoke the assumption (2.5a) to obtain that for m + n ≥ 3, n ≥ 2

λs X n
1 n−1
X 
n−ℓ

2−i
 
T11 . a m,n−1 ϕk1 suppχ m+n q n−1 i
∂v ∂v Z ∂v Γ ℓ
k |k|m (E)
φk kL2 .
(m + n)σ∗ ℓ
i=0 ℓ=0

We distinguish between the ℓ = 0 and ℓ ≥ 1 cases as follows

λs
1
X X n 
n−1
n−ℓ 2−i
 
n−ℓ i ℓ−1 ℓ−1 m (E)
T11 . a m,n−1 ϕ q ∂v (χ n−ℓ ∂v Z) q Γ k ∂v χ m+ℓ−1 Γ k |k| φk
(m + n)σ∗ ℓ L2
i=0 ℓ=⌈n/2⌉

95
 λs X X n
1 ⌊n/2⌋
n−ℓ−1
 
n−ℓ−1 i+1 ℓ 2−i ℓ m (E)
+ a m,n−1 ϕ χ n−ℓ−1 q ∂v (∂v Z) q ∂v χ m+ℓ Γ k |k| φk
(m + n)σ∗ ℓ L2
i=0 ℓ=1

λs X1
am,n−1 ϕ  
i+1 n−1 2−i (E)
+ B0,n−1 kq n−1 ∂v (χn−1 ∂v Z)kL2 am,0 ∂v χm |k|m φk
(m + n)σ∗ B0,n−1 am,0 L∞
i=0
3
X
=: T11j . (6.68)
j=1

To estimate the first term, we observe that

       
n n(n − 1)! n−1 n n−1 m+n−1
= = ≤2 ≤2 , ℓ ≥ ⌈n/2⌉. (6.69)
ℓ ℓ(n − ℓ)!(ℓ − 1)! ℓ−1 ℓ ℓ−1 m+ℓ−1

For the first term T11j in (6.68), we apply the commutator relations (6.100a), (6.100b), the relation
(2.1), and the relation (6.69) to estimate it as follows

λs X1 X m + n − 1−σ−σ∗
n−1
i n−ℓ
T111 . ϕn−ℓ B0,n−ℓ kq n−ℓ ∂v (χn−ℓ ∂v Z)kHy1
(m + n)σ∗ m+ℓ−1
i=0 ℓ=⌈n/2⌉
 
ℓ−1 3−i m ℓ−1 (E) ℓ−1 2−i m ℓ−1 (E)
× am,ℓ−1 kq ∂v (χm+ℓ−1 |k| Γk φk )kL2 + |k|kq ∂v (χm+ℓ−1 |k| Γk φk )kL2
!
λs X1 X m + n − 1−σ−σ∗ X
n−1 1
(≤1+i)
n−ℓ−i1
. ϕ B0,n−ℓ−i1 kJ0,n−ℓ−i1 ZkL2
(m + n)σ∗ m+ℓ−1
i=0 ℓ=⌈n/2⌉ i 1 =0
2
!
X (≤3−i) (E)
× am,ℓ−i2 kJm,ℓ−i2 φk kL2 .
i2 =0

Hence, we can invoke the estimate (6.88) to obtain that T111 is bounded as follows
1 n−1
X X X 1 2
X
(≤1+i) 2 (≤3−i) (E) 2
2 2s 2(n−ℓ−i1 )+ 2
T111 .λ ϕ B0,(n−ℓ−i 1 )+
J0,(n−ℓ−i1 )+ Z 2 am,(ℓ−i2 )+ Jm,(ℓ−i2 )+ φk .
L L2
i=0 ℓ=0 i1 =0 i2 =0

For the second term T112 in (6.68), we recall the relation (6.69) and obtain that in this parameter
regime,
       
n n−1 n n−1 m+n−1
= ≤2 ≤2 , ℓ ≤ ⌊n/2⌋.
ℓ ℓ n−ℓ ℓ m+ℓ
Hence,
   
n am,n−1 (m + n − 1)! (n − ℓ − 1)!s (m + ℓ)!s n−ℓ−1 m + n − 1 −σ−σ∗ n−ℓ−1
≤2 ϕ =2 ϕ .
ℓ Bn−ℓ−1 am,ℓ (n − ℓ − 1)!(m + ℓ)! (m + n − 1)!s m+ℓ

Now we estimate the T112 -term with (6.100), (6.101d) and (6.88), as follows,

1  ⌊n/2⌋
X X m + n − 1−σ∗
2 2s i+1 n−ℓ−1
T112 .λ B0,n−ℓ−1 ϕn−ℓ−1 q n−ℓ−1 ∂v (χn−ℓ−1 ∂v Z)
m+ℓ L2
i=0 ℓ=1

96
   2
ℓ 2−i (E)
× am,ℓ q ∂v χm+ℓ Γℓk |k|m φk
H1
1 ⌊n/2⌋ 1
! 2
!
X X X (≤1+i)
X (≤3−i) (E)
.λ2s ϕ2n−2ℓ−2−2i1 B0,n−ℓ−1−i
2
1
kJ0,n−ℓ−1−i1 Zk2L2 a2m,ℓ−i2 kJm,ℓ−i2 φk k2L2 .
i=0 ℓ=1 i1 =0 i2 =0

Finally, we estimate the T113 term in (6.68) with (6.100a),

X 1
2 λ2s 2 1+i n−1 2−i (E) 2
T113 . σ
B0,n−1 ϕ2n−2 kq n−1 ∂v (χn−1 ∂v Z)k2L2 a0,0 ∂v φk
(m + n) ∗ H1
i=0
X1 2
X
λ2s 2 (≤1+i) (≤3−i) (E) 2
. B0,(n−1−i 1 )+
ϕ2(n−1−i1 )+ kJ0,n−1−i1 Zk2L2 a0,0 J0,0 φk .
(m + n)σ∗ L2
i=0 i1 =0

This is concludes the Step # 1a.

Step # 1b: Estimation of the T22 term. For the T22 term in (6.67), we estimate it with the
relation (6.59) as follows:
!2
2
X X m + n−σ
n−1
n−ℓ+i 3−i (E)
2 n−ℓ ℓ
T12 . χm+n q ∂v Z q ∂v (χm+ℓ Γℓk |k|m φk )
m+ℓ L2 L∞
i=1 ℓ=0
!2
X
n−1
m+n
−σ
n−ℓ 3 (E)
n−ℓ ℓ
+ χm+n q ∂v Z q ∂v (χm+ℓ Γℓk |k|m φk ) .
m+ℓ L∞ L2
ℓ=0

Now we invoke the combinatorial fact (6.88), the commutator relations (6.100), (6.101d), (6.101),
and Lemma 6.20,
2 n−1
X X n−ℓ+i 2
2 2
T12 . B0,n−ℓ ϕ2n−2ℓ χn−ℓ q n−ℓ ∂v Z
L2
i=1 ℓ=0
   
(0,3−i,0) (E) 3−i (E)
× a2m,ℓ Jm,ℓ φk + [q ℓ , ∂v ] χm+ℓ Γℓk |k|m φk
H1 H1
n−1
X 2    
2 (≤1) (0,3,0) (E) 3 (E)
+ B0,n−ℓ ϕ2n−2ℓ J0,n−ℓ Z a2
2 m,ℓ
Jm,ℓ φk + [q ℓ , ∂v ] χm+ℓ Γℓk |k|m φk
L H1 L2
ℓ=0
1 n−1
X X 2
X 2
X
(≤1+i) 2 (≤3−i) (E)
2
. B0,(n−ℓ−i 1 )+
ϕ2(n−ℓ−i1 )+ J0,n−ℓ−i1 Z a2m,(ℓ−i2 )+ kJm,ℓ−i2 φk k2L2 .
L2
i=0 ℓ=0 i1 =0 i2 =0

Now summing all the estimates in Step #1, and applying a similar argument as in (6.64) yield
(6.66a).
Step # 2: Proof of (6.66b) We can estimate the T2;m,n in (6.24) as follows 1

(E)
am,n k∂y T2;m,n kL2 .am,n χ′m+n (n(n − 1)q n−2 (q ′ )2 + nq n−1 q ′′ + 2nq n−1 q ′ ∂y )|k|m Γnk φk
L2
 2 
n m n n (E)
+ am,n χm+n ∂y |k| q Γk φk
q2 L2
1

97
  
(E)
+ am,n n χm+n ∂y q n−1 Γk (χm+n−1 |k|m Γkn−1 φk )
L2
 
(E)
+ am,n n χm+n ∂y q n−1 (∂v Γk )(χm+n−1 |k|m Γkn−1 φk )
L2
4
X
=: T2ℓ . (6.70)
ℓ=1

We estimate the first term in (6.70) using the estimate (2.5a) and the commutator estimates (6.100),
(6.101a) as follows

T21 .am,n (m + n)1+σ

    
(E) m n−1 (E)
× n2 q −1 ∂v χm+n−1 |k|m q n−1 Γkn−1 φk + n 2 −1 n−1
q [q , ∂v ] χ m+n−1 |k| Γ k φk
L2 L2
   
(E) (E)
+ n∂v χm+n−1 |k|m q n−1 Γkn−1 φk + n[q n−1 , ∂v ] χm+n−1 |k|m Γkn−1 φk
L2 L2
   
2 (E) (E)
+ n∂v χm+n−1 |k|m q n−1 Γkn−1 φk + t n∂v |k| χm+n−1 |k|m q n−1 Γkn−1 φk
L2 L2
  
2 (E)
+ n[q n−1 , ∂v + ikt∂v ] χm+n−1 |k|m Γkn−1 φk 2 L
s −σ∗ (≤3)
.λ am,n−1 (m + n) kJm,n−1 φ(E) kL2 .

The estimate of the other three terms are similar,

  2    
n m n−1 n−1 (E) n2 n−1 m n−1 (E)
T22 .am,n ∂v Γk (χm+n−1 |k| q Γk φk ) + ∂v [q , ∂v ](χm+n−1 |k| Γk φk )
q L2 q L2
X (a,b,0) (E)
.λs am,n−1 (m + n)−σ−σ∗ kJm,n−1 φk kL2 ;
a+b=3
     
(E) (E)
T23 .am,n n ∂v Γk (χm+n−1 |k|m q n−1 Γkn−1 φk ) + ∂v [q n−1 , ∂v ](χm+n−1 |k|m Γkn−1 φk )
L2 L2
s −σ−σ∗ (3) (E)
.λ am,n−1 (m + n) kJm,n−1 φk kL2 ;
     
(E) (E)
T24 .am,n n ∂v (∂v Γk )(χm+n−1 |k|m q n−1 Γkn−1 φk ) + ∂v [q n−1 , ∂v Γk ](χm+n−1 |k|m Γkn−1 φk )
L2 L2
s −σ−σ∗ (3) (E)
.λ am,n−1 (m + n) kJm,n−1 φk kL2 .

Now summing all the estimates in Step # 2 yields (6.66b).

Step # 3: Proof of (6.66c) We estimate the T3;m,n as follows
(E)
am,n k∂y T3;m,n kL2 ≤am,n k∂y (2χ′m+n ∂y + χ′′m+n )q n Γnk |k|m φk kL2

(E)
.am,n kχ′′m+n ∂y (q n (∂v + ikt)χm+n−1 |k|m Γkn−1 φk )kL2
(E)
+ kχ′m+n ∂yy (q n (∂v + ikt)χm+n−1 |k|m Γkn−1 φk )kL2
m n−2 (E)
+ kχ′′′ n 2
m+n q (∂v + ikt) χm+n−2 |k| Γk φk kL2

(E)
+ kχ′′m+n ∂y (q n (∂v + ikt) 2
χm+n−2 |k|m Γkn−2 φk )kL2
4
X
=: T3ℓ . (6.71)
ℓ=1

98
Now we estimate the T31 term in the expression (6.71)
 
T31 .am,n (m + n)2(1+σ) k∂y (qΓk φ˚ (E) )k
m,n−1;k L 2 + k∂y (q[q n−1
, ∂ v ](χ m+n−1 |k|m n−1 (E)
Γ k φk ))kL 2

m+n 2 ˚ m+n
.λs am,n−1 (m + n)−σ∗ +σ ∂v φ(E) m,n−1;k + ∂v |k|φ˚(E)
m,n−1;k
q L2 q L2
 2  2 
m+n m+n
+ ∂v φ˚
(E)
m,n−1;k + φ˚ (E)
m,n−1;k
q 2
q 2
L L
s −2σ (3) (E)
.λ am,n−1 (m + n) kJm,n−1 φk kL2 .

Here in the last line, we have used that σ∗ ≥ 4σ.

Now we estimate the T32 term in the expression (6.71)
 
T32 .am,n (m + n)1+σ k∂yy (qΓk φ˚ (E) )k
m,n−1;k L 2 + k∂yy (q[q n−1
, ∂v ](χ m+n−1 |k|m n−1 (E)
Γ k φk ))kL 2

(3) (E)
.am,n−1 (m + n)−σ∗ kJm,n−1 φk kL2 .

Next we estimate T33 term in the expression (6.71) with the commutator relations (6.100),
(6.101a) as follows
 
T33 .am,n (m + n)3(1+σ) kq 2 Γ2k φ˚
(E) 2 n−2 2 m n−2 (E)
k
m,n−2;k L 2 + kq [q , ∂v + 2ikt∂v ](χ m+n−2 |k| Γ k φk )kL 2

m+n 2 ˚ m+n m+n 2 ˚
.λ2s am,n−2 (m + n)−2σ∗ +σ ∂v φ(E) m,n−2;k + ∂v |k|φ˚
(E)
m,n−2;k + |k| φ(E) m,n−2;k
q L 2 q L 2 q
  
X m+n a b c ˚
+ ∂v |k| φ(E) m,n−2;k
q L2
a+b+c=3,a≥1
(3) (E)
.λ am,n−2 (m + n)−2σ kJm,n−2 φk kL2 .
2s

The estimate of the T34 term is similar to the estimate of the T33 term in (6.71). We invoke the
relations (6.100), (6.101c) to obtain that
 
T34 .am,n (m + n)2(1+σ) k∂y (q 2 Γ2k φ˚
2 (E)
(E) 2 n−2
m,n−2;k )kL2 + k∂y (q [q , ∂v + 2ikt∂v ](χm+n−2 |k|m Γkn−2 φk ))kL2

∂v (q 2 ∂v φ˚ ˚ 2 ˚
2 (E)
.λ2s am,n−2 (m + n)−2σ∗ 2
m,n−2;k ) 2 + ∂v (q ∂v |k|φ
(E) 2
m,n−2;k ) 2 + ∂v (q |k| φ
(E)
m,n−2;k )
L L L2
  
X m+n a b c ˚ (E)
+ ∂v |k| φ m,n−2;k
q L2
a+b+c=3,a≥1
(3) (E)
.λ2s am,n−2 (m + n)−2σ∗ kJm,n−2 φk kL2 .

Now summing all the estimates in Step # 3 yields (6.66c).

As a consequence of the above lemma and the elliptic maximal estimate, we obtained the lemma:

Lemma 6.11. Assume m + n ≥ 3, n ≥ 2. Then,

M
X X
1n≥2 a2m,n (0,b,c)
kJm,n
(E)
φk k2L2
m+n=3 b+c=3

99
 M
X X (E)
. a2m,n kJm,n
(0,b,c)
ek k2L2
w
m+n=0 b+c=1
1 M
! M
!
X X (≤1+i) 2 X 2
2 (≤3−i) (E)
+ B0,n ϕ2n J0,n Z 2 a2m,n Jm,n φk
L (suppχ
e1 ) L2
i=0 n=0 m+n=0
M
X −1 M
X
(E) (I) (≤2)
+ λ2s a2m,n kJm,n
(3)
ek k2H 1
φk k2L2 + kw 2
B0,n ϕ2n+2 kJ0,n Zk2L2 (suppχe1 ) . (6.72)
k
m+n=0 n=0

Here the am,n is defined in (2.12a) and Bm,n is defined in (2.9).

Proof. The proof is similar to the proof of Lemma 6.9. Since n ≥ 2, we consider the equation (6.24)
and derive the following two equations
(E) (E) 2 (I)
ek ) + ∂y (χm+n |k|m q n Γnk (e
∆k ∂y (χm+n |k|m q n Γnk φk ) =∂y (χm+n |k|m q n Γnk w χ1 (∂v − ∂y2 )φk ))
+ ∂y T1;m,n + ∂y T2;m,n + ∂y T3;m,n ,
(E)
∂y (χm+n |k|m q n Γnk φk ) y=±1
= 0. (6.73)

We invoke the maximal regularity estimate (6.84) on the equation (6.73) to obtain that
M
X X (E)
a2m,n k∂yb+1 |k|c Jm,n
(0)
φk k2L2
m+n=3 b+c=2
M
X  
(0) (E) 2
. a2m,n k∂y Jm,n w (I) 2
ek kL2 + k∂y T1;m,n k2L2 + k∂y T2;m,n k2L2 + k∂y T3;m,n k2L2 + k∂y Cm,n k(6.74)
L2 .
m+n=3

The estimate of the second to fourth terms are summarized in (6.66a), (6.66b) and (6.66c). For
the last term, we invoke the estimate (6.29). In conclusion, the left hand side of (6.74) is consistent
with (6.72).
(I)
Next, from J (2) -estimate (6.56), and the bound of Cm,n (6.28), we have that
M
X X
1n≥2 (0,b,c+1)
kJm,n
(E)
φk k2L2
m+n=3 b+c=2
M
X M
X −1 X
(E) (E)
. a2m,n k|k|Jm,n
(0)
ek k2L2 + λ2s
w a2m,n (a,b,c+1)
kJm,n φk k2L2
m+n=0 m+n=0 a+b+c=2
M
! M
!
X (≤1) 2 X 2
2 (≤3) (E)
+ B0,n 1
ϕ2n1 J0,n1 Z a2m,n2 Jm,n φ
2 k
L2 L2
n1 =0 m+n2 =0
M
X X
ϕ2n+2 k1suppχe1 J0,n
(I) (a,b,0)
ek k2L2
+ k|k|w 2
B0,n Zk2L2 .
n=0 a+b≤1

Combining all the estimates developed thus far, we obtained (6.72).

6.3.4 Conclusion
Proof of Proposition 6.2 b). Recalling the induction relation (6.108b), the low-n regularity esti-
(I)
mates (6.35a), (6.35c), (6.48), the high-n regularity estimate (6.56) and the Cm,n -estimate (6.28),

100
we obtain that
M
X (E)
a2m,n kJm,n
(≤2)
φk k2L2
m+n=0
M M
! M
!
X X (≤1) 2 X 2
(≤2) (E)
. a2m,n kw (E) 2
em,n kL2 + 2
B0,n 1
ϕ2n1 J0,n1 Z a2m,n2 Jm,n φ
2 k
L2 (suppχ
e1 ) L2
m+n=0 n1 =0 m+n2 =0
M
X −1 M
X
(E) (≤1) (I)
+ λ2s a2m,n kJm,n
(2)
φk k2L2 + Bn2 ϕ2n kJ0,n Zk2L2 (suppχe1 ) kw
ek k2L2 .
m+n=0 n=0

Now we apply the assumption that

M
X
2 (≤1)
B0,n ϕ2n kJ0,n Zk2L2 (suppχe1 ) . exp{−ν −1/6 }, (6.75)
n=0

and take the parameters λ, ν small compared to universal constants to guarantee (6.10).
To prove the estimate (6.11), the strategy is similar. We invoke the induction relation (6.108b),
the low-n regularity estimates (6.35b), (6.35d), (6.49), the high-n regularity estimate (6.72) and
(I)
the Cm,n -estimate (6.29), and apply the assumption (6.75) to obtain that
M
X (E)
a2m,n kJm,n
(≤3)
φk k2L2
m+n=0
M M
! M
!
X X 2 X 2
(0) (E) 2 (2) (≤2) (E)
. a2m,n kJm,n ek kH 1
w + 2
B0,n ϕ2n J0,n Z 2 a2m,n Jm,n φk
k L (suppχ
e1 ) L2
m+n=0 n=0 m+n=0
M
X M
X
(E) (I) (≤2)
+ (exp{−ν −1/6 } + λ2s ) a2m,n kJm,n
(≤3)
ek k2H 1
φk k2L2 + kw 2
B0,n ϕ2n kJ0,n Zk2L2 (suppχe1 )
k
m+n=0 n=0
(I)
+ |vyy | y=±1 ke
ωk kL2v .

Hence for λ, ν chosen small enough, we can apply the bound (6.10) to derive the following estimate
M
X (E)
a2m,n kJm,n
(3)
φk k2L2
m+n=0
M M
! ∞
!
X X 2 X
(0) (E) 2 (2) (0) (E) 2 (I)
. a2m,n kJm,n w
ek kH 1 + 2
B0,n ϕ2n J0,n Z 2 a2m,n kJm,n w
ek kL2 + ek k2H 1
kw
k L (suppχ
e1 ) k
m+n=0 n=0 m+n=0
(I) (I)
+ exp{−ν −1/9 }kw
ek k2H 1 + |vyy | y=±1
ke
ωk kL2v .
k

This concludes the proof.

(E)
6.4 Bounds on Fell
In this section, we prove Proposition 6.1 b). We take Fourier transform in x and reformulate the
problem (6.1) as follows
(E) 2 2 (E)
∆k Ψ k =wkE − (2 + (1 + χI )H) H E ∂v ψk + (2 + (1 + χI )H) H E ∂v Ψk

101


(E)
+ H E ∂v H + (H I − 1) ∂v H E ∂v ψk − H E ∂v H + (H I − 1) ∂v H E ∂v Ψk
=:wkE + R1 + R2 + R3 + R4 ,
(E)
Ψk (±1) =0.
Here F I = χI F, F E = χE F for functions F ∈ {w, H}. During the reformulation, we have used
(I) (E)
the decomposition that ψk = Ψk + Ψk . We further note that the support of the right hand side
forcing are supported in the region where χm+n ≡ 1 (2.4) and hence one has sufficient regularity
for the coordinate function H.
To derive the conormal Gevrey estimates, we can write down a similar equation as (6.24),
 
(E)
∆k χm+n |k|m q n Γnk Ψk
=|k|m q n Γnk wkE + |k|m q n Γnk R1 + |k|m q n Γnk R2 + |k|m q n Γnk R3 + |k|m q n Γnk R4
   
(E) (E) (E)
+ χm+n |k|m q n [∂yy , Γnk ]Ψk + χm+n [∂yy , q n ] |k|m Γnk Ψk + [∂yy , χm+n ] |k|m q n Γnk Ψk
7
X
=:|k|m q n Γnk wkE + Tj;m,n . (6.76)
j=1

Here we note that the wkE , Ri , i ∈ {1, 2, 3, 4} are supported in supportχE and χE χm+n = χE .
(E)
Hence, we do not have χm+n cut-off for the first few terms. However, the solution Ψk is supported
on the whole channel, we have to include the cutoff χm+n in the last three terms.

Proof of Proposition 6.1 b). We decompose the proof into steps:

Step # 1: General Plan. To estimate the solutions to (6.76), we invoke the exterior estimates
of wk , ψk and H to control the wkE , T1;m,n and T3;m,n contribution. Then we use the fact that
H is extremely small on the support of T2;m,n , T4;m,n to absorb their contribution. Finally, the
treatment of the T5;m,n , T6;m,n and T7;m,n are treated in the same fashion as previous subsections.
(E)
With this plan in mind, we test the equation (6.76) by χm+n |k|m q n Γnk Ψk to obtain that for
arbitrary M ∈ N,
M
!2/r
X e m+n
(2λ) (E)
kχm+n Ψ̊m,n;k k2H 1
(m + n)! k
m+n=0
!2/r !2/r
M
X e m+n
(2λ) X7 M
X (2 e m+n
λ) D E
E (E)
. kχm+n ẘm,n;k k2L2 + Tj;m,n, χm+n Ψ̊m,n;k
(m + n)! (m + n)!
m+n=0 j=1 m+n=0
M
!2/r 7
X e m+n
(2λ) X
E
=: kχm+n ẘm,n;k k2L2 + Tj . (6.77)
(m + n)!
m+n=0 j=1

Here we recall the norms that we employed in (3.5) and the simplified notation f˚m,n;k = |k|m q n Γnk fk ,
(2.29). Hence to derive the bound (6.2), it is enough to estimate the right hand side of (6.77).
Step # 2: Estimates of T1 , T2 terms.
For the T1 , T2 contributions in (6.77), we estimate them through integration by parts. Since
(E)
they have similar structures, we use Bk to denote {ψk , Ψk } and estimate them in the following
unified fashion,
!2/r
XM e m+n
(2λ) D E
Tj 1j∈{1,2} := Tι;m,n , χm+n Ψ̊m,n;k 1j∈{1,2}
(E)
(m + n)!
m+n=0

102
 M
!2/r n  
X e m+n
(2λ) X n
=
(m + n)! n1
m+n=0 n1 =0
D E

(E)
q n−n1 Γn−n1 (2 + (1 + χI )H) H E [q n1 , ∂v ]∂v (Γnk 1 |k|m Bk ), χm+n Ψ̊m,n;k
D

 E
(E)
− vy−1 q n−n1 Γn−n1 (2 + (1 + χI )H) H E q n1 ∂v Γnk 1 |k|m Bk , ∂y χm+n Ψ̊m,n;k
D

n1 E
(E)
− ∂y vy−1 q n−n1 Γn−n1 (2 + (1 + χI )H) H E q ∂v Γn1 |k|m Bk , χm+n Ψ̊m,n;k

Now we start to estimate the above expression. First we observe that the commutator term with
[q n1 , ∂v ] is non-vacuous only if n ≥ n1 ≥ 1. Hence, we apply the Sobolev embedding to estimate
the coordinate factors in the above expressions, and suitably adjust the position of the q-weight to
end up with the following

M n  1−1/r
!1/r !1/r !1/r
X X n e n−n1
(2λ) e m+n1
(2λ) e m+n
(2λ)
Tj 1j=1,2 .
m+n=0 n1 =0
n1 (n − n1 )! (m + n1 )! (m + n)!

1 (E)
kq[q n1 , ∂v ]∂v Γn1 |k|m Bk kL2 ([−1/2,1/2]c ) χm+n Ψ̊m,n;k 1n≥n1 ≥1
q L2
 
(E)
+ kq n1 ∂v (Γn1 |k|m Bk )kL2 ([−1/2,1/2]c ) ∇k χm+n Ψ̊m,n;k 2
L
(E) n−n1

+ kq n1 ∂v (χm+n1 Γn1 |k|m Bk )kL2 ([−1/2,1/2]c ) kχm+n Ψ̊m,n;k kL∞ q n−n1 ∂v (2 + (1 + χI )H)H E .
H1

Bm,n
Here in the last inequality, we have invoked a property of the Gevrey coefficient that nn1 Bm,n B ≤
1 0,n−n1


−1/r

, and the fact that χm+n 1m+n≥1 ≡ 1 on the support of χE . Now we
n m+n 1−1/r
n−n1 n−n1 ≤ C nn1
(a,b,c)
recall the definition of the Jm,n fk (3.6b), (3.8b), the combinatorial bound (6.88) and bound the
above expression as follows:

M n  1−1/r
!1/r !1/r !1/r
X X n e n−n1
(2λ) e m+n1
(2λ) e m+n
(2λ)
Tj 1j=1,2 .
n1 (n − n1 )! (m + n1 )! (m + n)!
m+n=0 n1 =0
!
X n−n1

(a,b,c) (E)
× Jm,n Ψk q n−n1 ∂v (2 + (1 + χI )H)H E
L2 H1
a+b+c=1
 X 
(a,b,0)
× kJm,n 1
Bk kL2 ([−1/2,1/2]c )
a+b=1
 !2/r 1/2  ! n !2/r
M
X e m+n M
X X n  2−2/r X e n−n1
(2λ) (E) 2 n (2λ)
. (1)
Jm,n Ψk  
m+n=0
(m + n)! L2
m+n=0 n =0
n1 n =0
(n − n1 )!
1 1

!2/r 1/2
n−n1
2 e m+n1
(2λ)
× q n−n1 ∂v (2 + (1 + χI )H)H E (1)
kJm,n 1
Bk k2L2 ([−1/2,1/2]c )  .
H1 (m + n1 )!

In the first estimate, we have invoked the fact that on the support of χE , it is permissible to intro-
(a,b,c)
duce the cut-off χm+n freely to generate the quantities Jm,n (· · · ). Thanks to the combinatorial

103
 P
2−2/r
bound (6.88), the sum nn1 =0 nn1 in the final expression is bounded by a constant Cr . Fur-
thermore, it is possible to switch the order of summation to estimate the last factor as we have
done previously:
 !2/r 1/2  !2/r 1/2
XM e m+n M
X e m+n
(2λ) (E) 2 (2λ)
Tj .  (1)
Jm,n Ψk   (1)
kJm,n Bk k2L2 ([−1/2,1/2]c ) 
m+n=0
(m + n)! L2
m+n=0
(m + n)!
 !2/r 1/2
M
X e n
(2λ) n
2
× q n ∂v (2 + (1 + χI )H)H E  .
n=0
n! H1

At this time, one might be tempted to apply the induction lemma (6.108b). However, the induction
estimate is valid when applied with the Gevrey coefficients am,n , which include ϕn ∼ hti−n factors.
The time-decaying ϕ compensates for the additional t-factors concealed in Γk . Here, the Gevrey
coefficients lack ϕ-factors. Nonetheless, thanks to the analysis in the proof of (6.108b), it is clear
that invoking the induction will only result in extra hti-growing factors.
To estimate the T1 contribution, we invoke the product estimates (6.119), (6.120), and the
hypotheses (2.22c) to obtain that
!2/r
1 X
M e m+n
(2λ)   2
(E)
T1 ≤ ∇k χm+n Ψ̊m,n;k
8 m+n=0 (m + n)! L2
!2/r
M
X (2 e m+n
λ) 1 −2/3+η
 
(γ) (α)
+ Cν 4 hti2 e− 8 ν k|k|m q n Γnk ψk k2H 1 ([−1/2,1/2]c )) EH + EH .
(m + n)! k
m+n=0

Here in the last line, we have the extra factor exp{− 81 ν −2/3+η } because there is a weight eW in
(α) (γ)
EH , EH . Now by Lemma 6.24, we can express the last term with Gevrey norms with ϕ weights.
(I) (E)
Combining the decomposition ψk = φk + φk , the estimates (3.12), (3.14), yields the following,
!2/r
M
1 X e m+n
(2λ)   2   
(E) −1/6 (γ) low (γ) (α) 2
T1 ≤ ∇k χm+n Ψ̊m,n;k + exp{−ν } E [wk ] + E Int [wk ] E H + E H .
8 (m + n)! L2
m+n=0
(6.78)
Next we consider the T2 contributions . Now combining a variation of the induction relation
e (m+n)/r
(6.108b) (with am,n replaced by (2(m+n)!
λ)
1/r ) , the Sobolev inequality, Young’s inequality and the

combinatorial bound (6.88) yields that

!2/r
XM e m+n
(2λ)   2
(E)
T2 ≤ ∇k χm+n Ψ̊m,n;k
m+n=0
(m + n)! L2
 !2/r 
∞
X e n
1 (4λ) n
2
×  + Chti2 q n ∂v (2 + (1 + χI )H)H E .
16 n! H1
n=0

Now we invoke the product estimates (6.119), (6.120), the hypotheses (2.22c), and Lemma 6.24 to
obtain that
!2/r
1 X
M e m+n
(2λ)   2
(E)
T2 ≤ ∇k χm+n Ψ̊m,n;k . (6.79)
8 (m + n)! L2
m+n=0

104
 This concludes Step # 2.
Step # 3: Estimates of T3 , T4 terms. The estimates in this step are similar to the ones in the
last step. Hence we will only highlight the main differences. The contributions from the T3 , T4 can
(E)
be expressed in the following unified fashion (with Bk = ψk for j = 3 and Bk = Ψk for j = 4)
!2/r
M
X e m+n
(2λ) D E
Tj 1j=3,4 = Tj;m,n , χm+n Ψ̊m,n;k 1j=3,4
(E)

m+n=0
(m + n)!
M
!2/r n  
X e m+n
(2λ) X n
=
m+n=0
(m + n)! n =0
n1
1
D
E
(E)
× q n−n1 Γn−n
k
1
H E
∂v H + (H I − 1) ∂v H E q n1 Γnk 1 ∂v |k|m Bk , χm+n Ψ̊m,n;k .

We observe that this expression is similar to the T1 , T2 terms in Step # 2. Hence, we apply the
Sobolev embedding, the combinatorial bound (6.88) to estimate the contributions as follows:

|Tj=3,4 | .
 !2/r 1/2  ! n !2/r
M
X e m+n M
X X n  2−2/r X e n−n1
 (2λ) (E) 2
  n (2λ)
χm+n Ψ̊m,n;k
m+n=0
(m + n)! L∞
m+n=0 n =0
n1 n =0
(n − n1 )!
1 1

!2/r 1/2
n−n1
2 e m+n1
(2λ)
× q n−n1 ∂v H E ∂v H + (H I − 1) ∂v H E (1)
kJm,n 1
Bk k2L2 ([−1/2,1/2]c ) 
L2 (m + n1 )!
 !2/r 1/2  !2/r 1/2
M
X e m+n M
X e m+n
(2λ) (E) 2 (2λ)
. ∇k (χm+n Ψ̊m,n;k )   (1)
kJm,n Bk k2L2 ([−1/2,1/2]c ) 
m+n=0
(m + n)! L2
m+n=0
(m + n)!
 !2/r 1/2
M
X e n
(2λ) n
2
× q n ∂v H E ∂v H + (H I − 1) ∂v H E  . (6.80)
n=0
n! L2

Now we estimate the last factor which is associated with the coordinate system quantity H. We
invoked a variant of the Gevrey product rule (6.121): for sufficiently smooth functions f, g, we have
the following bound for 1/r = 1/p + 1/q,
 !2/r 
X∞ e n
 (2λ) n
kq n ∂v (f g)k2Lr 
m+n=0
n!
 !2/r  !2/r 
X∞ e n ∞
X e n
(2 λ) n (2λ) n
. kq n ∂v f k2Lp   kq n ∂v gk2Lq  .
n! n!
m+n=0 m+n=0

The proof of this estimate is similar to the proof of (6.121) and we omit the details. By invoking
the above product rule, we obtain the following
 !2/r 1/2
M
X e n
(2λ) n
2
 q n ∂v H E ∂v H + (H I − 1) ∂v H E 
n! L2
n=0

105
  !2/r 1/2  !2/r 1/2
M
X e n M
X e n
(2λ) n 2 (2λ) n+1 2
. q n ∂v H E   q n ∂v H 
n! L∞ n! L2 (supportχE )
n=0 n=0
 !2/r 1/2  !2/r 1/2
M
X e n M
X e n
(2 λ) n+1 E 2 (2λ) n 2
+ q n ∂v H   q n ∂v (H I − 1) 
n! L2 n! L∞ (supportχE )
n=0 n=0
 
(α) (γ)
. exp{−ν −1/5 } EH + EH .

Here in the last line, we have invoked the assumption (2.22).

Now we are ready to conclude from the estimate (6.80) and the coordinate control. Similar to
the estimate of (6.78), we have that
!2/r
1 X
M e m+n
(2λ)   2 −1/6
 (γ) 
(E) γ 2
T3 ≤ ∇k χm+n Ψ̊m,n;k + Ce−ν (EH α 2
) + (EH ) E [wk ] + EInt [wk ] .
8 m+n=0 (m + n)! L2

(6.81)

Similar to the estimate of (6.79), we have that the T4 contribution is bounded as follows
  X !2/r
1
M e m+n
(2λ)   2
−ν −1/7 γ 2 α 2 (E)
T4 ≤ + Ce ((EH ) + (EH ) ) ∇k χm+n Ψ̊m,n;k . (6.82)
8 m+n=0
(m + n)! L2

This concludes Step # 3.

Step # 4: Conclusion. The T5;m,n , T6;m,n and T7;m,n terms in (6.76) are similar to the terms in
(6.24). Hence they can be estimated in a similar fashion as in the previous subsections. We omit
further details for the sake of brevity. 2 Combining the above estimates ((6.77), (6.78), (6.79),
(6.81), (6.82)) and the definitions of E (γ) [wk ], EInt [wk ], we have obtained that
!2/r
M
X e m+n
(2λ)   
(E) −1/6 (γ) (α) 2
kχm+n |k|m q n Γnk Ψk k2H 1 . e−ν E (γ) [wk ] + EInt [wk ] 1 + EH + EH .
m+n=0
(m + n)! k

By taking M → ∞, we have the result.

6.5 Proof of Proposition 3.1

We are now ready to bring together the bounds established in this section in order to prove Propo-
sition 3.1.

Proof of Proposition 3.1. Combining Proposition 6.1 and Proposition 6.2 yields the result.

6.6 Technical Lemmas: ICC-Method

We first present a simple lemma, which is useful when we derive maximal regularity estimates.

Lemma 6.12. Assume that Fk ∈ L2 , k 6= 0, and Ψk solves the following elliptic equation

−∂yy Ψk + |k|2 Ψk = Fk (6.83)

2
In previous sections, we treat these terms with maximal regularity trick H 2 . Here we only do H 1 estimates. So
should be easier.

106
subject to homogeneous boundary conditions Ψk (y = ±1) = 0 or ∂y Ψk (y = ±1) = 0. Then the
following estimates hold
√
kΨk kḢ 2 :=k∂yy Ψk kL2 + k|k|∂y Ψk kL2 + k|k|2 Ψk kL2 ≤ (3 + 2)kFk kL2 ; (6.84)
k
2
kΨk kḢ 2 :=k∂v Ψk kL2 + k|k|∂v Ψk kL2 + k|k|2 Ψk kL2 ≤ C(kvy−1 kL∞
t,y
, kvy kL∞ Wy1,∞ )kFk kL2 (. 6.85)
k,v t

Proof. We first test the equation against Ψk and implement integration by parts. Thanks to the
Dirichlet/Neumann boundary condition, we obtain

k∂y Ψk k22 + |k|2 kΨk k22 ≤ kFk k2 kΨk k2 .

As a result, we have that

k|k|2 Ψk k2 ≤ kFk k2 .

By taking the L2 -norm on both side of the equation (6.83), we have that

k∂yy Ψk k2 ≤ kFk k2 + k|k|2 Ψk k2 ≤ 2kFk k2 .

Finally, we observe that, direct integration by parts yields the inequality

ˆ 1 ˆ 1
2
kk∂y Ψk k2 = |k|∂y Ψk |k|∂y Ψk dy = − |k|2 Ψk ∂yy Ψk dy ≤ k∂yy Ψk k2 k|k|2 Ψk k2 ≤ 2kFk k22 .
−1 −1

Combining the estimates we developed so far, we obtained (6.84).

To derive (6.85), it is enough to observe that

k|k|∂v Ψk kL2y ≤Ckvy−1 kL∞

t,y
k|k|∂y Ψk kL2y ,
2
 
k∂v Ψk kL2y ≤Ckvy−1 kL∞t,y
k∂y vy
−1 ∞
kL t,y
k∂y Ψ k k 2
Ly + kv −1 ∞
y kL t,y
k∂yy Ψ k kLy .
2

Now direct application of (6.84) yields (6.85).

Lemma 6.13 (Maximal regularity estimates with nonvanishing boundary condition). Assume that
Fk ∈ L2 .
a) Consider the solution Ψk to the following equation

−∂yy Ψk + |k|2 Ψk = Fk , ∂y Ψk (y = ±1) = a± ∈ R.

Then the following estimate holds

k∂yy Ψk k2 + k|k|∂y Ψk k2 + k|k|2 Ψk k2 . kFk k2 + |k|1/2 |a+ | + |k|1/2 |a− |. (6.86)

b) Consider the solution Φk to the following equation

−∂yy Φk + |k|2 Φk = Fk , Φk (y = ±1) = b± ∈ R.

Then the following estimate holds

k∂yy Φk k2 + k|k|∂y Φk k2 + k|k|2 Φk k2 . kFk k2 + |k|3/2 |b+ | + |k|3/2 |b− |. (6.87)

c) In either of the scenarios above, if the following boundary constraint is satisfied,

ReΨk ∂y Ψk y=±1
= 0,

then the estimate can be improved to be (6.84).

107
Proof. To derive the estimate with nonzero boundary conditions, we decompose the solution into
the free part Ψfk and the boundary corrector Ψbk , i.e.,

−∆k Ψfk = Fk , ∂y Ψfk (±1) = 0;

−∆k Ψbk = 0, ∂y Ψbk (±1) = a± .

The solution Ψfk can be estimated with (6.84), whereas the boundary corrector can be solved
explicitly:
a+ a−
Ψbk = cosh(|k|(1 + y)) − cosh(|k|(1 − y)).
|k| sinh(2|k|) |k| sinh(2|k|)

Given the explicit form of the solution, one can derive the Ḣk2 bound for Ψbk as follows

k∂yy Ψbk k2 + k|k|∂y Ψbk k2 + k|k|2 Ψbk k2 . |k|1/2 (|a+ | + |a− |).

The proof of this inequality is direct and we omit the details for the sake of brevity. Combining
this estimate of the boundary corrector and the estimate of the free part, yields the estimate (6.86).
The proof for the (6.87) is similar and we omit the details.

Before implementing the estimations, we recall Lemma 4.4. The product rule (without c cor-
rection) should be simplified to a simple combination fact
Lemma 6.14. For any positive ζ > 0, there exists a constant C = C(ζ) such that the following
estimates hold
Xn  −ζ
n
≤C(ζ); (6.88)
ℓ
ℓ=0
X n−ζ
n−1
≤C(ζ)n−ζ , ∀n ∈ N\{0}. (6.89)
ℓ
ℓ=1

Proof. First we observe that (6.88) is a direct consequence of (6.89) thanks to the fact that the
ℓ = 0 and ℓ = n term are bounded by 1. Hence, in the remaining part of the proof, we focus on
(6.89). For a fixed ζ > 0 in (6.89), we first choose a large enough N (ζ) := max{10, ⌈ ζ2 ⌉ + 1}. For
n ≤ 4N , the estimate (6.89) is direct because n−ζ has a fixed lower bound. Hence we only consider
the n > 4N case. Now we expand the combinatoric numbers
 
⌊n ⌈ 3n
X  (n − ℓ)!ℓ! ζ
n−1 XN X4
⌋
X 4
⌉ n−N
X n−1
X 
(n − ℓ)!ℓ!
ζ X 5
= + + + +  =: Ij .(6.90)
n! n n
n!
ℓ=1 ℓ=1 ℓ=N +1 ℓ=⌊ 4 ⌋ ℓ=⌈ 4 ⌉ ℓ=n−N +1 j=1

Thanks to the symmetric property

of the nℓ , it is enough to estimate the first three terms in (6.90).
For the first term, we use nℓ ≥ n for ℓ ≥ 1 to obtain that

I1 ≤N n−ζ ≤ C(ζ)n−ζ . (6.91)

For the second term, we recall the choice of N to obtain that

⌊n ⌋  ζ
X4
ℓ!
I2 =
n(n − 1)...(n − ℓ + 1)
ℓ=N +1

108
 ⌊n ⌋  ζ  ζ
X4
1 · 2 · (...) · N ℓ(ℓ − 1)...(N + 1)
≤
n(n − 1)(n − 2)...(n − N + 1) (n − N )....(n − ℓ + 1)
ℓ=N +1
n n
⌊4⌋ ⌊4⌋
X 2ζN (N !)ζ X 2ζN (N !)ζ
≤ ≤ ≤ C(ζ)n−ζ . (6.92)
nζN n2+ζ
ℓ=N +1 ℓ=N +1

For the third term, we apply the Stirling’s formula n! ∼ nn+1/2 e−n to obtain that for n, ℓ, n − ℓ ≥
N, ζ > 0
⌈ 3n ⌉  −ζ ⌈ 3n ⌉ !ζ
X 4
n X 4
ℓℓ+1/2 (n − ℓ)n−ℓ+1/2
≤C
n
ℓ n nn+1/2
ℓ=⌊ 4 ⌋ ℓ=⌊ 4 ⌋

⌈ 3n
X  7 nζ √
4
⌉
≤C ( n)ζ ≤ C(ζ)n−ζ . (6.93)
n
8
ℓ=⌊ 4 ⌋

Here we have used the fact that in the range of the sum, ℓ, n−ℓ ≤ 87 n. Combining the decomposition
(6.90) and the argument after, the estimates (6.91), (6.92), and (6.93), we have proven the estimate
(6.89).

In the following, we will present four general lemmas regarding the ICC-method (Lemma 6.15,
Lemma 6.16, Lemma 6.17, Lemma 6.18).
Lemma 6.15. The j-th derivative (1 ≤ j ≤ 4) of the q n has the following representation
j
∂v (q n ) = qg j (n−j)+
n,j n q . (6.94)
Here (n − j)+ := max{n − j, 0} and the qg
n,j are bounded functions such that

qn,j kL∞ ≤C(kqkW j,∞ , kvy−1 kL∞ , kvy kW j−1,∞ );

kg (6.95a)
k∂y qg
n,j kL∞ ≤C(kqkW j+1,∞ , kvy−1 kL∞ , kvy kW j,∞ ). (6.95b)
Here the constants are independent of n.
As a consequence, the following commutator relation holds
j  
X
j n j ℓ (n−ℓ)+ j−ℓ
[∂v , q ]f = qg
n,ℓ n q ∂v f. (6.96)
ℓ
ℓ=1

Proof. We start the proof by recalling the Faà di Bruno’s formula:

X Yn
dj j! (m1 +...+mj )
f (g(x)) = f (g(x)) · (g(ℓ) (x))(6.97)
mℓ
,
dxj m1 !(1!)m1 m2 !(2!)m2 ...mj !(j!)mj
(m1 ,m2 ,...,mj )∈Sj ℓ=1
( j
)
X
Sj := (m1 , m2 , ..., mj ) ∈ Nj ℓmℓ = j .
ℓ=1

For a fixed t ≥ 0, we write the physical variable y as a function of the variable v, i.e., y = y(v). A
j
direct application of the formula (6.97) yields the expression of the quantity ∂v q(y), j ≤ 4,
j ℓ
!mℓ
j X j! Y ∂v y(v)
j (m1 +m2 +...+mj )
∂v q(y) =(∂v ) q(y(v)) = q (y(v))
m1 !m2 !...mj ! ℓ!
(m1 ,m2 ,...,mj )∈Sj ℓ=1

109
 j ℓ−1 −1 !mℓ
X j! Y ∂v vy
= q (m1 +m2 +...+mj ) (y(v)) .
m1 !m2 !...mj ! ℓ!
(m1 ,m2 ,...,mj )∈Sj ℓ=1

Pj
We observe that ℓ=1 mj ≤ j ≤ 4. Hence the j-th ∂v -derivative of the q-weight is bounded,
j

k∂v qkL∞ ≤ C kqkW j,∞ , kvy−1 kL∞ , kvy kW j−1,∞ , j ≤ 4. (6.98)

We apply the Faà di Bruno formula again to derive the following

j j
∂v (q n (y)) =∂v (q n (y(v)))
 (Pj j ℓ
!mℓ
X mℓ ) Y
j! d ℓ=1
n ∂v q
= (q )
m1 !m2 ! ... mj ! dq ℓ!
(m1 ,m2 ,...,mj )∈Sj ℓ=1
X j! n!
= q n−m1 −...−mj
m1 !m2 ! ... mj ! (n − m1 − m2 ... − mj )!
(m1 ,m2 ,...,mj )∈Sj
j ℓ
!mℓ
Y ∂v q
× 1n−m1 −...−mj ≥0
ℓ!
ℓ=1
j max{n−j,0}
=:g
qn,j n q , j ≤ 4.

Here the explicit form of the coefficients are as follows

X j! n!
qg
n,j := j
m1 !m2 ! ... mj ! n (n − m1 − m2 ... − mj )!
(m1 ,m2 ,...,mj )∈Sj
m1 +m2 +...+mj ≤n
j ℓ
!mℓ
Y ∂v q
× q n−max{n−j,0}−m1 −...−mj . (6.99)
ℓ!
ℓ=1
Pj
Since for (m1 , ..., mj ) ∈ Sj , the relation ℓ=1 mj ≤ j ≤ 4 is satisfied (6.97). Combining this fact
with the upper bound (6.98) yields that
j
|∂v (q n )| =|g
qn,j |nj q max{n−j,0} , qn,j kL∞ ≤ C(kqkW j,∞ , kvy−1 kL∞ , kvy kW j−1,∞ ),
kg 1 ≤ j ≤ 4.

Hence we conclude the proof of (6.94), (6.95a). Next we note that there are at most 2j copies of
(··· )
∂v q factors in the above expression (6.99), hence the estimate (6.95b) is a direct consequence of
differentiating the expression (6.99). Moreover, an application of the product rule and the relation
(6.94) yields (6.96). This concludes the proof of the lemma.

Lemma 6.16. The following estimates hold for any n ≥ 1 and j ∈ {1, 2, 3},
j
Bm,n ϕn+1 k[∂v , q n ](χm+n |k|m Γnk fk )kL2
Xn  ′ σ
(n−n′ )s n′ +1 n ! (≤j)
. Bm,n′ λ ϕ kJm,n′ fk kL2
′
n!
n =(n−2)+
m
X n−1
X X  σ
|m′ , n′ |!
+ 1n<j (|m,n|−|m′ ,n′ |)s n′ +1 (0,b,c)
Bm′ ,n′ λ ϕ kJm′ ,n′ fk kL2 (. 6.100a)
|m, n|!
m′ =(m−1)+ n′ =0 b+c≤j

110
Here, (n − 2)+ = max{n − 2, 0}, |m, n| = m + n and the implicit constants depend only on
kvy−1 kL∞ , kvy kW 2,∞ and independent of m, n. Moreover,
n
X

′ (≤2)
Bm,n ϕn+1 ∂v [∂v , q n ](χm+n |k|m Γnk fk ) L2
. Bm,n′ ϕn +1 kJm,n′ fk kL2 ; (6.100b)
n′ =(n−1)+
  Xm n
X
n+1 2 ′ (≤3)
Bm,n ϕ ∂v [∂v , q n ] (χm+n |k|m Γnk fk ) . Bm′ ,n′ ϕn +1 kJm′ ,n′ f(6.100c)
k kL2 .
L2
m′ =(m−1)+ n′ =(n−2)+

Here the implicit constants depends only on kvy−1 kL∞ , kvy − 1kW 2,∞ .
Proof. Step # 0: Setup. Without loss of generality, we assume that k ≥ 1 in our proof. The
k = 0 and k < 0 cases are treated in a similar fashion. We note that if a + b + c = j > n, the
(a,b,c)
parameter a in the Jm,n fk (3.6b) must be zero. Hence there is a natural distinction between the
j ≤ n and j > n cases. We prove the following relations in these two different regimes
j X
[∂v , q n ] (χm+n |k|m Γnk fk ) = Aa,b,0;j (a,b,0)
m,n Jm,n fk , j ∈ {1, 2, 3}, n ≥ j; (6.101a)
a+b=j,a>0
Xm X
2 ′ 0,b,c;2 (0,b,c)
ϕ[∂v , q] (χm+1 |k|m Γk fk ) = hmi(m−m )+ (1+σ) Am ′ ,0 Jm′ ,0 fk ;

m′ =(m−1)+ b+c≤2
m
X n−1
X X
n 3 n ′ ′ (0,b,c)
|ϕ [∂v , q ] (χm+n |k|m Γnk fk ) | . hmi(n−n )(1+σ) ϕn |Jm′ ,n′ fk |, 1 ≤ n ≤ 2.
m′ =(m−1)+ n′ =(n−2)+ b+c≤3

(a,b,c)
Here, most of the indices in the coefficients Aa,b,c;j
m,n match their companion Jm,n fk , and the extra
2
index ‘j’ highlights the associated commutator (e.g., j = 2 if one is dealing with [∂v , q n ]). These
coefficients satisfy the following estimates for ℓ ∈ {0, 1}, ℓ + j ≤ 3,
 
k∂yℓ Aa,b,0;j
m,n kL∞ y
≤C kvy−1 kL∞
y
, kvy − 1kW j−1+ℓ,∞ , j ≤ n; (6.101b)
y
 
k∂yℓ Am,n
0,b,c;j
kL∞
y
≤C kvy−1 kL∞
y
, kvy − 1kW j−1+ℓ,∞ hmiℓ(1+σ) , j > n.
y

Moreover,
n
X X

′ (a,b,c)
ϕn
∂v [∂v , q n
] (χm+n |k|m Γnk fk ) ≤ Ba,b,c
m,n′ ϕ
n
Jm,n′ fk , kBa,b,c
m,n kL∞
y
. 1, n ≥ 1.
n′ =(n−1)+ a+b+c≤2
(6.101c)

Moreover,
  n
X X
2 ′ (a,b,c)
ϕn
∂v [∂v , q n ] (χm+n |k|m Γnk fk ) ≤ Ca,b,c
m,n′ ϕ
n
Jm,n′ fk , kCa,b,c
m,n kL∞
y
. 1, n ≥ 3;
n′ =(n−2)+ a+b+c≤3
(6.101d)
m
X n
X X
2 ′ ′ (0,b,c)
ϕn ∂v [∂v , q n ] (χm+n |k|m Γnk fk ) . hmi(m+n−m −n )(1+σ) Jm′ ,n′ fk , n ∈ {1, 2}.
m′ =(m−1)+ n′ =(n−2)+ b+c≤3

Here the implicit constant only depends on the kvy−1 kL∞

y
, kvy − 1kHy3 .

111
 These estimates, combined with the bound (4.29), yields the results (6.100a), (6.100b) and
(6.100c). For example, to derive the third estimate in (6.100a), we invoke the (4.29) to get that
3 X
Bm,n ϕn+1 k[∂v , q n ](χm+n |k|m Γnk fk )kL2 . kAa,b,0;3
m,n kL∞ Bm,n ϕ
n+1 (a,b,0)
kJm,n fk kL2
a+b=3,a>0
X
n+1 (a,b,0)
. Bm,n ϕ kJm,n fk kL2 , n ≥ 3;
a+b=3,a>0
3
Bm,n ϕn+1 k[∂v , q n ](χm+n |k|m Γnk fk )kL2
m
X n−1
X X ′ Bm,n ′ (0,b,c)
. hmi(n−n )(1+σ) Bm′ ,n′ ϕn +1 kJm′ ,n′ fk kL2
Bm′ ,n′
m′ =(m−1)+ n′ =(n−2)+ b+c=3
m
X n−1
X X  σ
(|m,n|−|m′ ,n′ |)s |m′ , n′ |! ′ (0,b,c)
. Bm′ ,n′ λ ϕn +1 kJm,n′ fk kL2 , n ≤ 2.
|m, n|!
m′ =(m−1)+ n′ =(n−2)+ b+c=3

We omit the derivation of other inequalities since they are similar.

Thanks to the above discussion, the remaining task is to develop (6.101). It turns out that
the most technical step is the proof of (6.101a), (6.101b). Because to get the estimates (6.101c),
(6.101d), it is also necessary to derive the explicit form of the coefficient Aa,b,c;j
m,n in (6.101a). Hence,
we further decompose the proof of (6.101a) into the high regularity regime and the low regularity
regime.
Step # 1a: Proof of (6.101a)j≤n , (6.101b)j≤n . We apply induction to prove the result:
h i X
j
∂v , q n (χm+n |k|m Γnk fk ) = Aa,b,0;j (a,b,0)
m,n Jm,n fk , 1 ≤ j ≤ 3, (6.102)
a+b=j
 a X       h
n j j − i1 j − i1 − i2 j − i1 − · · · − ih−1 Y
Aa,b,0;j
m,n = (−1) h+1
··· q]
n,ih′ .
m+n i1 i2 i3 ih ′
i1 +···+ih =a h =1
i1 ,··· ,ih 6=0

Here (i1 , · · · , ih ) is a partition of the integer a and the terms with a = 0 are vacuous. The qg n,ih are
a,b,0;j
defined in (6.99). Moreover, thanks to the bounds (6.95a), (6.95b), the coefficients Am,n has the
following bound for n ≥ j, m ∈ N,

kAa,b,0;j
m,n kL∞ y
≤C(kvy−1 kL∞
y
, kvy − 1kWyj−1,∞ ), a + b = j ∈ {1, 2, 3};
k∂y Aa,b,0;j
m,n kL∞ y
≤C(kvy−1 kL∞
y
, kvy − 1kWyj,∞ ), a + b = j ∈ {1, 2}.

This is the first estimate in (6.101b).

As a start, it is clear to see that for j = 1, the commutator matches the expression (6.102),
 
n m n m n n n m n n n (1,0,0)
[∂v , q ](χm+n |k| Γk fk ) = |k| ∂v (q )χm+n Γk fk = qg n,1 (χm+n |k| q Γk fk ) = q
g n,1 Jm,n fk .
q m+n

Assume that the expression (6.102) holds for the 1, · · · , (j − 1)-th levels. Then for the j-th level,
we have that an application of the Lemma 4.3 and the expressions (6.94), (6.96) yields the relation
 j n
∂v , q (χm+n |k|m Γnk fk )

112
 j  
X j ni1 n j−i1
= qg
n,i1 q ∂v (χm+n |k|m Γnk fk )
i1 q i1
i1 =1
j  
X   i1
j n (m + n)i1 j−i1
= qg
n,i1 ∂v (χm+n |k|m q n Γnk fk )
i1 m+n q i1
i1 =1
j  
X   i1
j n (m + n)i1 n j−i1
+ qg
n,i1 [q , ∂v ](χm+n |k|m Γnk fk ).
i1 m+n q i1
i1 =1

(m+n)i1 (a,b,0)
Now we invoke the induction hypothesis (6.102) and the relation q i1
1
Jm,n fk a+i1 +b≤n =
fk 1a+i1 +b≤n to obtain
(a+i ,b,0)
Jm,n 1
 
j n
∂v , q (χm+n |k|m Γnk fk )
j 
X  i1   j
X X   i1  
n j (i1 ,j−i1 ,0) n j a,b,0;j−i1 (m + n)i1 (a,b,0)
= qg
n,i1 Jm,n fk − Am,n qg
n,i1 Jm,n fk
m+n i1 m+n i1 q i1
i1 =1 i1 =1 a+b=j−i1
j
X   i1  
n j (i1 ,j−i1 ,0)
= qg
n,i1 Jm,n fk
m+n i1
i1 =1
j
X X X  i1 +a   
h+1 n j j − i1
+ (−1)
m+n i1 i2
i1 =1 (a+i1 )+b=j i2 +···+ih =a
i2 ,··· ,ih 6=0
    h
!
j − i1 − i2 j − i1 − i2 − · · · − ih−1 Y
(a+i1 ,b,0)
× ··· qg
n,i1 q]
n,ih′ Jm,n fk
i3 ih ′
h =2
X X  a      
n h+1 j j − i1 j − i1 − i2 j − i1 − i1 − · · · − ih−1
= (−1) ···
m+n i1 i2 i3 ih
a+b=j i1 +···+ih =a
i1 ,··· ,ih 6=0
h
!
Y
(a,b,0)
× qg
n,ih Jm,n fk
h′ =1
X
= Aa,b,0;j (a,b,0)
m,n Jm,n fk .
a+b=j

Here in the second to last line, we have redefined a. As a result, we have obtained (6.101a) for
j ≤ n.
Step # 1b: Proof of (6.101a)j>n , (6.101b)j>n . Next we consider the j > n case. Since j ≤ 3,
we have n ∈ {1, 2}. Hence we only have 5-sub-cases to consider. The main technical difficulty
(0,b,c)
here is to keep track of the coefficients before Jm,n fk . As the number of derivative ∂v increases,
one will quickly find that the expression of the coefficients becomes complicated. To make the
presentation clear, we will first do a precise computation on one of the cases. For other cases,
we will use the following notation convention: We say that the coefficient Fb,c;jm,n (associated with
j n
[∂v , q ] commutators) is admissible if it satisfies the following two properties:
 ℓ j−1
• Fb,c;j −1 −1
m,n depends only on vy , ∂y vy ℓ=0 , |k| , t;

113
 ℓ
• k∂v Fb,c;j
m,n kL∞ . hmi
ℓ(1+σ) , ℓ = 0, 1. Here the implicit constant depends only on kv −1 k ∞ , kv k
y Lt,y y L∞ Wy2,∞ .
t

When we do the computation, the admissible coefficients will be changing line by line to incorporate
all the contributions. We choose to drop the a index because it is automatically zero in the low
regularity setting.
We invoke the commutator relation (4.6), the cutoff bound (2.5a), the definition (6.94) and the
computation facts that adℓ∂ (q) = qf ℓ 2
1,ℓ . 1, ad∂ (q ) = qf
ℓ (2−ℓ)+ . 1 to obtain that
2,ℓ 2 q
v v

ϕn [∂v , q n ](χm+n |k|m Γnk fk ) = ϕn ng
qn,1 χm+n q n−1 (∂v + ikt) |k|m Γkn−1 fk


qn,1 χm+n ) 1n=2 [q n−1 , ∂v ] Γkn−1 (χm |k|m fk ) + Jm,n−1 fk + itJm,n−1 fk
(0,1,0) (0,0,1)
=ϕn (ng
(0,1,0) (0,0,1)
=ϕn (ng
qn,1 χm+n ) Jm,n−1 fk + ϕn it (ng qn,1 χm+n ) Jm,n−1 fk
 
− 1n=2 (2g 2 (0,1,0) (0,0,1)
q2,1 qg
1,1 χm+2 )ϕ Jm,0 fk + itJm,0 fk .

It is direct to see that

n−1
X X ′ ′ (0,b,c)
|ϕn [∂v , q n ](χm+n |k|m Γnk fk )| . hmi(n−1−n )+ (1+σ) ϕn Jm,n′ fk , n ∈ {1, 2}.
n′ =(n−2)+ b+c=1

Moreover, we can clearly see that the coefficients are admissible because n = 1, 2 and the derivative
of χm+n is bounded by Chmi1+σ , (2.5a). Next we consider the following case,
2
ϕ[∂v , q](χm+1 |k|m Γk fk )
 
= ϕ ad2∂ [q](χm+1 Γk (χm |k|m fk )) + 2ad∂v [q]∂v (χm+1 Γk (|k|m fk ))
v
   
(0,1,0) (0,0,1) (0,2,0) (0,1,1)
1,2 χm+1 ϕJm,0 fk + i(ϕt)Jm,0 fk + 2g
= qg q1,1 χm+1 ϕJm,0 fk + itϕJm,0 fk
   
+ 1m≥1 2gq1,1 [∂v χm+1 ] ϕJm−1,0 fk + i(ϕt)Jm−1,0 fk + 1m=0 2g
(0,1,1) (0,0,2) (0,1,0) (0,0,1)
q1,1 [∂v χ1 ] ϕJ0,0 fk + i(ϕt)J0,0 fk
m
X X ′ (0,b,c)
= hmi(m−m )+ (1+σ) Fb,c;2
m′ ,0 Jm′ ,0 fk .
m′ =(m−1)+ b+c≤2

This is consistent with (6.101a). With the above two examples, we can summarize the idea. First
of all, we apply the commutator expansion (4.6) to expand the commutator properly. Then we
commute the cut-off function χ··· out of the main battle ground. Finally, we continuously commute
(0,b,c)
the boundary weight q n to obtain various Jm,n fk terms. Consequently,
2
ϕ2 [∂v , q 2 ](χm+2 |k|m Γ2k fk )




= ϕ2 ad2∂ [q 2 ] χm+2 Γ2k (χm |k|m fk ) + 2ϕ2 ad∂v [q 2 ] ∂v χm+2 Γ2k (|k|m fk )
X v b,c;2 (0,b,c) 2
= Fm,0 Jm,0 fk + 4g q1,1 ϕ2 (∂v χm+2 )qΓ2k (|k|m fk ) + 4g
q1,1 ϕ2 χm+2 q(∂v + ikt∂v )Γk (|k|m fk )
b+c=2
X ′ ′ (0,b,c) 2
= hmi(1−n )(1+σ) Fb,c;2 n
q1,1 ϕ2 χm+2 ([q, ∂v ] + ikt[q, ∂v ])Γk (χm |k|m fk )
m,n′ ϕ Jm,n′ fk + 4g
n′ ∈{0,1}
b+c=2
1
X X ′ ′ (0,b,c)
= hmi(1−n )(1+σ) Fb,c;2 n
m,n′ ϕ Jm,n′ fk .
n′ =0 b+c=2

114
This is consistent with (6.101a). Through similar but more complicated computations, we obtain
the estimate of the other two commutator. Since we do not need to keep track of their derivatives,
we choose to bound them directly (with implicit constants depends only on kvy−1 kL∞
t,y
, kvy kL∞ Wy2,∞ ),
t

2
X
3 3−ℓ ℓ
ϕ|[∂v ,q](χm+1 |k|m Γk fk )| . |∂v q|ϕ ∂v (χm+1 Γk |k|m fk )
ℓ=0
2
X ℓ
X  
ℓ−ℓ1 1+ℓ1 ℓ1
. ϕ ∂v χm+1 ∂v + ikt∂v (|k|m fk )
ℓ=0 ℓ1 =0
2 X
X ℓ X X
. 1(ℓ,ℓ1 )6=(2,0) hmi(1+σ) Jm,0 fk + 1m≥1 hmi2(1+σ)
(0,b,c) (0,b,c+1)
|Jm−1,0 fk |
ℓ=0 ℓ1 =0 b+c=1+ℓ1 b+c=1
X
+ 1m=0
(0,b,c)
|J0,0 fk |.
b+c=1

Finally, we estimate the most complicated expression:

2
X
3 3−ℓ 2 ℓ
ϕ2 [∂v , q 2 ](χm+2 |k|m Γ2k fk ) . |∂v q |ϕ2 ∂v (χm+2 Γ2k |k|m fk )
ℓ=0
1
X ℓ 2
. ϕ2 |∂v (χm+2 |k|m Γ2k fk )| + ϕ2 |∂v q||q∂v (χm+2 |k|m Γ2k fk )|
ℓ=0
X (0,b,c) 2 2
. hmi(1+σ) Jm,0 fk + ϕ2 |∂v q||q[∂v , χm+2 ](|k|m Γ2k fk )| + ϕ2 |∂v q|χm+2 |q∂v (|k|m Γ2k fk )|
b+c=3
X 1
X
(0,b,c) 2−ℓ ℓ
. hmi(1+σ) Jm,0 fk + ϕ2 |∂v χm+2 ||∂v (|k|m Γ2k fk )|
b+c=3 ℓ=0
2 2
+ ϕ2 χm+2 |[q, ∂v Γk ](|k|m Γk fk )| + ϕ2 χm+2 |∂v Γk (χm+1 |k|m qΓk fk )|
1
X X 1
X ℓ
(n−n′ )(1+σ) n′
hmi(1+σ) |∂v Γ2k (χm |k|m fk )|1supp χm+2
(0,b,c) 2
. hmi ϕ Jm,n′ fk + ϕ
n′ =0 b+c=3 ℓ=0
2 2 m
+ ϕ χm+2 |[q, ∂v Γk ]Γk (χm |k| fk )|
1
X X ′ ′ (0,b,c)
. hmi(n−n )(1+σ) ϕn Jm,n′ fk .
n′ =0 b+c≤3

This concludes the proof of (6.101a) in the j < n case.

Step # 2: Proof of (6.101c). We apply the expression (6.101a)j=1 to derive the following

ϕn ∂v [∂v , q n ](χm+n |k|m Γnk fk ) = ϕn ∂v A1,0,0;1 (1,0,0) n 1,0,0;1

m,n Jm,n fk + ϕ Am,n ∂v Jm,n fk
(1,0,0)
 1,0,0;1 (1,0,0)
 n n 1,0,0;1 ∂v q (2,0,0) n 1,0,0;1 (1,1,0)
ϕ ∂v Am,n Jm,n fk − ϕ Am,n m+n Jm,n fk + ϕ Am,n Jm,n fk , n ≥ 2;
=

 (1,0,0) (0,2,0) (0,1,1)
ϕ∂v A1,0,0;1 1,0,0;1
m,1 Jm,1 fk + Am,1 (ϕJm,0 fk + i(ϕt)Jm,0 fk ), n = 1,
n
X X ′ (a,b,c)
=: Ba,b,c n
m,n′ ϕ Jm,n′ fk . (6.103)
n′ =(n−1)+ a+b+c≤2

115
Thanks to the estimate (6.101b), we have that

A1,0,0;1
m,n L∞
+ ∂y A1,0,0;1
m,n L∞
≤ C(kvy−1 kL∞ , kvy − 1kL∞ Wy1,∞ ).
t

Combining this estimate with the expression (6.103), we have obtained (6.101c) for n ≥ 1.
Step # 3: Proof of (6.101d). We start by considering the case n ≥ 3. We apply the expression
(6.101a)j=2 to derive the following
2
ϕn ∂v [∂v , q n ](χm+n |k|m Γnk fk )
= ϕn ∂v A2,0,0;2 (2,0,0) n 2,0,0;2 (2,0,0) n 1,1,0;2 (1,1,0) n 1,1,0;2 (1,1,0)
m,n Jm,n fk + ϕ Am,n ∂v Jm,n fk + ϕ ∂v Am,n Jm,n fk + ϕ Am,n ∂v Jm,n fk
2∂v q (3,0,0)
= ϕn ∂v A2,0,0;2 (2,0,0) n 2,0,0;2
m,n Jm,n fk − ϕ Am,n J fk + ϕn ∂v A1,1,0;2 (1,1,0)
m,n Jm,n fk
m + n m,n
∂v q (2,1,0)
− ϕn A1,1,0;2
m,n J fk + ϕn A1,1,0;2 (1,2,0)
m,n Jm,n fk
m + n m,n
X
(a,b,0)
=: Cm,n fk .
a+b≤3

Thanks to the estimate (6.101b), we have that

1
X X
∂yℓ Aa,b,0;2
m,n ≤ C(kvy−1 kL∞ , kvy − 1kL∞ Wy2,∞ ).
L∞ t
ℓ=0 a+b=2,a>0

(a,b,0)
Hence the coefficients {Cm,n }n≥3 are consistent with (6.101d).
For the n ≤ 2, j = 1, 2 case, we invoke the second relation in (6.101a), and the bound (6.101b),
j
ϕn ∂v [∂v , q n ] (χm+n |k|m Γnk fk )
m
X n−1
X X ′ ′
 
n′ (0,b,c) 0,b,c;j n′ (0,b+1,c)
. hmi(m+n−1−m −n )+ (1+σ) |∂v A0,b,c;j
′
m ,n ′ |ϕ |J ′
m ,n ′ f k | + |A ′
m ,n ′ |ϕ |J ′
m ,n ′ f k |
m′ =(m−1)+ n′ =(n−2)+ b+c=j
m
X n−1
X X ′ ′ ′ (0,b,c)
. hmi(m+n−m −n )+ (1+σ) ϕn |Jm′ ,n′ fk |.
m′ =(m−1)+ n′ =(n−2)+ b+c=j+1

This is consistent with (6.101d) and hence complete the proof.

(a,b,c)
The following lemma help us to control various type of Jm,n -terms.
Lemma 6.17 (Induction lemma). Assume that σ∗ ≥ 10σ and fix an arbitrary pair (m, n) ∈
(a,b,c)
N × (N\{0}). The Jm,n fk can be estimated as follows
X
a2m,n (a,b,c)
kJm,n fk k2L2
a+b+c=m0 ,
a≥1
n−1
X  2σ
′ |m, n′ |! (m )
. λ2s(n−n ) a2m,n′ kJm,n0 ′ fk k2L2
|m, n|!
n′ =(n−m0 −2)+
m
X 1
X  2σ X
|m′ , n′ |!
+ 1n<m0 2s(|m,n|−|m′ ,n′ |) (0,b,c)
λ a2m′ ,n′ kJm′ ,n′ fk k2L2 . (6.104)
|m, n|!
m′ =(m−1)+ n′ =0 b+c≤m0

116
 (m ) (≤m )
Here, we recall the notations Jm,n0 fk , Jm,n 0 fk (3.8b), the index m0 ∈ {1, 2, 3} and highlight that
the implicit constant depends on kvy−1 kL∞t,y
, kvy kL∞ Wy2,∞ .
t

Proof. Step #1: Setup. We will derive the following bound which implies the estimate (6.104)
X
(a,b,c)
am,n kJm,n fk kL2
a+b+c=m0
m0 ≤n, a≥1
n−1
X  σ
s(n−n′ ) |m, n′ |! (m )
. λ am,n′ kJm,n0 ′ fk kL2
|m, n|!
n′ =(n−m0 −2)+
m
X 1
X  σ X
|m′ , n′ |!
+ 1n<m0 s(|m,n|−|m′ ,n′ |) (0,b,c)
λ am′ ,n′ kJm′ ,n′ fk kL2 , m0 ∈ {1, 2, 3}.
|m, n|!
m′ =(m−1)+ n′ =0 b+c≤m0
(6.105)
(a,b,c)
Thanks to the definition of Jm,n (3.6b) and the restriction of its indices (3.7), we observe that
the left hand side of (6.105) indeed contains all the nontrivial terms on the left hand side of the
relation (6.104). Now, the implication is a direct consequence of the fact that there are only at
most twenty terms in the sum (m0 ∈ {1, 2, 3}) and the Hölder inequality.
The plan now is to decompose the left hand side of (6.105) into manageable pieces. We will
commute the χm+n -cutoff out first and then identify four main contributions. We start by rewriting
(a,b,c)
the Jm,n fk term, for a + b + c = m0 , m0 ∈ {1, 2, 3}, as follows,
X  a
c m+n b
(6.105)L.H.S. = 1a+b+c≤n am,n |k| ∂v (χm+n q n Γnk |k|m fk )
a≥1 q L2
a+b+c=m0
 
X m+n a b n n m
≤ 1a+b+c≤n am,n χm+n |k|c ∂v (q Γk |k| fk )
a≥1 q L2
a+b+c=m0
 
X m+n a b
+ 1a+b+c≤n am,n |k|c [∂v , χm+n ](q n Γnk |k|m fk ) .
a≥1 q L2
a+b+c=m0

Here we apply a variant of the χm+n -estimate (2.5a)

ℓ
∂v χm+n ≤ C(kvy−1 kL∞
t,y
, kvy kL∞ Wy2,∞ )(m + n)ℓ(1+σ) χm+n−1 , σ∗ ≥ 10σ, ℓ ≤ m0 ≤ 3,
t

and Lemma 4.3 to conclude that

(6.105)L.H.S.
X  a
m+n b
. 1a+b+c≤n am,n 1supp(χm+n ) |k| c
∂v (q n Γnk |k|m fk )
a≥1 q L2
a+b+c=m0
b  
X X m + n a+ℓ b−ℓ n n m
+ 1a+b+c≤n am,n (m + n) 1supp(χm+n ) |k| σℓ
∂v (q Γk |k| fk ) c
a≥1 q
a+b+c=m0 ℓ=1 L2
X  a
m+n b
. am,n (m + n)3σ 1a+b+c≤n 1supp(χm+n ) |k|c ∂v (q n Γnk |k|m fk ) . (6.106)
a≥1 q L2
a+b+c=m0

This concludes the setup.

117
Step # 2: Estimates. We estimate the right hand side of (6.106) with simple commutator:

X b
X  a
m+n b−i1 i1
(6.106)R.H.S. . am,n (m + n) 3σ
1supp(χm+n ) |k| c
∂v (q n ) ∂v Γnk (|k|m fk )
q L2
a+b+c=m0 i1 =0
m0 ≤n, a≥1

X b
X i1
. am,n (m + n)3σ+a 1supp(χm+n ) |k|c q n−a−b+i1 Γa+b−i
k
1
∂v (Γkn−a−b+i1 |k|m fk )
L2
a+b+c=m0 i1 =0
m0 ≤n, a≥1

X b
X i
. am,n (m + n) 3σ+a
1supp(χm+n ) |k|c Γa+b−i
k
1
∂v (q n−a−b+i1 Γkn−a−b+i1 |k|m fk )
1

L2
a+b+c=m0 i1 =0
m0 ≤n, a≥1

X b
X h i
i1
+ am,n (m + n) 3σ+a
1supp(χm+n ) |k|c q n−a−b+i1 , Γa+b−i
k
1
∂v (Γkn−a−b+i1 |k|m fk )
L2
a+b+c=m0 i1 =0
m0 ≤n, a≥1

=: T1 + T2 .

For the T1 term, we apply the facts that 10σ ≤ σ∗ , ϕt . 1 and estimate as follows

X b a+b−i
X X1 (m + n − a − b + i1 )!σ
T1 . am,n−a−b+i1 (m + n)−σ∗ +3σ (ϕt)a+b−i1 −i2 λ(a+b−i1 )s
(m + n)!σ
a+b+c=m0 i1 =0 i2 =0
m0 ≤n, a≥1

× 1supp(χm+n ) |k|a+b+c−i1 −i2 ∂v i1 +i2 (χn−a−b+i1 |k|m q n−a−b+i1 Γkn−a−b+i1 fk )

L2
n−1
X X (m + n′ )!σ (n−n′ )s (0,b,c)
. am,n′ λ kJm,n′ fk kL2 .
(m + n)!σ
n′ =(n−m0 )+ b+c=m0

Here in the last line, we use the observation that the sum of the order of the x-derivative (a+ b+ c−
(0,b,c)
i1 − i2 ) and the order of the ∂v -derivative (i1 + i2 ) is m0 and they can be combined to Jm,n -form.
Then we relabel the summation indices to compactify the notation.
To treat the T2 -term, we invoked the Gevrey coefficient relation (4.29) and the product rule to
decompose it as follows

X b a+b−i
X X1  σ
|m, n − a − b + i1 |!
T2 . am,n−a−b+i1 λ(a+b−i1 )s (m + n)3σ−σ∗ ϕa+b−i1
|m, n|!
a+b+c=m0 i1 =0 i2 =0
m0 ≤n, a≥1
i1 +i2
× |k|a+b+c−i1 −i2 ta+b−i1 −i2 [q n−a−b+i1 , ∂v ](Γkn−a−b+i1 |k|m fk ) .
L2 (supp{χm+n })

Thanks to the assumption that 10σ ≤ σ∗ , the factor (m + n)3σ−σ∗ ≤ 1. Moreover, we observe
that the last factor can be estimated by the commutator estimate (6.100). Finally, since a > 0,
b − i1 ≥ 0, we have that n − a − b + i1 ≤ n − 1. Hence support{χm+n−a−b+i1 } ⊂ support{χm+n },
and the resulting Jm,n′ fk terms will have lower n′ -regularity (which guarantees that the induction
process will finish in finite steps). By combining all the observations above, we have that

T2

118
 X X X 1  |m, n − a − b + i1 |! σ
b a+b−i
. λ(a+b−i1 )s |k|m0 −i1 −i2 (ϕt)a+b−i1 −i2
|m, n|!
a+b+c=m0 i1 =0 i2 =0
m0 ≤n, a≥1
n−a−b+i
X 1  σ
|m, n′ |! ′ (≤i +i2 )
× a m,n′ λ(n−a−b+i1 −n )s Jm,n1′ fk
|m, n − a − b + i1 |! L2 (supp{χm+n })
n′ =(n−a−b+i1 −2)+
m
X n−a−b+i
X 1 −1  σ
|m′ , n′ |!
+ 1n−a−b+i1 <m0 am′ ,n′
|m, n − a − b + i1 |!
m′ =(m−1)+ n′ =0
!
X (0,b′ ,c′ )
|−|m′ ,n′ |)s
× λ(|m,n−a−b+i1 Jm′ ,n′ fk 2
L (supp{χm+n })
b′ +c′ ≤i1 +i2
n−1
X X |m, n′ |!σ (n−n′ )s (a,b,c)
. am,n′ λ kJm,n′ fk kL2
|m, n|!σ
n′ =(n−m0 −2)+ a+b+c=m0
m
X 1
X X  σ
|m′ , n′ |!
+ 1n<m0 s(|m,n|−|m′ ,n′ |) (0,b,c)
λ am′ ,n′ kJm′ ,n′ fk kL2 .
|m, n|!
m′ =(m−1)+ n′ =0 b+c≤m0

(≤i +i2 ) (≤m )

Here in the last line, we have used the fact that |k|m0 −i1 −i2 kJm,n1 fk kL2 ≤ kJm,n 0 fk kL2 . This
concludes the proof.

Lemma 6.18 (Translation Lemma). Assume that σ∗ ≥ 8σ.

a) Assume that m0 ∈ {1, 2}. There exists a universal C ≥ 1 such that the following two estimates
hold
X 2
a2m,n (a,b,c)
Jm,n fk 2 (6.107a)
L
a+b+c=m0
n
X X  2σ
s 2(n−n′ ) |m, n′ |! ′ ′ 2
. (Cλ ) a2m,n′ χm+n′ ∂yb |k|c (|k|m q n Γnk fk )
|m, n|! L2
n′ =0 b+c≤m0
m
X 1
X X  2σ
|m′ , n′ |! 2
+ 1n<m0 s 2(|m,n|−|m′ ,n′ |) ′ ′ ′
(Cλ ) a2m′ ,n′ χm′ +n′ ∂yb |k|c (|k|m q n Γnk fk ) .
|m, n|! L2
m′ =(m−1)+ n′ =0 b+c≤m0

Here the implicit constant depends on kvy−1 kL∞

t,y
, kvy − 1kL∞ 3 and is independent of m + n. More-
t Hy
s 1
over, if λ ≤ 2C , then
M
X X M
X X
2 2
a2m,n (a,b,c)
Jm,n fk . a2m,n χm+n ∂yb |k|c (|k|m q n Γnk fk ) , m0 ∈ {1, 2}.
L2 L2
m+n=0 a+b+c=m0 m+n=0 b+c≤m0
(6.107b)

Here the implicit constant depends on kvy−1 kL∞

t,y
, kvy − 1kL∞ 3 and is independent of M .
t Hy
b) Assume that m0 ∈ {1, 2, 3}. There exists a universal constant C ≥ 1 such that the following
estimate holds
X 2
a2m,n (a,b,c)
Jm,n fk 2 (6.108a)
L
a+b+c=m0
a≥1

119
 n−1
X X  2σ
′ |m, n′ |! (0,b,c) 2
. (Cλs )2(n−n ) a2m,n′ Jm,n′ fk
|m, n|! L2
n′ =0 b+c≤m0
m
X 1
X X  2σ
|m′ , n′ |! 2
+ 1n<m0
′ ′ (0,b,c)
(Cλs )2(|m,n|−|m ,n |) a2m′ ,n′ Jm′ ,n′ fk .
|m, n|! L2
m′ =(m−1)+ n′ =0 b+c≤m 0

Here the implicit constant depends on kvy−1 kL∞

t,y
, kvy − 1kL∞ 3 and is independent of m + n. More-
t Hy
over,
M
X X M
X −1 X
2 2
a2m,n Jm,n
(a,b,c)
fk . a2m,n Jm,n
(0,b,c)
fk , m0 ∈ {1, 2,(6.108b)
3}.
L2 L2
m+n=0 a+b+c=m0 m+n=0 b+c≤m0
a≥1

Here the implicit constant depends on kvy−1 kL∞

t,y
, kvy − 1kL∞ 3 and is independent of M .
t Hy

Remark 6.19. Here we highlight that the estimates (6.107) cost more regularity than (6.108).
The (6.107) estimates are used to convert the J-estimates to the diffusive bounds D. On the other
(a,b,c) (E)
hand, the (6.108) estimates are used to treat various {Jm,n φk }a6=0 terms in the elliptic section.
Proof. To make the flow of the proof more accessible, we organize the material in three steps. In
Step # 1, we prove estimate (6.108a) because it is a direct consequence of the previous induction
lemma and combinatorics; In Step # 2, we use the trick in Step # 1 to handle more complicated
situation (6.107a), (6.107b); In the final Step # 3, we will comment that the remaining estimate
(6.108b) is similar to the treatment in Step # 2.
Step # 1: Proof of the estimate (6.108a). We plan to apply the induction relation (6.104).
For presentation purposes, we first consider the case where n ≥ 5(m0 + 2) to avoid complicated
lower order terms that might appear in the first few steps of induction. The n < 5(m0 + 2) case will
be naturally handled when we complete the induction argument. We apply the relation (6.104),
n−1  
2
X
(a,b,c) 2
X
2(n−n1 )s 2 |m, n1 |! 2σ (m0 )
am,n kJm,n fk kL2 . λ am,n1 kJm,n1 fk k2L2
|m, n|!
a+b+c=m0 , n1 =(n−m0 −2)+
a≥1
X  2σ X
|m, n1 |!
. λ2(n−n1 )s a2m,n1 (0,b,c)
kJm,n 1
fk k2L2
n1 =n−1
|m, n|!
b+c=m0
n−2
X  2σ X
|m, n1 |!
+ λ2(n−n1 )s a2m,n1 (0,b,c)
kJm,n 1
fk k2L2
|m, n|!
n1 =(n−m0 −2)+ b+c=m0
n−1
X nX
1 −1  2σ
|m, n1 |! |m, n2 |!
+ λ2(n−n2 )s a2m,n2 (m0 )
kJm,n f k2 .
2 k L2
|m, n|! |m, n1 |!
n1 =(n−m0 −2)+ n2 =(n1 −m0 −2)+

Here in the second inequality, we single out the terms with n1 = n − 1, identify the J (a,b,c) -terms
with a 6= 0 and apply the induction estimate (6.104). First, we note that the n1 = n − 1 will no
longer appear in the latter expansion because n2 ends at n1 − 1 ≤ n − 2. Next we observe that
the |m, n1 |! cancels out in the last term. Hence, the coefficients in the sum only “remembers” the
starting position (m, n) and the ending position (m, n2 ). With these observations, we do another
induction and end up with
X
a2m,n (a,b,c)
kJm,n fk k2L2
a+b+c=m0 ,
a≥1

120
 n−1
X  2σ X
|m, ℓ|! (0,b,c)
. Cn−ℓ λ2(n−ℓ)s a2m,ℓ kJm,ℓ fk k2L2
|m, n|!
ℓ=n−2 b+c=m0
n−3
X X
|m, n1 |!2σ
+ λ2(n−n1 )s a2m,n1 (0,b,c)
kJm,n fk k2L2
|m, n|!2σ 1
n1 =(n−m0 −2)+ b+c=m0
n−1
X nX
1 −1  2σ X
|m, n2 |!
+ 1 n2 6=n−2 λ
2(n−n2 )s 2
am,n2 (0,b,c)
kJm,n 2
fk k2L2
|m, n|!
n1 =(n−m0 −2)+ n2 =(n1 −m0 −2)+ b+c=m0
n−1
X nX
1 −1 nX
2 −1  2σ
|m, n3 |!
+ λ2(n−n3 )s a2m,n3 (m0 )
kJm,n f k2 .
3 k L2
|m, n|!
n1 =(n−m0 −2)+ n2 =(n1 −m0 −2)+ n3 =(n2 −m0 −2)+

This process will be stopped after ≈ n steps. Here the constant C = C(m0 ) is introduced to account
(0,b,c)
for the repetition of the Jm,ℓ fk terms. For example, in the first sum, the term ℓ = n − 2 appears
at most m0 + 3 times and it will not appear in the latter induction expansion. For other terms
(0,b,c)
Jm,ℓ , the total number of repetition increases as the distance |m, n| − |m, ℓ| increases. Thanks
to the iterative sum structure in the last term, the number of repetition is exponential in terms of
|m, n| − |m, ℓ|. Hence,
n−1
X X  2σ
|m, ℓ|!
a2m,n 1a≥1 kJm,n
(m0 ) (0,b,c)
fk k2L2 . Cn−ℓ λ2(n−ℓ)s a2m,ℓ kJm,ℓ fk k2L2
|m, n|!
ℓ=0 b+c=m0
m
X 1
X X  2σ
|m,n|−|m′ ,ℓ| 2s(|m,n|−|m′ ,ℓ|) |m′ , ℓ|! (0,b,c) 2
+ C λ a2m′ ,ℓ Jm′ ,ℓ fk .
|m, n|! L2
m′ =(m−1)+ ℓ=0 b+c≤m0

(0,b,c)
Now all the terms appeared in the sum only have Jm,n fk components and are consistent with
(6.108a). Since the n < 5(m0 + 2) case can be treated in an identical fashion, this concludes the
proof of (6.108a).
Step # 2: Proof of the estimates (6.107a), (6.107b). Thanks to the induction relation
(6.104), we can first focus on the case where a = 0, b + c = m0 because the a > 0 case can be
reduced to this case:
X
(0,b,c)
am,n kJm,n fk kL2
b+c=m0
X b X b
≤am,n kχm+n ∂v |k|c (q n Γnk |k|m fk )kL2 + am,n k|k|c [∂v , χm+n ](q n Γnk |k|m fk )kL2
b+c=m0 b+c=m0
X b c
.am,n kχm+n ∂v |k| (q n Γnk |k|m fk )kL2
b+c=m0
b  X
X X i1    
b i1 i1 b−i1 i1 n−i1 i2
+ am,n |k|c ∂v χm+n ∂v (q q ∂v (ikt)i1 −i2 Γn−i
k
1
|k|m fk )
i1 i2 L2
b+c=m0 i1 =1 i2 =0
X
.am,n kχm+n ∂yb |k|c (q n Γnk |k|m fk )kL2
b+c≤m0
i1
b X
X X am,n−i1 ϕi1 ti1 −i2 λi1 s c+i1 −i2 b−i1 i1 i2 n−i1 n−i1
+ i σ
|k| kχm+n−i1 ∂v (q ∂v (q Γk |k|m fk ))kL2
(m + n) 1 ∗
b+c=m0 i1 =1 i2 =0

121
 i1
b X
X X am,n−i1 ϕi1 ti1 −i2 λi1 s c+i1 −i2 b−i1 i1 i2
+ |k| k∂v (q [∂v , q n−i1 ]Γn−i
k
1
|k|m fk ))kL2 (supp{χm+n })
(m + n)i1 σ∗
b+c=m0 i1 =1 i2 =0
=:T1 + T2 + T3 . (6.109)

The T1 term is consistent with the (6.107a). For the T2 -term, we estimate it as follows

X X i1 b−i
b X X1 am,n−i λi1 s i2 +i3 n−i1 n−i1
T2 . 1
i σ
kχm+n−i1 |k|c+i1 −i2 ∂v (q Γk |k|m fk ))kL2
(m + n) 1 ∗
b+c=m0 i1 =1 i2 =0 i3 =0
n−1
X X  σ∗
(n−n′ )s |m, n′ |! ′ ′
. a m,n′ λ kχm+n′ ∂yb |k|c (q n Γnk |k|m fk )kL2 . (6.110)
|m, n|!
n′ =(n−m0 )+ b+c≤m0

Next we estimate the T3 -term using the product rule,

X X b X i1 b−i
X1 am,n−i ϕi2 λi1 s
T3 . 1
|k|c+i1 −i2
(m + n)i1 σ∗
b+c=m0 i1 =1 i2 =0 i3 =0
h i 
(b−i1 −i3 ) i2
× q (i1 −i3 )+ ∂v ∂v , q n−i1 |k|m Γn−i
k
1
f k .
L2 (supp χm+n )

Here we note that (b − i1 − i3 ) + i2 ≤ b ≤ m0 ≤ 2. Hence we invoke the estimates (6.101a)j=2 ,

(6.101c) to obtain that
n−1
X X b
am,n−i1 λi1 s (≤m0 )
T3 . i σ
Jm,n−i f
1 k
(m + n) 1 ∗ L2 (supp χm+n )
n′ =(n−m0 −1)+ i1 =1
X m  σ∗ X
m′ !
+ 1n<2m0 (|m,n|−m′ )s ′
am′ ,0 λ χm′ ∂yb |k|c (|k|m fk ) (6.111)
.
′
|m, n|! L2 (supp χm+n )
m =(m−1)+ b+c≤m0

Hence, combining the decomposition (6.109) for the {a = 0, b + c = m0 } case and the estimates
(6.110), (6.111), we further apply the induction relation (6.104) to take the {a 6= 0, a + b + c = m0 }
cases into account and end up with the following
X
(a,b,c)
am,n kJm,n fk kL2
a+b+c=m0
Xn X  σ∗
′ |m, n′ |! ′ ′
. am,n′ λ(n−n )s kχm+n′ ∂yb |k|c (q n Γnk |k|m fk )kL2
|m, n|!
n′ =(n−m0 −1)+ b+c≤m0
n−1  
X X |m, n′ |! σ∗ (a,b,c)
+ am,n′ λ (n−n′ )s
kJm,n′ fk kL2 1a≥1
|m, n|!
n′ =(n−m0 −1)+ a+b+c≤m0
Xm  σ∗ X
m′ !
+ 1n<2m0
′ ′
am′ ,0 λ(|m,n|−m )s χm′ ∂yb |k|c (|k|m fk ) .
′
|m, n|! L2
m =(m−1)+ b+c≤m0

Now we apply the induction estimates (6.104) on the second term in the expression to translate all
(0,b,c) (0,b,c)
the terms in to the form Jm′ ,n′ fk . We further note that each distinct Jm,n fk can only appear at
most exponentially in m + n many times during the inductive expansion. Now we apply the Hölder

122
inequality and the combinatorial estimate (5.34) to obtain (6.107a). To derive (6.107b), we sum
over both side of the expression (6.107a) and apply the Fubini
M
X X
a2m,n (a,b,c)
kJm,n fk k2L2
m+n=0 a+b+c=m0
M M
X X −m X
n X  2σ
s 2(n−ℓ) |m, ℓ|! 2
≤C (Cλ ) a2m,ℓ χm+ℓ ∂yb |k|c (|k|m q ℓ Γℓk fk )
m=0 n=0 ℓ=0 b+c≤m0
|m, n|! L2

M M
X X −m m
X X  2σ
s 2(|m,n|−m′ ) m′ ! ′ 2
+C (Cλ ) a2m′ ,0 χm′ ∂yb |k|c (|k|m fk )
m=0 n=0 m′ =(m−1)+ b+c≤m0
|m, n|! L2

M M
X X −m M
X −m X
2
≤C (Cλs )2(n−ℓ) a2m,ℓ χm+ℓ ∂yb |k|c (|k|m q ℓ Γℓk fk )
L2
m=0 ℓ=0 n=ℓ b+c≤m0
M
X X 2
+C a2m,0 χm ∂yb |k|c (|k|m fk )
L2
m=0 b+c≤m0
M
X X 2
≤C a2m,ℓ χm+ℓ ∂yb |k|c (|k|m q ℓ Γℓk fk ) .
L2
m+ℓ=0 b+c≤m0

This is (6.107b).
Step # 3: Proof of the estimate (6.108b). Now the derivation of (6.108b) is similar to Step
# 2 and we omit the details.

As an application of the ICC method, we present two lemmas that provide bounds on the
coordinate quantity Z = vy2 − 1.
Lemma 6.20. Assume σ∗ ≥ 3σ. The following estimate holds for Z = vy2 − 1 and n ≥ 1,
n+1 X
k1suppχn J0,n−1 ZkL2 ; (6.112)
(0,1,0) (a,b,0)
B0,n kχn q n ∂v ZkL2 .B0,n kJ0,n ZkL2 + B0,n−1 n−σ∗
a+b=1
n+2 X
k1suppχn J0,n−1 ZkL2
n (0,2,0) −σ (a,b,0)
B0,n kχn q ∂v ZkL2 .B0,n kJ0,n ZkL2 + B0,n−1 n
a+b=2
2
X ℓ
+ 1n∈{1,2} k1suppχ1 ∂v ZkL2 . (6.113)
ℓ=0
| {z }
P
0≤b≤2 k1suppχ1 J0,0
(0,b,0)
= ZkL2

Proof. To derive the inequality (6.112),

n+1
B0,n kχn q n ∂v ZkL2
n n−1
.B0,n k∂v (χn q n ∂v Z)kL2 + B0,n k∂v (χn q n )∂v (χn−1 ∂v Z)kL2
n−1
k1suppχn q
(0,1,0) 1+σ n−1
.B0,n kJ0,n ZkL2 + B0,n n ∂v (χn−1 ∂v Z)kL2
λs B0,n−1  
n−1
(0,1,0)
.B0,n kJ0,n ZkL2 + 1suppχn J0,n−1
(0,1,0)
Z 2 + 1suppχn [q n−1 , ∂v ](χn−1 ∂v Z)
nσ∗ L L2
λs B0,n−1 X
k1suppχn J0,n−1 ZkL2 .
(0,1,0) (a,b,0)
.B0,n kJ0,n ZkL2 +
nσ∗
a+b=1

123
To derive the inequality (6.113), we expand the expression and obtain that,
n+2
B0,n kχn q n ∂v ZkL2
2 n 2 n−1
.B0,n k∂v (χn q n ∂v Z)kL2 + B0,n k∂v (χn q n )∂v (χn−1 ∂v Z)kL2
2 n−1
+ B0,n k∂v (χn q n )∂v (χn−1 ∂v Z)kL2
2 n−1
+ B0,n n1+σ k1suppχn q n−1 ∂v (χn−1 ∂v
(0,2,0)
.B0,n kJ0,n ZkL2 Z)kL2
n−1
+ B0,n n2+2σ k1suppχn q (n−2)+ ∂v (χn−1 ∂v Z)kL2
 
2 n−1
.B0,n kJ0,n ZkL2 + B0,n n1+σ k1suppχn J0,n−1 ZkL2 + 1suppχn [q n−1 , ∂v ](χn−1 ∂v
(0,2,0) (0,2,0)
Z)
L2
n−1
+ B0,n n2+2σ k1suppχn q (n−2)+ ∂v (χn−1 ∂v Z)kL2 =: T1 + T2 + T3 . (6.114)
The first term is consistent with the result (6.113). Then we can estimate the last term T3 in
(6.114) with the relation (2.1), and the facts that σ∗ ≥ 3σ to as follows,
λs B0,n−1 n  n−1 n−1

2
T3 .1n≥3 σ
1 suppχ n ∂v χ n−1 q ∂v Z + 1n=2 B0,0 k1suppχn ∂v ZkL2
n q L2
+ 1n=1 B0,0 k1suppχn ∂v ZkL2
λs B0,n−1 2
=1n≥3 1suppχn J0,n−1 + 1n=2 B0,0 k1suppχn ∂v ZkL2 + 1n=1 B0,0 k1suppχn ∂v ZkL2 .
(1,1,0)
σ
Z
n L2

Now we invoke the relation (6.96) and the estimate (6.95a) to obtain a bound for the second term
λs B0,n−1 n−1
k1suppχn J0,n−1 kL2 + B0,n n2+σ k1suppχn q (n−2)+ ∂v (χn−1 ∂v
(0,2,0)
T2 . Z)kL2
nσ
λs B0,n−1 n−1
+ σ
(n − 1)2 k1suppχn q (n−3)+ −(n−1) (χn−1 q n−1 ∂v Z)kL2 .
n
We observe that the second term can be estimated in a similar fashion as T3 in (6.114). Hence,
2
λs B0,n−1 λs B0,n−1 X X ℓ
T2 . k1 suppχ n J
(0,2,0)
0,n−1 kL 2 + 1n≥3 1 (a,b,0)
suppχn J0,n−1 Z + 1n∈{1,2} k1suppχ1 ∂v ZkL2 .
nσ nσ L2
a+b=2 ℓ=0

Lemma 6.21. Assume (2.22). The following estimates hold

∞ X
X ∞ X
X 1
2 2
1supp{eχ1 } J0,n
2 (a,b,0) (0,b,0) −1/9 (γ) (α)
B0,n ϕ2n J0,n χ1 (vy2 − 1))
(e . 2
B0,n ϕ2n H . e−ν (EH + EH );
2 2
n=0 a+b=1 n=0 b=0
(6.115)
∞
X X ∞ X
X 2 X
2
1supp{eχ1 } J0,n
2 (a,b,0) (0,b,0) −1/9 (ι) (ι)
B0,n ϕ2n kJ0,n χ1 (vy2 − 1))k22 .
(e 2
B0,n ϕ2n H . e−ν (EH + DH ).
2
n=0 a+b=2 n=0 b=0 ι=α,γ
(6.116)
Proof. First of all, we observe that the last inequalities in (6.115) and (6.116) are natural conse-
(α) (α)
quences of the definitions of EH (2.19) and DH (2.20). We can explicitly estimate the left hand
side of (6.115) with the translation relation (6.107b) as follows
∞ X
X 1   2
2 b n
L.H.S of (6.115) . B0,n ϕ2n χn ∂v q n ∂v (e
χ1 (vy2 − 1))
2
n=0 b=0

124
We recall that vy2 − 1 = H(H + 2). Hence to prove (6.115), it is enough to prove
∞ X
X 1 ∞ X
X 1
b n 2 2
2
B0,n ϕ2n χn ∂v (q n ∂v (e
χ1 H(H + 2))) . 2
B0,n ϕ2n 1supp χe1 J0,n
(0,b,0)
H (6.117)
.
2 L2
n=0 b=0 n=0 b=0

Similarly, by the translation estimate (6.107b), the following estimate guarantees (6.116)
∞
X ∞ X
X 2
2 n
ϕ2n k1supp
2 (0,b,0)
B0,n ϕ2n k∂v (χn q n ∂v (e
χ1 H(H + 2)))k22 . 2
B0,n e1 J0,n
χ Hk2L2 . (6.118)
n=0 n=0 b=0

We comment that the estimates (6.117) and (6.118) are simple exercises in Gevrey product rule
and we omit further details for the sake of brevity.

To conclude the subsection, we present the last two technical lemma.

Lemma 6.22 (Gevrey Product Rule). Consider two Gevrey smooth functions f, g ∈ Gr;Λ ([−1, 1]).
Let
Λm+n
B m,n := .
|m, n|!1/r
The following two product estimates hold
∞
X 2
B m,n ϕ2n kq n ∂xm Γn (f g)k2L2
m+n=0
∞
! ∞
!
X 2 X 2
. B m,n ϕ2n kq n ∂xm Γn f k2L2 B m,n c2n ϕ2n k∂y (q n ∂xm Γn g)k2L2
m+n=0 m+n=0
∞
! ∞
!
X 2 X 2
+ B m,n c2n ϕ2n k∂y (q n ∂xm Γn f )k2L2 B m,n ϕ2n kq n ∂xm Γn gk2L2 ; (6.119)
m+n=0 m+n=0
∞
X 2
B m,n ϕ2n k∂y (q n ∂xm Γn (f g))k2L2
m+n=0
∞
! ∞
!
X 2 X 2
. B m,n ϕ2n k∂y (q n ∂xm Γn f )k2L2 B m,n ϕ2n k∂y (q n ∂xm Γn g)k2L2 . (6.120)
m+n=0 m+n=0

Here 0 < c = c(r) < 1.

In general, for 1/p1 + 1/q1 = 1/p2 + 1/q2 = 1/r,
∞
X 2
B m,n ϕ2n kq n ∂xm Γn (f g)k2Lr
m+n=0
∞
! ∞
!
X 2 X 2
. B m,n ϕ2n kq n ∂xm Γn f k2Lp1 B m,n c2n ϕ2n kq n ∂xm Γn gk2Lq1
m+n=0 m+n=0
∞
! ∞
!
X 2 X 2
+ B m,n c2n ϕ2n kq n ∂xm Γn f k2Lp2 B m,n ϕ2n kq n ∂xm Γn gk2Lq2 . (6.121)
m+n=0 m+n=0

Here 0 < c = c(r) < 1.

125
Remark 6.23. We comment that the lemma has another form in our companion paper [5], see,
e.g., Lemma 6.27.

Proof. Throughout the proof, we use the shorthand notation |m, n| := m+n. We recall the relation
    
|m, n| m n
′ ′
≥ ′
, m′ ≤ m, n′ ≤ n.
|m , n | m n′
1
Now we prove the estimate (6.121). To this end, we consider a finite sum (Here 0 < ζ ≤ 16 (1/r −1))
and apply the combinatorial bound (6.88),
M
X 2
B m,n ϕ2n k∂xm q n Γn (f g)k2Lr
m+n=0
M M −m m X n     !2
X X Λ2m+2n 2n X m 1+ζ−ζ n 1+ζ−ζ m′ n′ n′ m−m′ n−n′ n−n′
. ϕ k∂x q Γ f ∂x q Γ gkLr
m=0 n=0
|m, n|!2/r
′ ′
m′ n′
m =0 n =0
M M −m    
X X X Λ2m+2n 2n m 2+2ζ n 2+2ζ m′ n′ n′ 2 ′ ′ ′
. 2/r
ϕ ′ ′
k∂x q Γ f kLp1 k∂xm−m q n−n Γn−n gk2Lq1
|m, n|! m n
m=0 n=0 m′ ≤m, n′ ≤n
|m′ ,n′ |≤1/2|m,n|
M M −m    
X X X Λ2m+2n 2n m 2+2ζ n 2+2ζ m′ n′ n′ 2 ′ ′ ′
+ 2/r
ϕ ′ ′
k∂x q Γ f kLp2 k∂xm−m q n−n Γn−n gk2Lq2
|m, n|! m n
m=0 n=0 m′ ≤m, n′ ≤n
|m′ ,n′ |>1/2|m,n|

=: T1 + T2 .

We focus on the T1 term, and treat the T2 term in a similar fashion.

We further decompose the T1 term as follows:
M |m,n| 2  
X X B m,n |m, n| 2+2ζ
T1 ≤ 1 m′ ≤m, n′ ≤n 2 2
B m−m′ ,n−n′ B m′ ,n′ |m′ , n′ |
|m,n|=0 |m′ ,n′ |=0
2 ′ ′ ′ ′ 2 ′ ′ ′ ′
× B m′ ,n′ ϕ2n k∂xm q n Γn f k2Lp1 B m−m′ ,n−n′ ϕ2n−2n k∂xm−m q n−n Γn−n gk2Lq1
 
XM min{N,⌊|m,n|/8⌋}
X ⌈|m,n|/8⌉
X |m,n|
X  −2(1/r−1)+2ζ
.  + +  1m′ ≤m, n′ ≤n |m, n|
′ ′ ′ ′
|m′ , n′ |
|m,n|=0 |m ,n |=0 min{N,⌊|m,n|/8⌋}+1 |m ,n |=⌈|m,n|/8⌉
2 ′ ′ ′ ′ 2 ′ ′ ′ ′
× B m′ ,n′ ϕ2n k∂xm q n Γn f k2Lp1 B m−m′ ,n−n′ ϕ2n−2n k∂xm−m q n−n Γn−n gk2Lq1
=:T11 + T12 + T13 .

We will pick a constant N = N (r) large. The estimate of the T11 -term is direct:
M
X N
X
1m′ ≤m, n′ ≤n B 2m′ ,n′ c2n ϕ2n k∂xm q n Γn f k2Lp1
′ ′ ′ ′ ′
T11 ≤C(N, c)
|m,n|=0 |m′ ,n′ |=0
2 ′ ′ ′ ′
× B m−m′ ,n−n′ ϕ2(n−n ) k∂xm−m q n−n Γn−n gk2Lq2
  
M
X M
X
2 2
≤C(N, c)  B m,n c2n ϕ2n k∂xm q n Γn f k2Lp1   B m,n ϕ2n k∂xm q n Γn gk2Lq1  .
|m,n|=0 |m,n|=0

126
Here in the last line, we commute the summation order. Now we estimate the T12 term. We note
the following relation
 
|m, n| −1 |m′ , n′ |!
1 1 ′ ′
′ ′
|m′ ,n′ |≤|m,n|/8 = ≤ 7−|m ,n | .
|m′ , n′ | |m, n|(|m, n| − 1) · · · (|m, n| − |m′ , n′ | + 1) |m ,n |≤|m,n|/8

As a result, we obtain

M ⌊|m,n|/8⌋
X X
1m′ ≤m, n′ ≤n 7−2|m ,n |(1/r−1−ζ) B 2m′ ,n′ ϕ2n k∂xm q n Γn f k2Lp1
′ ′ ′ ′ ′ ′
T12 .
|m,n|=0 |m′ ,n′ |=0
2 ′ ′ ′ ′
× B m−m′ ,n−n′ ϕ2(n−n ) k∂xm−m q n−n Γn−n gk2Lq1
  
XM M
X
2 2
. B m,n c2n ϕ2n k∂xm q n Γn f k2Lp1   B m,n ϕ2n k∂xm q n Γn gk2Lq1  .
|m,n|=0 |m,n|=0

Finally, we estimate T13 . Note the Stirling’s formula |m, n|! ∼ |m, n||m,n|+1/2 e−|m,n| implies that
for |m, n|, |m′ , n′ |, |m, n| − |m′ , n′ | ≥ N ≫ 1, 1/r > 1
  ′ ′ ′ ′
!1/r−1−ζ
|m, n| −1/r+1+ζ |m′ , n′ ||m ,n |+1/2 (|m, n| − |m′ , n′ |)|m,n|−|m ,n |+1/2
.
|m′ , n′ | |m, n||m,n|+1/2
 |m,n|(1/r−1−ζ)   |m,n|(1/r−1−ζ)
7 p 1/r−1−ζ 7 2
. ( |m, n|) . , |m′ , n′ |, ||m, n| − |m′ , n′ || ≤ 7|m, n|/8.
8 8

Hence,
  
M
X M
X
2 2
T13 .  B m,n c2n ϕ2n k∂xm q n Γn f k2Lp1   B m,n ϕ2n k∂xm q n Γn gk2Lq1  .
|m,n|=0 |m,n|=0

Apply similar argument to the T2 term yields the similar bounds. Combining the estimates above
and taking M → ∞ yield (6.121). The estimate (6.119) can be derived using (6.121) and Sobolev
embedding. The estimate for the (6.120) is similar.
1
Lemma 6.24. Consider s ∈ (1, 5/4), t ∈ [0, ν −1/3−η ], η ≤ 78 . Then,
∞
X e2(m+n)es X∞  
λ m n n 2 2 m n n 2 1 −2/3+η
2e
s
k|k| q Γk fk kL2 ([−1/2,1/2]c ) . am,n (t)kχm+n |k| q Γk fk kL2 exp ν ,
m+n=0
((m + n)!) m+n=0
C

1 1 34 e = C(e
< = , λ s, s, C, λ).
2 se 19(2s + 1)
34 68
Note that if s ≤ 5/4, we have that 1/e
s= 19(2 54 +1)
≥ 133 > 0.51 > 1/2.

Proof. On the time interval t ∈ [0, ν −1/3−η ], we have the following estimate:
     
1 −2/3+η 2n+2 1 −2/3+η 2(m+n)+2 1 1 −2/3+η
exp ν ϕ ≥ exp ν ϕ ≥ exp ν (1 + t2 )−m−n−1
C C C C

127
  
1 1 −2/3+η 1 1 1 1
≥ exp ν
C 2C N ! (2C)N ν ( 23 −η)N (1 + t2 )m+n 1 + t2
 
1 1 −2/3+η 1 1 2 2
≥ exp ν N m+n
ν −( 3 −η)N +( 3 +2η)(m+n) .
C 4C N ! (2C) 2
j k
By setting N = 2/3+2η
2/3−η (m + n) and G = 2C, we have
 
1 −2/3+η 1
exp ν ϕ2n+2 ≥ j
2/3+2η
k j k.
C (m+n) 2/3+2η
CG 2/3−η
2m+n 2/3−η (m + n) !

By invoking the facts that χm+n ≡ 1, ∀ m + n ∈ N, |y| ∈ [−1/2, 1/2]c , we obtain the following
∞
X  
λ2(m+n)s ϕ2n+2 m n n 2 1 −2/3+η
kχm+n |k| q Γk fk kL2 exp ν
((m + n)!)2s C
m+n=0
∞
X λ2(m+n)s m n 2
& j
2/3+2η
k j k kχm+n |k| Γk fk kL2
(m+n) m+n 2/3+2η
n+m=0 G 2/3−η
2 ((m + n)!)2s 2/3−η (m + n) !
∞
X λ2(m+n)s
k 2s+1 k1|y|≥1/2 |k| q Γk fk kL2 .
m n n 2
& √ j (6.122)
2/3+2η
n+m=0 ( 2G)2(m+n)
2/3−η (m + n) !

To estimate the right hand side of (6.122), we use the Gamma function Γ(n) = (n − 1)!, n∈
N\{0} and the log convexity of the Gamma function:

Γ(θx1 + (1 − θ)x2 ) ≤ Γ(x1 )θ Γ(x2 )1−θ , θ ∈ [0, 1], x1 , x2 > 0.

1 2/3+2η 18
Since η = 78 , so 2/3−η = 17 , and
    
2/3 + 2η 18
(m + n) ! =Γ (m + n) + 1
2/3 − η 17
2(m + n) − ⌊ 18
17 (m + n)⌋ 16
≤Γ(m + n + 1)θ Γ(2(m + n) + 1)1−θ , θ= ≥ .
m+n 17
Now we estimate the Γ(2(m + n) + 1) = (2(m + n))!:
(m+n) (m+n−1)
Y Y
(2(m + n))! = (2ℓ1 ) (2ℓ2 + 1) ≤ 22(m+n) ((m + n)!)2 .
ℓ1 =1 ℓ2 =0

Hence,
 
2/3 + 2η 1
(m + n) ! ≤((m + n)!)θ [(2(m + n))!]1−θ ≤ (m + n)![(2(m + n))!] 17
2/3 − η
2 19
≤2(m+n) 17 ((m + n)!) 17 .

Therefore, the right hand side of (6.122) has the following lower bound,
∞
X λ2(m+n)s
k 2s+1 k1|y|≥1/2 |k| q Γk fk kL2
m n n 2
√ j
2(m+n) 2/3+2η
m+n=0 ( 2G)
2/3−η (m + n) !

128
 ∞
X λ2(m+n)s
& √ 2 19 k1|y|≥1/2 |k|m q n Γnk fk k2L2
2(m+n) (m+n)(2s+1) (2s+1)
m+n=0 ( 2G) 2 17 ((m + n)!) 17

∞
X λ2(m+n)s
& 19 k1|y|≥1/2 |k|m q n Γnk fk k2L2 .
(2s+1)
m+n=0 (2G)2(m+n) ((m + n)!) 17

Hence the resulting Gevrey index se and radius of analyticity are

19 1 34 e = λ(λ,
e s, se, C),
2e
s= (2s + 1) ⇒ = , λ
17 se 19(2s + 1)

which completes the proof.

Acknowledgments: JB was supported by NSF Award DMS-2108633. JB would also like to

thank Ryan Arbon for helpful discussions. The work of SH is supported in part by NSF grants
DMS 2006660, DMS 2304392, DMS 2006372. SH would like to thank Ruth Luo for teaching him
many fancy combinatoric tricks. SH would also like to thank Yiyue Zhang for many suggestion on
Lie Algebra. The work of SI is supported by NSF DMS-2306528 and a UC Davis Society of Hellman
Fellowship award. The work of FW is supported by the National Natural Science Foundation of
China (No. 12101396, 12161141004, and 12331008).

References
[1] Y. Almog and B. Helffer, On the stability of laminar flows between plates, Archive for Rational Mechanics and
Analysis 241 (2021), no. 3, 1281–1401.
[2] J. Bedrossian, M. Coti Zelati, and M. Dolce, Taylor dispersion and phase mixing in the non-cutoff boltzmann
equation on the whole space, To appear in Proc. London Math. Soc. (2022).
[3] J. Bedrossian and S. He, Inviscid damping and enhanced dissipation of the boundary layer for 2d navier–stokes
linearized around couette flow in a channel, Communications in Mathematical Physics 379 (2020), no. 1, 177–226.
[4] J. Bedrossian, S. He, S. Iyer, and F. Wang, Stability threshold of nearly-Couette shear flows with Navier boundary
conditions in 2d, arXiv:2311.00141 (2023).
[5] J. Bedrossian, S. He, S. Iyer, and F. Wang, Uniform Inviscid Damping and Inviscid Limit of the 2d Navier-Stokes
equation with Navier Boundary Conditions, arXiv (2024).
[6] J. Bedrossian and N. Masmoudi, Inviscid damping and the asymptotic stability of planar shear flows in the 2d
euler equations, Publications mathématiques de l’IHÉS 122 (2015), no. 1, 195–300.
[7] J. Bedrossian, N. Masmoudi, and V. Vicol, Enhanced dissipation and inviscid damping in the inviscid limit of the
Navier–Stokes equations near the two dimensional Couette flow, Archive for Rational Mechanics and Analysis
219 (2016), no. 3, 1087–1159.
[8] R. Beekie, S. Chen, and H. Jia, Uniform vorticity depletion and inviscid damping for periodic shear flows in the
high Reynolds number regime, arXiv preprint arXiv:2403.13104 (2024).
[9] J. Benameur, Smoothing effects for navier-stokes equations, 2008.
[10] S. Chaturvedi, J. Luk, and T. Nguyen, The Vlasov-Poisson-Landau system in the weakly collisional regime, J.
Amer. Math. Soc. 36 (2023), no. 4, 1103–1189. MR4618956
[11] J.-Y. Chemin, Le système de navier-stokes incompressible soixante dix ans après jean leray, Séminaire et Congrès
9 (200401).
[12] Q. Chen, T. Li, D. Wei, and Z. Zhang, Transition threshold for the 2-d couette flow in a finite channel, Archive
for rational mechanics and analysis 238 (2020), no. 1, 125–183.
[13] Q. Chen, D. Wei, and Z. Zhang, Linear inviscid damping and enhanced dissipation for monotone shear flows,
Communications in Mathematical Physics 400 (2023), no. 1, 215–276.
[14] Q. Chen and Z. Li, Enhanced dissipation and transition threshold for the poiseuille-couette flow, Journal of
Differential Equations 386 (2024), 404–434.

129
[15] M. Coti Zelati, Stable mixing estimates in the infinite Péclet number limit, J. Funct. Anal. 279 (2020), no. 4,
108562, 25. MR4095800
[16] M. Coti Zelati, T. M Elgindi, and K. Widmayer, Enhanced dissipation in the navier–stokes equations near the
poiseuille flow, Communications in Mathematical Physics 378 (2020), no. 2, 987–1010.
[17] Y. Deng and N. Masmoudi, Long-time instability of the Couette flow in low Gevrey spaces, Communications on
Pure and Applied Mathematics 76 (2023), no. 10, 2804–2887.
[18] E. Grenier, T. T Nguyen, F. Rousset, and A. Soffer, Linear inviscid damping and enhanced viscous dissipation
of shear flows by using the conjugate operator method, Journal of Functional Analysis 278 (2020), no. 3, 108339.
[19] A. D Ionescu, S. Iyer, and H. Jia, Linear inviscid damping and vorticity depletion for non-monotonic shear flows,
arXiv preprint arXiv:2301.00288 (2022).
[20] A. D Ionescu and H. Jia, Inviscid damping near the couette flow in a channel, Communications in Mathematical
Physics 374 (2020), no. 3, 2015–2096.
[21] A. D Ionescu and H. Jia, Nonlinear inviscid damping near monotonic shear flows, To appear in Acta Math.
(2020).
[22] H. Jia and V. Šverák, Local-in-space estimates near initial time for weak solutions of the navier-stokes equations
and forward self-similar solutions, Inventiones mathematicae 196 (2014), no. 1, 233–265.
[23] H. Jia, Linear inviscid damping in gevrey spaces, Archive for Rational Mechanics and Analysis 235 (2020), no. 2,
1327–1355.
[24] H. Jia, Linear inviscid damping near monotone shear flows, SIAM Journal on Mathematical Analysis 52 (2020),
no. 1, 623–652.
[25] H. Jia, Uniform linear inviscid damping and enhanced dissipation near monotonic shear flows in high reynolds
number regime (i): the whole space case, Journal of Mathematical Fluid Mechanics 25 (2023), no. 3, 42.
[26] I. Kukavica, T. T. Nguyen, V. Vicol, and F. Wang, On the Euler+Prandtl expansion for the Navier-Stokes
equations, J. Math. Fluid Mech. 24 (2022), no. 2, Paper No. 47, 46.
[27] I. Kukavica, V. Vicol, and F. Wang, The inviscid limit for the navier–stokes equations with data analytic only
near the boundary, Archive for Rational Mechanics and Analysis 237 (2020), 779–827.
[28] I. Kukavica, V. Vicol, and F. Wang, Remarks on the inviscid limit problem for the Navier-Stokes equations, Pure
Appl. Funct. Anal. 7 (2022), no. 1, 283–306.
[29] Y. Maekawa, On the inviscid limit problem of the vorticity equations for viscous incompressible flows in the
half-plane, Comm. Pure Appl. Math. 67 (2014), no. 7, 1045–1128.
[30] Y. Maekawa, H. Miura, and C. Prange, Local energy weak solutions for the Navier-Stokes equations in the half-
space, Commun. Math. Phys. (2019), 517–580.
[31] N. Masmoudi and W. Zhao, Nonlinear inviscid damping for a class of monotone shear flows in finite channel,
To appear in Ann. Math. (2020).
[32] W. Orr, The stability or instability of steady motions of a perfect liquid and of a viscous liquid, Part I: a perfect
liquid, Proc. Royal Irish Acad. Sec. A: Math. Phys. Sci. 27 (1907), 9–68.
[33] F. Wang, The three-dimensional inviscid limit problem with data analytic near the boundary, SIAM J. Math.
Anal. 52 (2020), no. 4, 3520–3545.
[34] D. Wei and Z. Zhang, Enhanced dissipation for the kolmogorov flow via the hypocoercivity method, Science China
Mathematics 62 (2019), no. 6, 1219–1232.
[35] D. Wei, Z. Zhang, and W. Zhao, Linear inviscid damping for a class of monotone shear flow in sobolev spaces,
Communications on Pure and Applied Mathematics 71 (2018), no. 4, 617–687.
[36] D. Wei, Z. Zhang, and W. Zhao, Linear inviscid damping and vorticity depletion for shear flows, Annals of PDE
5 (2019), no. 1, 1–101.
[37] D. Wei, Z. Zhang, and W. Zhao, Linear inviscid damping and enhanced dissipation for the kolmogorov flow,
Advances in Mathematics 362 (2020), 106963.
[38] D. Wei, Z. Zhang, and H. Zhu, Linear inviscid damping for the β-plane equation, Comm. Math. Phys. 375
(2020), no. 1, 127–174. MR4082187
[39] C. Zillinger, Linear inviscid damping for monotone shear flows, Transactions of the American Mathematical
Society 369 (2017), no. 12, 8799–8855.

130