On the global well-posedness for the Boussinesq system with

horizontal dissipation

Changxing Miao1 and Xiaoxin Zheng2

arXiv:1203.1372v1 [math.AP] 7 Mar 2012

1
Institute of Applied Physics and Computational Mathematics,
P.O. Box 8009, Beijing 100088, P.R. China.
(miao changxing@iapcm.ac.cn)
2
The Graduate School of China Academy of Engineering Physics,
P.O. Box 2101, Beijing 100088, P.R. China.
(xiaoxinyeah@163.com)

Abstract
In this paper, we investigate the Cauchy problem for the tridimensional Boussinesq
equations with horizontal dissipation. Under the assumption that the initial data is an
axisymmetric without swirl, we prove the global well-posedness for this system. In the
absence of vertical dissipation, there is no smoothing effect on the vertical derivatives. To
r
make up this shortcoming, we first establish a magic relationship between ur and ωrθ by
taking full advantage of the structure of the axisymmetric fluid without swirl and some
tricks in harmonic analysis. This together with the structure of the coupling of (1.2) entails
the desired regularity.

Mathematics Subject Classification (2000): 76D03, 76D05, 35B33, 35Q35

Keywords: Boussinesq system, losing estimate, horizontal dissipation, anisotropic inequal-
ity, global well-posedness.

1 Introduction

The Boussinesq system describes the influence of the convection phenomenon in the dynamics
of the ocean or atmosphere. In fact, it is used as a toy model for geophysical flows whenever
rotation and stratification play an important role (see [28]). This system is described by the
following equations:


 (∂t + u · ∇)u − κ∆u + ∇p = ρen , (t, x) ∈ R+ × Rn , n = 2, 3,


(∂ + u · ∇)ρ − ν∆ρ = 0,
t
(1.1)

 divu = 0,


(u, ρ)|
t=0 = (u0 , ρ0 ),

where, the velocity u = (u1 , · · · , un ) is a vector field with zero divergence and ρ is a scalar
quantity such as the concentration of a chemical substance or the temperature variation in a
gravity fields, in which case ρen represents the buoyancy force. The nonnegative parameters κ
and ν denote the viscosity and the molecular diffusion respectively. In addition, the pressure
p is a scalar quantity which can be expressed by the unknowns u and ρ.

1
 In the case where ν and κ are nonnegative constants, the local well-posedness of (1.1) can
be easily established by using the energy method. When variables κ and ν are both positive,
the classical methods allow to establish the global existence of regular solutions in dimension
two and for three dimension with small initial data. Unfortunately, for the inviscid Boussinesq
system (1.1), whether or not smooth solution for some nonconstant ρ0 blows up in finite time is
still an open problem. The intermediate situation has been attracted considerable attentions
in the past years and important progress has been made. When ν is a positive constant
and κ = 0; or ν = 0 and κ is a positive constant, D. Chae [12], and T.Y. Hou and C.
Li [24] proved the global well-posedness independently for the two-dimensional Boussinesq
system. It is also shown the global well-posedness in the critical spaces, see [1]. In addition,
C. Miao and L. Xue [27] proved the global well-posedness of the two-dimensional Boussinesq
equations with fractional viscosity and thermal diffusion when the fractional powers obey mild
condition. Other interesting results on the two-dimensional Boussinesq equations can be found
in [4, 5, 22, 23].
Recently, there are many works devoted to the study of the tridimensional axisymmetric
Boussinesq system without swirl for different viscosities. In [2], a global result was established
but under some restrictive conditions on the initial density, namely it does not intersect the
axis r = 0. Subsequently, T. Hmidi and F. Rousset [21] removed the assumption on the support
of the density and proved the global well-posedness for the Navier-Stokes-Boussinesq system
by virtue of the structure of the coupling between two equations of (1.1) with ν = 0. In [20],
they also proved the global well-posedness for the tridimensional Euler-Boussinesq system with
axisymmetric initial data without swirl.
In the present paper, we consider the case that the diffusion and the viscosity only occur
in the horizontal direction. More precisely,


 (∂t + u · ∇)u − ∆h u + ∇p = ρez , (t, x) ∈ R+ × R3 ,


(∂ + u · ∇)ρ − ∆ ρ = 0,
t h
(1.2)

 divu = 0,


(u, ρ)|
t=0 = (u0 , ρ0 ),

Here ∆h = ∂12 +∂22 . Let us point out that the anisotropic dissipation assumption is natural in the
studying of geophysical fluids. It turns out that, in certain regimes and after suitable rescaling,
the vertical dissipation (or the horizontal dissipation) is negligible as compared to the horizontal
dissipation (or the vertical dissipation)(see [15] for details). In fact, there are several works
devoted to study of the two-dimensional Boussinesq system with anisotropic dissipation. In
[17], R. Danchin and M. Paicu proved the global existence for the two-dimensional Boussinesq
system with horizontal viscosity in only
√ one equation. They mainly exhibited a polynomial
control of k∇uk√L , where the space L stands for the space of functions f in ∩2≤p<∞ Lp such
that
1
kf k√L := sup p− 2 kf kLp ≤ ∞. (1.3)
2≤p<∞
Combining this with the following estimate


k∇uk∞ ≤ C 1 + k∇uk√L log(e + kukH s ) , s>2
yields the global well-posedness of smooth solutions. Next, they observed the fact k∇uk√L
1 1
implies that u ∈ L2loc (R+ , LogLip 2 ), where LogLip 2 stands for the set of bounded functions f
such that
|f (y) − f (x)|
sup 1 ≤ +∞. (1.4)
x6=y;|x−y|≤ 21 |x − y| log 2 (|x − y|)−1

2
And then they established the global existence with uniqueness for rough data with the help
of a losing estimate. Recently, A. Adhikari, C. Cao and J. Wu also established some global
results for different model under various assumption on dissipation in a series of recent papers,
see in particular [4, 5, 10, 11]. In [11], C. Cao and J. Wu proved the global well-posedness for
the two-dimensional Boussinesq system with vertical viscosity and vertical diffusion in terms
of a Log-type inequality. In their proof, they first find that Lp -norm
√ on vertical component of
velocity with 2 ≤ p < ∞ at any time does not grow faster than p logp as p increase by means
of the low-high decomposition techniques.
To better understand the axisymmetric fields, let us recall some algebraic and geometric
properties of the axisymmetric vector fields and discuss the special structure of the vorticity
of system (1.2), see for example [19, 26]. First, we give some general statement in cylindrical
coordinates: we say that a vector field u is axisymmetric if it satisfies

R−α {u(Rα x)} = u(x), ∀α ∈ [0, 2π], ∀x ∈ R3 , (1.5)

where Rα denotes the rotation of axis (Oz) and with angle α. Moreover, an axisymmetric
vector field u is called without swirl if it has the form:
q
u(t, x) = ur (r, z)er + uz (r, z)ez , x = (x1 , x2 , x3 ), r = x21 + x22 and z = x3 ,

where (er , eθ , ez ) is the cylindrical basis of R3 . Similarly, a scalar function f : R3 → R is called

axisymmetric if the vector field x 7→ f (x)ez is axisymmetric, which means that

f (Rα x) = f (x), ∀ x ∈ R3 , ∀ α ∈ [0, 2π]. (1.6)

This is equivalent to say that f depends only on r and z. Direct computations show us that
the vorticity ω := curlu of the vector field u takes the form

ω = (∂z ur − ∂r uz )eθ := ωθ eθ .

On the other hand, we know that

ur ur
u · ∇ = ur ∂r + uz ∂z , divu = ∂r ur + + ∂z uz and ω · ∇u = ω (1.7)
r r
in the cylindrical coordinates. Therefore, the vorticity ω satisfies
ur
∂t ω + u · ∇ω − ∆h ω = −∂r ρeθ + ω. (1.8)
r
Since the horizontal Laplacian operator has the form ∆h = ∂rr + 1r ∂r in the cylindrical coor-
dinates then the ωθ satisfies
ωθ ur
∂t ωθ + u · ∇ωθ − ∆h ωθ + 2
= −∂r ρ + ωθ . (1.9)
r r
In this paper, we are going to establish the global well-posedness for the system (1.2) corre-
sponding to large axisymmetric data without swirl. Since the dissipation only occurs in the
horizontal direction, it seems not obvious to get the global regularity of solutions following
from [2] directly. Indeed, their proof relies on the smoothing effect on vertical direction. Also,
we do not expect to obtain the growth estimate of Lp -norm about vertical component of veloc-
ity as in [11] for the tridimensional
√ axisymmetric Boussinesq equations. Besides, as the space
H 1 (R3 ) fails to be embedded in L(R3 ), it is impossible to obtain the bound of k∇uk√L in

3
terms of kωk√L just as in [17]. This requires us to further study the structure of axisymmetric
flows and establish priori estimate to control the vorticity in L1loc (R+ , L∞ ). Now, let us briefly
to sketch the proof of results. According to (1.9) and the properties of axisymmetric flows, we
find that the quantity ωrθ satisfies

ωθ 2
ωθ ∂r ρ
∂t + u · ∇ − ∆h + ∂r =− . (1.10)
r r r r
We observe that the main difficulty is the lack of information about the influence of the term
in the right side of (1.10) and how to use some priori estimates on ρ to control it. Therefore
we need to study the properties of the operator ∂rr so as to analyze the influence of the forcing
term ∂rrρ on the motion of the fluid. Indeed, the behavior of ∆h + ∂rr is like that of ∆h , which
be derived from the fact that ∂rr is a part of the operator ∆h = ∂r2 + ∂rr . This induces us to
consider the structure of the coupling between two equation of (1.2). From this observation,
we introduce a new quantity Γ := ωrθ − 21 ρ and then Γ solves the following transport equation
2
(∂t + u · ∇)Γ − (∆h + ∂r )Γ = 0. (1.11)
r
It follows that
kΓ(t)kLp ≤ kΓ0 kLp , ∀p ∈ [1, ∞].
This together with the Lp -estimate of ρ gives that
ω ω
θ θ
(t) p ≤ (0) p , ∀p ∈ [1, ∞].
r L r L

This estimate enables us to establish a global H 1 -bound of the velocity. Now, by taking the
L2 -inner product of (1.9) with ωθ and using the anisotropic inequality which will be described
in Appendix A, we obtain
1 d ω
θ 2
kωθ (t)k2L2 + k∇h ωθ (t)k2L2 + (t) 2
2 dt r L
ur 3  ur  1 (1.12)
4 4 1
2
1
2 2 1 2 1
ωθ 2

≤ 6 ∂z 2 kωθ k2 k∇h ωθ k2 kωθ k2 + kρkL2 + k∇h ωθ (t)kL2 + (t) 2 .
r L r L 4 4 r L

As a consequence, it is impossible to use the information of ωrθ to control the quantity

k∂z (ur /r)kL2 via the following pointwise estimate established by T. Shirota and T. Yanag-
isawa (abbr. S-Y)
 ur  1 ω 
   θ
  ≤ C 2 ∗  .
r |x| r
r
This forces us to establish the new relationship between ur and ωrθ instead of the S-Y estimate.
To fulfill the goal, we find the following algebraic identity deduced from the geometric structure
of axisymmetric flows and the Biot-Savart law:
ur ω  ∂r ω 
θ θ
= ∂z ∆−1 − 2 ∆−1 ∂z ∆−1 · (1.13)
r r r r
This identity allows us to conclude the S-Y estimate, one can see Propositions 2.5 and 2.6 for
more details.
Before stating our results, let us introduce the space L of those functions f which belong
to every space Lp with 2 ≤ p < ∞ and satisfy
kf kL := sup p−1 kf kLp ≤ ∞. (1.14)
2≤p<∞

Our results are stated as follows.

4
Theorem 1.1. Let u0 ∈ H 1 be an axisymmetric divergence free vector field without swirl
such that ωr0 ∈ L2 and ∂z ω0 ∈ L2 . Let ρ0 ∈ H 0,1 be an axisymmetric function. Then there is
a unique global solution (u, ρ) of the system (1.2) such that

u ∈ C(R+ ; H 1 ) ∩ L2loc (R+ ; H 1,2 ∩ H 2,1 ), ∂z ω ∈ C(R+ ; L2 ) ∩ L2loc (R+ ; H 1,0 ),

ω
∈ L∞ 2 2
loc (R+ ; L ) ∩ Lloc (R+ ; H
1,0
), ρ ∈ C(R+ ; H 0,1 ) ∩ L2loc (R+ ; H 1,1 ).
r
Here and in what follows, we can refer to Section 2.1 for the definition of spaces such as H 1 ,
H 0,1 , etc.

Remark 1.1. The main difficulty is how to establish the H 1 -estimates of velocity due to the
lack of dissipation in the vertical direction. To overcome this difficulty, we explore an algebraic
r
identity between ur and ωrθ , which strongly rely on the geometric structure of axisymmetric
flows, and control the stretching term in vorticity equation
ur
∂t ω + u · ∇ω − ∆h ω = −∂r ρeθ + ω.
r
We observe the diffusion in a direction perpendicular to the buoyancy force, and this helps us
to control the source term ∂r ρeθ by virtue of the horizontal smoothing effect.

Theorem 1.2. Let u0 ∈ H 1 be an axisymmetric divergence free vector field without swirl
such that ωr0 ∈ L2 and ω0 ∈ L∞ . Let ρ0 ∈ H 0,1 be an axisymmetric function. Then the system
(1.2) admits a unique global solution (ρ, u) such that

u ∈ Cw (R+ ; H 1 ) ∩ L2loc (R+ ; H 1,2 ∩ H 2,1 ), ∇u ∈ L∞

loc (R+ ; L),

ω
∈ L∞ 2 2
loc (R+ ; L ) ∩ Lloc (R+ ; H
1,0
), ρ ∈ Cw (R+ ; H 0,1 ) ∩ L2loc (R+ ; H 1,1 ) ∩ Cb (R+ ; L2 ).
r
Remark 1.2. Compared with Theorem 1.1, the condition ∂z ω0 ∈ L2 has been replaced by
ω0 ∈ L∞ in Theorem 1.2. It enables us to extend the global well-posedness theory to vector-
field lying in space L (which ensures the vector-field belongs to LogLip space) instead of being
Lipschitz. Our choice is motivated by the well-known result that the velocity in the LogLip
space LL (see (6.15) in Appendix A) seems to be the minimal requirement for uniqueness to
the incompressible Euler equations. Indeed, the vorticity equation can provides us the Lp -
norms of vorticity with 2 ≤ p < ∞. This allows us to get that ∇u ∈ L∞ loc (R+ ; L) by means
of the relation k∇ukLp ≤ Cp kωkLp . Furthermore, we can obtain the global well-posedness by
exploring losing estimates.

The paper is organized as follows. In Section 2 we shall give the definitions of the functional
spaces that we shall use and state some useful propositions and algebraic identity. Next, we
shall obtain a priori estimate for sufficiently smooth solutions of (1.2) in Section 3. The last
two sections will be devoted to proving Theorems 1.1 and 1.2. In Appendix, we shall give
a few technical lemmas used throughout the paper. We shall also prove an existence result
and a losing estimate for the anisotropic equations with a convection term, which are the key
ingredients in the proof of the results.
Notations: Throughout the paper, we write R3 = R2h × Rv The tridimensional vector field
u is denoted by (uh , uz ), and we agree that ∇h = (∂1 , ∂2 ). Finally, the Xh (resp.,Xv ) stands
for that Xh is a function space over R2h (resp.,Rv ).

5
2 Preliminaries

2.1 Littlewood-Paley Theory and Besov spaces

In this subsection, we provide the definition of some function spaces based on the so-called
Littlewood-Paley decomposition.
Let (χ,
 ϕ) ben a couple
of smooth functions with values
 in n[0,
 31] such that
χ is supported in
 4  8
the ball ξ ∈ R |ξ| ≤ 3 , ϕ is supported in the shell ξ ∈ R 4 ≤ |ξ| ≤ 3 and
X
χ(ξ) + ϕ(2−j ξ) = 1 for each ξ ∈ Rn . (2.15)
j∈N

For every u ∈ S ′ (Rn ), we define the dyadic blocks as

∆−1 u = χ(D)u and ∆j u := ϕ(2−j D)u for each j ∈ N.

We shall also use the following low-frequency cut-off:

Sj u := χ(2−j D)u.

One can easily show that the formal equality

X
u= ∆j u (2.16)
j≥−1

holds in S ′ (Rn ), and this is called the inhomogeneous Littlewood-Paley decomposition. It has
nice properties of quasi-orthogonality:

∆j ∆j ′ u ≡ 0 if |j − j ′ | ≥ 2. (2.17)

∆j (Sj ′ −1 u∆j ′ v) ≡ 0 if |j − j ′ | ≥ 5. (2.18)

Next, we first introduce the Bernstein lemma which will be useful throughout this paper.
Lemma 2.1. There exists a constant C such that for q, k ∈ N, 1 ≤ a ≤ b and for f ∈ La (Rn ),
1 1
sup k∂ α Sq f kLb ≤ C k 2q(k+n( a − b )) kSq f kLa ,
|α|=k

C −k 2qk k∆q f kLa ≤ sup k∂ α ∆q f kLa ≤ C k 2qk k∆q f kLa .

|α|=k

Definition 2.1. For s ∈ R, (p, q) ∈ [1, +∞]2 and u ∈ S ′ (Rn ), we set

 X 1
2jsq k∆j ukqLp (Rn )
q
kukBp,q
s (Rn ) := if r < +∞
j≥−1

and
js
kukBp,∞
s (Rn ) := sup 2 k∆j ukLp (Rn ) .
j≥−1

Then we define the inhomogeneous Besov spaces as

 
s
Bp,q (Rn ) := u ∈ S ′ (Rn ) kukBp,q
s (Rn ) < +∞ .

s by H s .
We also denote B2,2

6
 Since the dissipation only occurs in the horizontal direction, it is natural to introduce the
following definition.
Definition 2.2. For s, t ∈ R, (p, q) ∈ [1, +∞]2 and u ∈ S ′ (R3 ), we set
 X q 1
jsq ktq h v q
kukBp,q
s,t 3
(R ) := 2 2 ∆ j ∆ k u p 3
if r < +∞
L (R )
j,k≥−1

and
js kt h v
kukBp,∞
s,t
(R3 ) := sup 2 2 ∆ ∆ u
j k .
j,k≥−1 Lp (R3 )

Then we define the anisotropic Besov spaces as

 
s,t
Bp,q (R3 ) := u ∈ S ′ (R3 ) kukBp,q
s,t 3
(R ) < +∞ .
s,t
We also denote B2,2 by H s,t .

Let us now state some basic properties for H s,t spaces which will be useful later.
Lemma 2.2. The following properties of anisotropic Besov spaces hold:

(i) Inclusion realtion: kukH s2 ,t2 (R3 ) ⊆ kukH s1 ,t1 (R3 ) if s2 ≥ s1 and t2 ≥ t1 .
(ii) Interpolation: for s1 , s2 , t1 , t2 ∈ R and θ ∈ [0, 1], we have
kukH θs1 +(1−θ)s2 ,θt1 +(1−θ)t2 (R3 ) ≤ kukθH s2 ,t2 (R3 ) kuk1−θ
H s1 ,t1 (R3 ) .

(iii) For s, t ≥ 0, kukH s,t (R3 ) is equivalent to

kukL2 (R3 ) + kΛsh ukL2 (R3 ) + Λtv u L2 (R3 ) + Λtv Λsh u L2 (R3 ) .

(iv) kukH s,t (R3 ) ≃ kukH s (R2 ) H t (Rv ) ≃ kukH t (Rv ) H s (R2 ) .
h h

1
(v) Algebraic properties: for s > 1 and t > 2, kukH s,t (R3 ) is an algebra.

Proof. We first point out that (i) and (ii) are obviously true. We only need to prove (iii),
(iv) and (v). From the definition of anisotropic Besov Spaces kukH s,t (R3 ) and the Plancherel
theorem, we can conclude that
X 2 X 2

kuk2H s,t (R3 ) = 22js 22kt ∆hj ∆vk u 2 3 = 22js 22kt ϕ2hj ϕ2vk û L2 (R3 )
L (R )
j,k≥−1 j,k≥−1
X X Z
= 22js 22kt ϕ2hj ϕ2vk |û|2 (ξ)dξ
j,k≥−1 j,k≥−1 R3
Z X Z X
= 22js ϕ2hj 22kt ϕ2vk |û|2 (ξh , ξ3 )dξ3 dξh .
R2h j≥−1 Rv k≥−1

This together with Equality (2.15) yields that

Z
2
kukH s,t (R3 ) ≃ (1 + ξh2 )s (1 + ξv2 )t |û|2 (ξ)dξ
R 3
Z Z Z Z
≃ 2
|û| (ξ)dξ + 2s 2
ξh |û| (ξ)dξ + 2t 2
ξv |û| (ξ)dξ + ξh2s ξv2t |û|2 (ξ)dξ (2.19)
R3 R3 R3 R3
2 2
= kuk2 2 3 + kΛsh uk2 2 3 + Λtv u 2 3 + Λsh Λtv u 2 3 .
L (R ) L (R ) L (R ) L (R )

7
This implies the desired result (iii).
From (2.19), it is clear that
Z Z
2 2 s 2 t 2
kukH s,t (R3 ) ≃ (1 + ξh ) (1 + ξv ) |û| (ξ)dξ ≃ |(1 + Λsh )(1 + Λtv )u|2 (x)dx
R 3 R 3
2 2
t s
≃ (1 + Λ2v ) 2 k(1 + Λ2h ) 2 ukL2 2 ≃ kukH s (R2 ) H t (Rv ) .
h Lv h

Similarly, we can show that kukH s,t (R3 ) ≃ kukH t (Rv ) H s (R2 ) .
h
n
Finally, according to the fact that H s (Rn )(s> 2)
is an algebra and (iv). Thus, for any
u, v ∈ H s,t (s > 1, t > 21 ),

kuvkH s,t ≤C kuvkH s (R2 ) H t (Rv ) ≤ C kukH s (R2 ) kvkH s (R2 ) H t (Rv )
h h h

≤C kukH s,t kvkH s,t .

This is exactly the last result.

2.2 Heat kernel and Algebraic Identity

In this subsection, we first review the properties of the heat equation. Next, we give two useful
algebraic identities and its properties.
Proposition 2.3. [13, 25] There exist c and C > 0 such that for every u solution of
(
∂t u − ∆u = 0, x ∈ Rn ,
u|t=0 = u0 ,
the following estimates hold true
n 1 1
(i) ku(t)kLp (Rn ) = ket∆ u0 kLp (Rn ) ≤ Ct− 2 ( q − p ) ku0 kLq (Rn ) , for 1 ≤ q ≤ p ≤ ∞.
2j
(ii) k∆j u(t)kLp (Rn ) = ket∆ ∆j u0 kLp (Rn ) ≤ Ce−ct2 ku0 kLp (Rn ) , for j ≥ 0.
∂r −1
Next, we intend to recall the behavior of the operator r ∆ over axisymmetric functions.
Proposition 2.4. [20] If u is an axisymmetric smooth scalar function, then we have
∂  x2 x2 x1 x2
r
∆−1 u(x) = 22 R11 u(x) + 21 R22 u(x) − 2 2 R12 u(x), (2.20)
r r r r
with Rij = ∂ij ∆−1 . Moreover, for p ∈]1, ∞[ there exists C > 0 such that
k(∂r /r)∆−1 ukLp ≤ CkukLp . (2.21)
Proposition 2.5. Let u be a free divergence axisymmetric vector-field without swirl and ω =
curlu. Then ω  ω 
ur θ ∂r θ
= ∂z ∆−1 − 2 ∆−1 ∂z ∆−1 . (2.22)
r r r r
Besides, there hold that
 ur  ω
θ
∂z p ≤C p, 1<p<∞ (2.23)
r L r L
and ur ω
θ
3−q3q ≤ C q , 1 < q < 3. (2.24)
r L r L

8
Proof. Using Biot-Savart law and the fact ω = ωθ eθ , we have
 ω 
θ
u1 = ∆−1 ((∂z ωθ ) cos θ) = ∂z ∆−1 x1 (2.25)
r
and  ω 
θ
u2 = ∆−1 ((∂z ωθ ) sin θ) = ∂z ∆−1 x2 . (2.26)
r
On the other hand, we observe that
 ω  ω     ωθ 
θ θ
∆−1 xi = xi ∆−1 − xi , ∆−1 , for i = 1, 2. (2.27)
r r r
 

Applying the Laplace operator to the commutator xi , ∆−1 ωrθ , we get
   ωθ  ωθ   ω 
θ
∆ xi , ∆−1 = − xi + ∆ xi ∆−1
r r r ω 
ωθ ωθ θ
= − xi + xi + 2∂i xi ∂i ∆−1
r  r r
ωθ
=2∂i ∆−1 .
r
This means that
   ωθ  ω 
θ ∂r ω 
θ
xi , ∆−1 = 2∂i ∆−2 = 2xi ∆−2 .
r r r r
Inserting this estimate in (2.27) gives
 ω  ω  ∂r ω 
θ θ θ
∆−1 xi = xi ∆−1 − 2xi ∆−2 . (2.28)
r r r r
Plugging (2.28) in (2.25) and (2.26), respectively, we get

ur x1 u1 + x2 u2 x21 + x22   x21 + x22 ∂r −1  

−1 ωθ −1 ωθ
= = ∂z ∆ − 2 ∆ ∂z ∆
r r2 r2 r r2 r r
ω  ∂  ω 
θ r θ
=∂z ∆−1 − 2 ∆−1 ∂z ∆−1 .
r r r
Furthermore,
 ur  ω 
∂r −1 2 −1  ωθ 
θ
∂z = ∂z2 ∆−1 ∆ ∂z ∆ −2 .
r r r r
Combining this with Proposition 2.4 ensures us to get the estimate (2.23).
Finally, by using Proposition 2.4, Lp -boundedness of Riesz operator and the Sobolev em-
bedding theorem,
ur   ω
−1 ωθ θ
3−q 3q ≤ C ∂z ∆ 3q ≤ C
3−q .
r L r L r Lq
This completes the proof.
r ωθ
Next, we will give a precise expression about ur and r by virtue of the algebraic identity
(2.22) and the harmonic analysis tools. More precisely:

9
Proposition 2.6. Let u be a free divergence axisymmetric vector-field without swirl and ω =
curlu. Then
ur x3 ω θ x2 x2 x3 ω θ x2 x2 x3 ω θ x1 x2 x1 x2 x3 ω θ
= (c1 −4γ1 i) 3 ∗ +6γ1 i 22 1 5 ∗ +6γ1 i 21 2 5 ∗ −12γ1 i 2 ∗ , (2.29)
r |x| r r |x| r r |x| r r |x|5 r
3 Γ( 12 ) 3 Γ( 21 )
where c1 = 2π 2 Γ(1) and γ1 = iπ 2 Γ(2) .

Proof. From the algebraic identities (2.22) and (2.20), we obtain

ur ω  x2 ω  x2 ω  x1 x2 ω 
θ θ θ θ
= ∂z ∆−1 − 2 22 R11 ∂z ∆−1 − 2 21 R22 ∂z ∆−1 + 4 2 R12 ∂z ∆−1 .
r r r r r r r r
(2.30)
On the one hand,
ω   1 ω  
θ θb 1 ωθ
∆−1 = −F −1 2
(ξ) = −c1 ∗ .
r |ξ| r |x| r

Thus, ω 
θ x3 ω θ
∂z ∆−1 = c1 ∗ .
r |x|3 r
On the other hand,
ω   ξ ω  
θ k θb
Rkj ∂z ∆−1 = −i∂z ∂j F −1 (ξ) . (2.31)
r |ξ|4 r

Since ξk is a harmonic polynomial of order one, then we can get by using Theorem 5 of Chap-4
in [30] that
ξ  xk
k
F −1 4
= γ1 .
|ξ| |x|
Inserting this equality to (2.31), we have
ω  x ωθ 
θ k
Rkj ∂z ∆−1 = − iγ1 ∂z ∂j ∗ (x)
r |x| r
x3 ω θ xk xj x3 ω θ
=iγ1 δkj 3 ∗ (x) − 3iγ1 ∗ (x).
|x| r |x|5 r

Plugging these equalities in (2.30) gives the desired result.

Remark 2.1. Let us point out that the equality (2.29) implies the S-Y estimate of [29]. More
precisely, by virtue of (2.29) and the triangle inequality, we can conclude that
 ur   x3 ω θ x2 x2 x3 ω θ x2 x2 x3 ω θ x1 x2 x1 x2 x3 ωθ 
  
  =(c1 − 4γ1 i) 3 ∗ + 6γ1 i 22 1 5 ∗ + 6γ1 i 21 2 5 ∗ − 12γ1 i 2 ∗ 
r |x| r r |x| r r |x| r r |x|5 r
1 ω 
 θ
≤C 2 ∗  .
|x| r

Proposition 2.7. Let u be a smooth axisymmetric vector field with zero divergence and we
denote ω = ωθ eθ . Then
ur ω 1  ω  1
θ 2 θ 2
∞ ≤ C 2 ∇h 2.
r L r L r L

10
Proof. From (2.30), it is clear that

ur  ω  2
X  ω 
θ θ
∞ 3 ≤ C ∂z ∆−1 ∞ 3 +C Rkj ∂z ∆−1 ∞ 3. (2.32)
r L (R ) r L (R ) r L (R )
k,j

For the first term in the right side of (2.32), by using Lemma F.1, we obtain
 ω   ω  1  ω  1
θ θ 2 θ 2
∂z ∆−1 ∞ 3 ≤ ∇∂z ∆−1 2 3 ∇h ∇∂z ∆−1 2 3
r L (R ) r L (R ) r L (R )
ω 1  ω  1
θ 2 θ 2
≤C 2 3 ∇h 2 3 .
r L (R ) r L (R )

Using Lemma F.1 again and applying the Lp -boundedness of Riesz operator, the second term
can be bounded by
 ω  1  ω  1 ω 1  ω  1
θ 2 θ 2 θ 2 θ 2
∇Rkj ∂z ∆−1 2 3 ∇h ∇Rkj ∂z ∆−1 2 3 ≤ C 2 3 ∇h 2 3 .
r L (R ) r L (R ) r L (R ) r L (R )

This ends the proof.

3 The priori estimate

In this section, we will give some useful priori estimates.

3.1 Energy estimate and higher-order estimate

We start with L2 energy estimates and the maximum principle.

Proposition 3.1. Let (u, ρ) be a solution of (1.2), then

ku(t)k2L2 + k∇h uk2L2 L2 ≤ (ku0 kL2 + t kρ0 kL2 )2 (3.33)

t

and
kρ(t)k2L2 + k∇h ρk2L2 L2 ≤ kρ0 k2L2 . (3.34)
t

Besides, for 2 ≤ p ≤ ∞,
kρ(t)kLp ≤ kρ0 kLp . (3.35)

Proof. The proof is standard, we also give the proof for reader convenience. We first prove the
estimate (3.35). Multiplying the second equation of (1.2) by |ρ|p−2 ρ and integrating by parts
yields that Z
1 d p
kρ(t)kLp + (p − 1) |ρ|p−2 |∇h ρ|2 dx = 0.
p dt
Thus we obtain
d
kρ(t)kpLp ≤ 0,
dt
which implies immediately
kρ(t)kLp ≤ kρ0 kLp .
For p = ∞, it is just the maximum principle.

11
 For the first one we take the L2 -inner product of the velocity equation with u. From
integration by parts and the fact that u is divergence free, we obtain
1 d
ku(t)k2L2 + k∇h u(t)k2L2 ≤ ku(t)kL2 kρ(t)kL2 . (3.36)
2 dt
Furthermore, we conclude that
d
ku(t)kL2 ≤ kρ(t)kL2 .
dt
By integration in time, we get that
Z t
ku(t)kL2 ≤ ku0 kL2 + kρ(τ )kL2 dτ ≤ ku0 kL2 + tkρ0 kL2 ,
0

where we used the fact kρ(t)kL2 ≤ kρ0 kL2 . Plugging this estimate into (3.36) yields
Z t
1 1

ku(t)k2L2 + k∇h u(τ )k2L2 dτ ≤ ku0 k2L2 + ku0 kL2 + tkρ0 kL2 kρ0 kL2 t.
2 0 2
This implies the first result.
Finally, by the same argument as in proof of (3.33), we obtain the estimate (3.34).

ωθ
Subsequently, we will establish the estimate of the quantities r and ω which enable us to
get the global existence of axisymmetric system (1.2).
Proposition 3.2. Assume that u0 ∈ H 1 , with ωr0 ∈ L2 and ρ0 ∈ L2 . Let (u, ρ) be a smooth
axisymmetric solution (u, ρ) of (1.2) without swirl, then we have

ω 2 Z t   2  ω 2
ω 0
(t) 2 + ∇h (τ ) 2 dτ ≤ 2 + kρ0 k2 ,
r L 0 r L r L2
and Z t
ku(t)k2H 1 + k∇h u(τ )k2H 1 dτ ≤ C0 eC0 t ,
0
where C0 depends only on the norm of the initial data.

ωθ
Proof. According to the equation (1.10), it is clear that r satisfies the following equation

ωθ 2
ωθ ∂r ρ
∂t + u · ∇ − ∆h + ∂r =− · (3.37)
r r r r
Now we recall that (∂t + u · ∇)ρ − ∆h ρ = 0, which can be rewritten as
2 2
(∂t + u · ∇)ρ − (∆h + ∂r )ρ = − ∂r ρ. (3.38)
r r
ωθ
In view of (3.37) and (3.38), we can set Γ := r − 12 ρ and then Γ solves the equation
2
(∂t + u · ∇)Γ − (∆h + ∂r )Γ = 0.
r
Taking the L2 -inner product with Γ and integrating by parts, we have
1 d
kΓ(t)k2L2 + k∇h Γ(t)k2L2 ≤ 0,
2 dt

12
 R
where we used the facts that u is divergence free and − ∂rrΓ Γdx ≥ 0. By integration in time,
we obtain that Z t
kΓ(t)k2L2 + k∇h Γ(τ )k2L2 dτ ≤ kΓ0 k2L2 .
0
This together with the estimate (3.34) yield that
ω Z t  
θ 2 ωθ 2
(t) 2 + ∇h (τ ) 2 dτ
r L 0 r L

2
2
≤ kΓ(t)kL2 + kρ(t)kL2 + k∇h Γ(t)kL2t L2 + k∇h ρ(t)kL2t L2

2
≤2 kΓ0 kL2 + kρ0 kL2 .
This gives the first claimed estimate.
To prove the second estimate. By taking the L2 -inner product of (1.9) with ωθ we get
ω Z Z
1d 2 2 θ 2 ur
kωθ (t)kL2 + k∇h ωθ (t)kL2 + (t) = ωθ ωθ dx − ∂r ρωθ dx.
2 dt r L2 R3 r R3
Integrating by parts,
Z Z Z Z
∂r ρωθ dx = 2π ∂r ρωθ rdrdz = 2π ρ∂r ωθ rdrdz + 2π ρωθ drdz
R3
Z Z
ωθ
= ρ∂r ωθ dx + ρ dx.
R 3 R 3 r
Thus, by the Hölder inequality, we have that
Z 

 
 ∂r ρωθ dx ≤kρkL2 k∇h ωθ kL2 + kωθ /rkL2
R3
1

≤2 kρ0 k2L2 + k∇h ωθ k2L2 + kωθ /rk2L2 .
4
Next, by virtue of the equality (6.6), Proposition 2.5 and the Young inequality, we obtain that
Z  r 3  r  1
 ur  u 4
 ≤ ∂z u
1 1
 ω ω dx
4
kω k 2
k∇ ω k 2
 3 r θ θ  θ 2 h θ 2 kωθ k2
R r L6 r L2
ω 3 1
θ
≤C 2 kωθ k22 k∇h ωθ k22
r L
ω 4 1
θ 3
≤C 2 kωθ k22 + k∇h ωθ k22 .
r L 4
Collecting these estimates with Proposition 3.1 yield
d ω 2 ω 4
θ θ 3
kωθ (t)k2L2 + k∇h ωθ k2L2 + . kρ0 k22 + 2 kωθ k22 .
dt r L2 r L
Therefore we get by the Gronwall inequality that
Z t ω 
2 θ 2
kωθ (t)kL2 + k∇h ωθ (τ )k2L2 + (τ ) dτ
0 r L2
R t ωθ 4  
k r (τ )k 32 dτ ωθ
≤Ce 0 L (0) 2 + kρ0 kL2 t .
r L
2
Since kωkL2 = kωθ kL2 and k∇h ωkL2 = k∇h ωθ kL2 + ωrθ L2 . So, we finally obtain that
2 2

Z t
2
kω(t)kL2 + k∇h ω(τ )k2L2 dτ ≤ C0 eC0 t .
0
This together with the energy estimates yields the second desired estimate. This ends the
proof.

13
3.2 One derivative estimate on vertical variable

In the absence of dissipation on vertical variable, we need to establish the following estimate
on vertical variable in order to compensate this deficiency.
Proposition 3.3. Assume that ∂z ρ0 ∈ L2 and ∂z ω0 ∈ L2 , then we have
Z t
k∂z ρ(t)k2L2 + k∇h ∂z ρ(τ )k2L2 dτ ≤ C1 eexp C1 t , (3.39)
0

and Z t
k∂z ω(t)k2L2 + k∇h ∂z ω(τ )k2L2 dτ ≤ C2 eexp C2 t , (3.40)
0
where C1 and C2 depend only on the norm of the initial data ρ0 and ω0 .

Proof. Applying the operator ∂z to the second equation of (1.2), we obtain

(∂t + u · ∇)∂z ρ − ∆h ∂z ρ = −∂z ur ∂r ρ − ∂z uz ∂z ρ. (3.41)
Taking the L2 -inner product of the equation (3.41) with ∂z ρ and integrating by parts, we get
Z Z
1 d 2 2
k∂z ρ(t)kL2 + k∇h ∂z ρ(t)kL2 = − ∂z u ∂r ρ∂z ρdx − ∂z uz ∂z ρ∂z ρdx
r
2 dt
Z Z r Z
u
= − ∂z ur ∂r ρ∂z ρdx + ∂z ρ∂z ρdx + ∂r ur ∂z ρ∂z ρdx
r
:=I + II + III,
ur
where we used the fact divu = ∂r ur + r + ∂z uz = 0.
For the first term I, by using (6.7), the Hölder and the Young inequalities, we have
1 1 1 1 1 1
I ≤ k∂z ur kL2 2 k∇h ∂z ur kL2 2 k∂r ρkL2 2 k∂z ∂r ρkL2 2 k∂z ρkL2 2 k∇h ∂z ρkL2 2
1
≤2 k∂z ur kL2 k∇h ∂z ur kL2 k∂r ρkL2 k∂z ρkL2 + k∇h ∂z ρk2L2
4
1
2 2 2 2
≤ k∇h ∂z ur kL2 + kωkL2 k∂r ρkL2 k∂z ρkL2 + k∇h ∂z ρk2L2 .
4
We now turn to bound the term II, by using (6.6), Proposition 2.5 and the Young inequality,
we have
3 1 1 1
II ≤ kur /rkL4 6 k∂z (ur /r)kL4 2 k∂z ρkL2 2 k∇h ∂z ρkL2 2 k∂z ρkL2
3 1 4 1
≤C kωθ /rkL2 k∂z ρkL2 2 k∇h ∂z ρkL2 2 ≤ C kωθ /rkL3 2 k∂z ρk2L2 + k∇h ∂z ρk2L2 .
4
Similarly, the term III can be bounded by
3 1 3 1
k∂r ur kL4 6 k∂z ∂r ur kL4 2 k∂z ρkL2 2 k∇h ∂z ρkL2 2
3 1 4 1
≤C k∂r ∇ukL2 k∂z ρkL2 2 k∇h ∂z ρkL2 2 ≤ C k∇h ωkL3 2 k∂z ρk2L2 + k∇h ∂z ρk2L2 .
4
Combining these estimates, we have
d
k∂z ρ(t)k2L2 + k∇h ∂z ρ(t)k2L2
dt
4 ω 4
θ 3
. k∇h ∂z ur k2L2 + kωk2L2 k∂r ρk2L2 k∂z ρk2L2 + k∇h ωkL3 2 k∂z ρk2L2 + 2 k∂z ρk2L2 .
r L

14
Since k∇h ∇ukL2 ≃ k∇h ωkL2 . By the Gronwall inequality and Proposition 3.2, we obtain the
first desired result (3.39).
Applying ∂z to the equation (1.8), we get

2 ∂z ur ur
∂t ∂z ω + u · ∇∂z ω − ∆h ∂z ω = −∂zr ρeθ + ω + ∂z ω − ∂z ur ∂r ω − ∂z uz ∂z ω.
r r
Taking the L2 -inner product to the above equation with ∂z ω and integrating by parts, we
obtain
1 d
k∂z ω(t)k2L2 + k∇h ∂z ω(t)k2L2
2 dt
Z Z Z r
2 ∂z ur u
= − ∂zr ρeθ ∂z ωdx + ω∂z ωdx + ∂z ω∂z ωdx
r r
Z Z
− ∂z ur ∂r ω∂z ωdx − ∂z uz ∂z ω∂z ωdx
Z Z  r Z r
2 u u
= − ∂zr ρeθ ∂z ωdx + ∂z ω∂z ωdx + 2 ∂z ω∂z ωdx
r r
Z Z
− ∂z ur ∂r ω∂z ωdx + ∂r ur ∂z ω∂z ωdx
5
X
:= Ji .
i=1

ur
Here we used the fact divu = ∂r ur + r + ∂z uz = 0.
By the Hölder inequality and the Cauchy-Schwarz inequality, we know
2 2 2
J1 ≤ ∂zr ρ L2 k∂z ωkL2 ≤ k∂z ωk2L2 + ∂zr ρ L2 .

By the inequality (6.6) and Proposition 2.5,

 ur  3 1 1 1

J2 ≤ ∂z 2 kωkL4 6 k∂z ωkL4 2 k∂z ωkL2 2 k∇h ∂z ωkL2 2
r L
 ur  3 3 1

≤C ∂z 2 k∇ωkL4 2 k∂z ωkL4 2 k∇h ∂z ωkL2 2
r L
ω 3 3 1 ω 3 1
θ θ
≤C 2 k∇h ωkL4 2 k∂z ωkL4 2 k∇h ∂z ωkL2 2 + C 2 k∂z ωkL2 2 k∇h ∂z ωkL2 2
r L r L
ω 4 ω 4 1
θ 3 θ 3
≤C 2 k∇h ωkL2 + C 2 k∂z ωkL2 + k∇h ∂z ωk2L2 .
2 2
r L r L 8
For the third term J3 , we get by virtue of the inequality (6.6) and Proposition 2.5 that
ur 3  ur  1 3
4 4
J3 ≤2 6 ∂z 2 k∂z ωkL2 2 k∇h ∂z ωkL2
r L r L
ω 4 1
3
≤C k∂z ωk2L2 + k∇h ∂z ωk2L2 .
r L2 8
Arguing as for proving J2 , the term J4 can be bounded as follows.
1 1 1 1 1 1
J4 ≤ k∂r ωkL2 2 k∂z ∂r ωkL2 2 k∂z ur kL2 2 k∇h ∂z ur kL2 2 k∂z ωkL2 2 k∇h ∂z ωkL2 2
1 1 1 1
≤ kωkL2 2 k∂r ωkL2 2 k∇h ∂z ur kL2 2 k∂z ωkL2 2 k∇h ∂z ωkL2

15
 1
≤C kωkL2 k∂r ωkL2 k∇h ∂z ur kL2 k∂z ωkL2 + k∇h ∂z ωk2L2
8
1
≤C kωk2L2 k∇h ∂z ur k2L2 k∂z ωk2L2 + C k∂r ωk2L2 + k∇h ∂z ωk2L2 .
8
We turn to bound the term J5 , by Lemma F.2 and the Young inequality, we obtain
1 1
J5 ≤ ∂1 u1 L2 2 ∂z ∂1 u1 L2 2 k∂z ωkL2 k∇h ∂z ωkL2
1
≤ kωkL2 ∂z ∂1 u1 L2 k∂z ωk2L2 + k∇h ∂z ωkL2 .
8
Putting this all together and using the fact that k∇h ∇ukL2 ≃ k∇h ωkL2 , we get

d 3
k∂z ω(t)k2L2 + k∇h ∂z ω(t)k2L2
dt 4
ω 4 ω 2
2 2 θ 3 θ
≤ k∂z ωk2L2 + ∂zr ρ L2 + C 2 k∇h ωk2L2 + C 2 k∂z ωk2L2
r L r L
+ C kωk2L2 k∇h ∂z ur k2L2 k∂z ωk2L2 + C k∂r ωk2L2 + kωkL2 k∇h ∇ukL2 k∂z ωk2L2
 ω 4  2 2 ω 4
θ 3 θ 3
. 1 + 2 + kωk2L2 k∇h ωk2L2 k∂z ωk2L2 + ∂zr ρ L2 + 2 k∇h ωk2L2 + k∂r ωk2L2 .
r L r L
This together with Proposition 3.2 and the Gronwall inequality yields the desired the result.

3.3 Strong a priori estimate

In the following, our target is to establish the global estimate about Lipschitz norm of the
velocity which ensures the global existence of solution.
Let us first give a useful lemma which provides the maximal smooth effect of the velocity
in horizontal direction.

Lemma 3.4. Let s1 , s2 ∈ R and p ∈ [2, ∞[. Assume that (u, ρ) be a smooth solution of the
system (1.2), then there holds that

kukL1 B s1 +2,s2 . ku0 kB s1 ,s2 + kukL1 B s1 ,s2 + ku ⊗ ukL1 B s1 +1,s2 ∩L1 B s1 ,s2 +1 + kρkL1 B s1 ,s2 . (3.42)
t p,1 p,1 t p,1 t p,1 t p,1 t p,1

Proof. Applying the operator ∆hq ∆vk to (1.2) and using Duhamel formula we get
Z t Z t
(t−τ )∆h
uq,k (t) =e t∆h
uq,k (0) − e ∆hq ∆vk P(u · ∇u)(τ, x)dτ − e(t−τ )∆h ∆hq ∆vk Pρ(τ, x)ez dτ,
0 0

where uq,k = ∆hq ∆vk u and P is the Leray projection on divergence free vector fields.
According to Proposition 2.3, we have the following estimate for q ≥ 0

2q
ket∆h ∆hq ∆vk f kLp (R3 ) ≤ ket∆h ∆hq ∆vk f kLp (R2 ) p ≤ Ce−ct2 k∆hq ∆vk f kLp (R3 ) .
h L (Rv )

Therefore, for q ≥ 0, we have

Z t
−2q −2q
kurq,k kL1t Lp .2 kuq,k (0)kLp + 2 k∆hq ∆vk (u · ∇u)(τ )kLp dτ + 2−2q kρq,k kL1t Lp
0

16
Multiplying 2q(s1 +2) 2ks2 and summing over q, k, we obtain
+∞
X
2q(s1 +2) 2ks2 kuq,k kL1t Lp
q=0,k=−1
+∞
X Z t +∞
X
. 2qs1 2ks2 kuq,k (0)kLp + 2q(s1 +1) 2ks2 k∆hq ∆vk (u ⊗ u)(τ )kLp dτ
q=0,k=−1 0 q=0,k=−1
Z t +∞
X Z t +∞
X
+ 2qs1 2k(s2 +1) k∆hq ∆vk (u ⊗ u)(τ )kLp dτ + 2qs1 2ks2 kρq,k (τ )kLp dτ
0 q=0,k=−1 0 q=0,k=−1

It follows that,
+∞
X
kukL1 B s1 +2,s2 . kukL1 B s1 ,s2 + 2q(s1 +2) 2ks2 kuq,k kL1t Lp
t p,1 t p,1
q=0,k=−1

. kuk s1 ,s2
L1t Bp,1 + ku0 kB s1 ,s2 + ku ⊗ ukL1 B s1 +1,s2 + ku ⊗ ukL1 B s1 ,s2 +1 + kρkL1 B s1 ,s2 .
p,1 t p,1 t p,1 t p,1

This ends the proof.

Proposition 3.5. Let u0 ∈ H 1 be a divergence free axisymmetric without swirl vector field
such that ωr0 ∈ L2 , ∂z ω0 ∈ L2 and ρ0 ∈ H 0,1 an axisymmetric function. Then any smooth
solution (u, ρ) of the system (1.2) satisfies

k∇ukL1t L∞ ≤ C0 eexp C0 t .

Here, the constant C0 depends on the initial data.

Proof. According to the structure of axisymmetric flows and the incompressible property of
r
velocity, we know that divu = ∂r ur + ur + ∂z uz = 0 and ωθ = ∂z ur − ∂r uz . Therefore

k∇ukL1t L∞ ≤k∂r ur kL1t L∞ + k∂z ur kL1t L∞ + k∂r uz kL1t L∞ + k∂z uz kL1t L∞

ur (3.43)

. 1 ∞ + k∂r ur kL1t L∞ + k∂z ur kL1t L∞ + k∂r uz kL1t L∞ .
r Lt L
r
For the quantity ur L1 L∞ . By virtue of Proposition 2.7 and Proposition 3.2, we get that
t

ur ω 1  ω  1
θ 2 θ 2
1 ∞ ≤ C ∞ 2 ∇h 1 2 ≤ CeCt .
r Lt L r Lt L r Lt L

Next, we turn to bound the quantity k∂z ur kL1t L∞ , by using Lemma F.1 and the Bernstein
inequality, we have
Z t 1 1 1 1
r
k∂z u kL1t L∞ ≤ C k∂z ∇ur kL2 2 k∇h ∂z ∇ur kL2 2 dτ ≤ Ck∂z ωkL2 ∞ L2 k∇h ∂z ωkL2 1 L2 .
t t
0

For the quantity k∂r ur kL1t L∞ and k∂r uz kL1t L∞ , by taking advantage of Lemma 2.1, we know

k∂r ur kL1t L∞ + k∂r uz kL1t L∞ ≤ C kuk 2, 1 .

L1t B2,12

Furthermore, by virtue of Lemma 3.4, we get

kuk 2, 1 . ku0 k 0, 1 + kuk 0, 1 + ku ⊗ uk 1, 1 + ku ⊗ uk 0, 3 + kρk 0, 1

L1t B2,12 B2,12 L1t B2,12 L1t B2,12 L1t B2,12 L1t B2,12

17
 . ku0 kH 1,1 + kukL1 H 1,1 + ku ⊗ ukL1 H 2,1 + ku ⊗ uk 5 7 + kρkL1 H 1,1
t t L1t H 4 , 4 t

. ku0 kH 1,1 + kukL1 H 1,1 + kuk2L2 H 2,1 + kuk2 2 5 , 7 + kρkL1 H 1,1 .

t t Lt H 4 4 t

On the other hand, from the definition of space, we have

kukL2 H 2,1 . kukL2 L2 + k∂z ukL2 L2 + ∇2h u L2 L2 + ∇2h ∂z u L2 L2
t t t t t

and
5 7 5 7
kuk 5 7 . kukL2 L2 + Λh4 u L2 L2 + Λv4 u L2 L2 + Λh4 Λv4 u L2 L2
L2t H 4 , 4 t t t t
2

. kukL2 L2 + ∇h u L2 L2 + ∇h u L2 L2 + ∂z u L2 L2 + ∂z2 u L2 L2 + ∇h ∂z ω L2 L2 .
t t t t t t

It remains to bound the norm of ρ. By the first estimate of Proposition 3.2 and Proposition 3.3,
we have

kρkL1 H 1,1 . ρ L1 L2 + ∂z ρ L1 L2 + ∇h ρ L1 L2 + ∇h ∂z ρ L1 L2 ≤ Ceexp Ct .
t t t t t

Collecting these estimates with Proposition 3.33, Proposition 3.2 and Proposition 3.3 yields
that
k∇ukL1t L∞ ≤ C0 eexp C0 t .
This ends the proof.

4 Proof of Theorem 1.1

Here we use the Friedrichs method (see [17] for more details): For n ≥ 1, let Jn be the spectral
cut-off defined by
Jd b
n f (ξ) = 1[0,n] (|ξ|)f (ξ), ξ ∈ R3 .
We consider the following system in the spaces L2n := {f ∈ L2 (R3 )| suppf ⊂ B(0, n)}:


∂t u + PJn div(PJn u ⊗ PJn u) − ∆h PJn u = PJn (ρe3 ),
∂t ρ + Jn div(Jn uJn ρ) − ∆h Jn ρ = 0, (4.44)


(ρ, u)|t=0 = Jn (ρ0 , u0 ).

The Cauchy-Lipschitz theorem entails that this system exists a unique maximal solution
(ρn , un ) in C 1 ([0, Tn∗ [; L2n ). On the other hand, we observe that Jn2 = Jn , P 2 = P and
Jn P = PJn . It follows that (ρn , Pun ) and (Jn ρn , Jn Pun ) are also solutions. The unique-
ness gives that Pun = un , Jn un = un and Jn ρn = ρn . Therefore


 ∂t un + PJn div(un ⊗ un ) − ∆h un = PJn (ρn e3 ),


∂ ρ + J div(u ρ ) − ∆ ρ = 0,
t n n n n h n
(4.45)

 divun = 0,


(ρ , u )|
n n t=0 = Jn (ρ0 , u0 ).

As the operators Jn and PJn are the orthogonal projectors for the L2 -inner product, the above
formal calculations remain unchanged. We will start with the following stability results.

18
Lemma 4.1. Let u0 be a free divergence axisymmetric vector-field without swirl and ρ0 an
axisymmetric scalar function. Then

(i) for every n ∈ N, u0,n and ρ0,n are axisymmetric and divu0,n = 0.
(ii) If u0 ∈ H 1 is such that (curl u0 )/r ∈ L2 and ρ0 ∈ H 0,1 . Then there exists a constant C
independent of n such that

ku0,n kH 1 ≤ ku0 kH 1 , (curl u0,n )/r L2 ≤ C (curl u0 )/r L2 ,

kρ0,n kL2 ≤ kρ0 kL2 , kρ0,n kH 0,1 ≤ Ckρ0 kH 0,1 .

Proof. The proof of (curl u0,n )/r L2 ≤ C (curl u0 )/r L2 is subtle, one can see [16] for more
details. Other estimates can be proved by the standard methods.

Now, we come back to the proof of the existence parts of Theorem 1.1. From Lemma 4.1,
we observe that the initial structure of axisymmetry is preserved for every n and the involved
norms are uniformly controlled with respect to this parameter n. This ensures us to construct
locally in time a unique solution (un , ρn ) to the approximate system (4.45). On the other
hand, we have seen in Proposition 3.5 that the Lipschitz norm of the velocity keeps bounded
in finite time. Therefore, this solution is globally defined. By standard compactness arguments
and Lions-Aubin Lemma we can show that this family (un , ρn )n∈N converges to (u, ρ) which
satisfies in turn our initial problem. And the Fatou Lemma ensures (u, ρ) ∈ X , where

X := L∞ 1 2
loc (R+ ; H ) ∩ Lloc (R+ ; H
2,1
) ∩ L∞ loc (R+ ; H
1,1
∩ H 0,2 ) ∩ L2loc (R+ ; H 2,1 ∩ H 1,2 )



∩ L1loc (R+ ; Lip) × L∞ loc (R+ ; H
0,1
) ∩ L2loc (R+ ; H 1,1 ) .

It remains to prove the time continuity of the solution (u, ρ). We only show that u belongs
to C(R+ ; H 1 ), the other terms can be treated the same way. First we show the continuity of u
in H 1 . Indeed, we just need to show that ω ∈ C(R+ ; L2 ). Let us recall the vorticity equation
ur
∂t ω + u · ∇ω − ∆h ω = −∂r ρeθ + ω.
r
It is easy to check that the source terms belong to L2loc (R+ ; L2 ). Using the fact ∇u ∈
L1loc (R+ ; Lip) and applying Proposition F.4, we get the desired result ω ∈ C(R+ ; L2 ).
Next, let us turn to prove the uniqueness. We assume that (ui , ρi ) ∈ X , 1 ≤ i ≤ 2 be two
solutions of the system (1.2) with the same initial data (u0 , ρ0 ). Then the difference (δρ, δu, δp)
between two solutions (ρ1 , u1 , p1 ) and (ρ2 , u2 , p2 ) satisfies
(
∂t δu + div(u2 ⊗ δu) − ∆h δu + ∇δp = −δu · ∇u1 + δρez ,
(4.46)
∂t δρ + div(u2 δρ) − ∆h δρ = −δu · ∇ρ1 .

Taking the L2 -inner product to the first equation of (4.46) with u, we obtain
Z Z
1 d 2 2
kδu(t)kL2 + k∇h δukL2 = − δu∇u1 δudx + δρez δudx
2 dt (4.47)
≤ k∇u1 kL∞ kδuk2L2 + kδρkL2 kδukL2 .
On the other hand, by the same computation, we get
Z Z Z
1 d 2 2
kδρ(t)kL2 + k∇h δρkL2 = − δu∇ρ1 δρdx = − (δu) ∂r ρ1 δρdx − (δu)z ∂z ρ1 δρdx.
r
2 dt

19
By Lemma F.3 and the Young inequality,
Z 1 1 1 1 1 1
(δu)r ∂r ρ1 δρdx ≤ k(δu)r kL2 2 k∇h (δu)r kL2 2 k∂r ρ1 kL2 2 k∂z ∂r ρ1 kL2 2 kδρkL2 2 k∇h δρkL2 2
1
≤C k∂r ρ1 kL2 k∂z ∂r ρ1 kL2 kδukL2 kδρkL2 + k∇h δukL2 k∇h δρkL2 .
2
Using Lemma F.3 and divδu = 0, we have
Z 1 1 1 1 1 1
(δu)z ∂z ρ1 δρdx ≤ k(δu)z kL2 2 k∂z (δu)z kL2 2 k∂z ρ1 kL2 2 k∇h ∂z ρ1 kL2 2 kδρkL2 2 k∇h δρkL2 2
1 1 1 1 1 1
≤ k(δu)z kL2 2 k∇h (δu)kL2 2 k∂z ρ1 kL2 2 k∇h ∂z ρ1 kL2 2 kδρkL2 2 k∇h δρkL2 2
1
≤C k∂z ρ1 kL2 k∇h ∂z ρ1 kL2 kδukL2 kδρkL2 + k∇h δukL2 k∇h δρkL2 .
2
The combination of these estimates yield
1 d
kδρ(t)k2L2 + k∇h δρk2L2
2 dt
≤C(k∇h ρ1 kL2 + k∂z ρ1 kL2 ) k∇h ∂z ρ1 kL2 kδukL2 kδρkL2 + k∇h δukL2 k∇h δρkL2 .

This together with (4.47) yields that

1 d  
kδρ(t)k2L2 + kδu(t)k2L2 + k∇h δρk2L2 + k∇h δuk2L2
2 dt

≤C k∇h ρ1 kL2 + k∂z ρ1 kL2 k∇h ∂z ρ1 kL2 kδukL2 kδρkL2 + k∇h δukL2 k∇h δρkL2
+ k∇u1 kL∞ kδuk2L2 + kδρkL2 kδukL2 .

Consequently,
d    
kδρ(t)k2L2 + kδu(t)k2L2 ≤ CF (t) kδρ(t)k2L2 + kδu(t)k2L2 ,
dt
where
F (t) = (k∇h ρ1 kL2 + k∂z ρ1 kL2 ) k∇h ∂z ρ1 kL2 + k∇u1 kL∞ + 1.
By Proposition 3.1 and Proposition 3.3, we know that F (t) is integrable. Therefore, we obtain
the uniqueness by using the Gronwall inequality.

5 Proof of Theorem 1.2

In this section, we intend to prove the global existence and the uniqueness of Theorem 1.2 for
another class of initial data.

Proposition 5.1. Assume that u0 ∈ H 1 , with ωr0 ∈ L2 and ω0 ∈ L∞ . Let ρ0 ∈ H 0,1 . Then
any smooth axisymmetric solution (u, ρ) of (1.2) without swirl satisfies


k∇u(t)kL ≤ Ceexp Ct kω0 kL2 ∩L∞ + 1 .

Here constant C depends on the initial data.

20
Proof. Multiplying the vorticity equation (1.9) with |ωθ |p−2 ωθ and performing integration in
space, we get
Z Z Z
1 d 2 ω2
|ωθ | dx + (p − 1) |∇h ωθ | |ωθ | dx + |ωθ |p−2 2θ dx
p p−2
p dt r
Z r Z (5.48)
u p p−2
= |ωθ | dx − ∂r ρ|ωθ | ωθ dx.
r
We consider the case p ≥ 4. For the first term in the last line, we deuce by the Hölder inequality
that Z r ur
u
|ωθ |p dx ≤ ∞ kωθ kpLp .
r r L
Since kukLp−2 ≤ kukαL2 kuk1−α
Lp with α =
4
(p−2)2
and
Z Z 4
2(p−4) 4 p−2 2(p−4)
p−4
|ωθ | 2
|∇h ωθ | dx = |ωθ |p−4 |∇h ωθ | p−2 |∇h ωθ | p−2 dx ≤ k∇h ωθ kLp−2
2
|ωθ | 2 ∇h ωθ 2p−2 .
L

By using Lemma F.3 and some based inequalities, the second term can be bounded as follows
Z
∂r ρ|ωθ |p−2 ωθ dx
1 1
p−2
2 p 1 p
1
2 1 p−2 2
1
2 2 2
≤ k∂r ρkL2 ∂rz ρ L2 ωθ ∇h ω 2 2
2
ωθ 2 ∇h ωθ2 2
2
θ L L
L2 L2
√ 1 1 p−2 1 p
1
2 2 2 4 p2 −2 2 p
4 2
2
∂ ρ kω k
≤C p k∂r ρkL2 rz L2 θ Lp−2 θ ω ∇ ω
h θ 2 kω k ∇
θ Lp h ω θ 2
L L
1 1 1 p(p−4) p p
1 +
(p−4) 1
1 2 2(p−2)
≤Cp p−2 k∂r ρkL2 2 ∂rz
2 2
ρ L2 kωθ k p−2
kωθ k
L 2
4(p−2)
kωθ kLp k∇h ωθ k
Lp
4
∇h ωθ2 2
p−2
L 2
L
p−2 p−2 2 2 2
1 p
2
2
∂rz
2 p−1 p−2−
≤Cp p−1 k∂r ρkLp−1
2 ρ L2 k∇h ωθ kLp−1 p−1
2 kωθ kL2 kωθ kLp
p−1
+ ∇h ωθ2 2
4 L

Without loss of generality, we assume that kωθ kLp ≥ 1, thus

d ur ur
2
kωθ k2Lp ≤ C ∞ kωθ k2Lp + Cp p−1 F (t) ≤ C ∞ kωθ k2Lp + 4CF (t),
dt r L r L
p−2 p−2 2 2
where F (t) := k∂r ρkLp−1
2
∂ 2 ρ p−1 k∇ ω k p−1 kω k p−1 . According to Proposition 3.2 and
rz h θ L2 θ L2
L2
Proposition 3.3, we know that F (t) is integrable. Therefore, by the Gronwall inequality and
the relation kωkLp = kωθ kLp , we obtain that
 Z t 
kωk2Lp ≤ Ce exp Ct
kω0 k2Lp + F (τ )dτ .
0

This together with the second estimate of Proposition 3.2 yields

kωk2Lp ≤ Ceexp Ct for 2 ≤ p < ∞.

p 2
Since k∇ukLp ≤ C p−1 kωkLp (see [13]. Chap-3). So, we finally obtain that k∇ukL ≤ Ceexp Ct .
This ends the proof.

21
 Let us first focus on the existence part of Theorem 1.2. Let

Y := L∞ 1 2
loc (R+ ; H ) ∩ Lloc (R+ ; H
1,1
) ∩ L∞loc (R+ ; H
1,1
∩ H 0,2 ) ∩ L2loc (R+ ; H 2,1 ∩ H 1,2 )



∩ L1loc (R+ ; L) × L∞ loc (R+ ; H
0,1
) ∩ L2loc (R+ ; H 1,1 ) .

To prove existence, we smooth out the initial data (u0 , ρ0 ) so as to obtain a sequence
(u0,n , ρ0,n )n∈N of smooth functions which converges to (u0 , ρ0 ). From Lemma 4.1, it is clear
that the initial structure of axisymmetry is preserved for every n . By preceding argument,
it is easy to check that (un , ρn ) ∈ Y. Combining this with the equations (1.2), one may con-
clude that ∂t ρn ∈ L2loc (R+ ; H −1 ) and ∂t un ∈ L2loc (R+ ; L2 ). On the other hand, we know that
L2 ֒→ H −1 and H 1 ֒→ L2 are locally compact. Therefore, by the classical Aubin-Lions argu-
ment and Cantor’s diagonal process, we can deduce that, up to extraction, family (un , ρn )n∈N
has a limit (u, ρ) satisfying the equations (1.2) and that (u, ρ) ∈ Y. The same arguments as
used in Proposition F.4 allows us to show that the time continuity of (u, ρ) in low norms and
the weak time continuity. In addition, in a similar way as used in Theorem 5 of [18], we can
conclude that ρ ∈ Cb (R+ ; L2 ).
Now let us turn to the prove the uniqueness. We assume that (ui , ρi ) ∈ Y, 1 ≤ i ≤ 2 be
two solutions of the system (1.2) with the same initial data (u0 , ρ0 ). One can write (4.46)
(
∂t δu + div(u2 ⊗ δu) + div(δu ⊗ u1 ) − ∆h δu + ∇δp = δρez ,
(5.49)
∂t δρ + div(u2 δρ) − ∆h δρ = −div(δuρ1 ).
1
For the sake of convenience, let (α, β, γ) such that 2 < α < β < γ ≤ 1. Note that

kSq ukL∞ 3 kSq uk q

L 3
≤ 2q ≤ 2 2 kukL .
q q
By the same argument as in Proposition F.5 with the vector-field u2 , then there exists T1 such
that Z t
kδρkH β−1 ≤ C kdiv(δuρ1 )kH γ−1 dτ for all t ∈ [0, T1 ].
0
The term on the right side can be bounded as follows. By virtue of the Bony decomposition:
2

X
div(δuρ1 ) = div Tδu ρ1 + R(δu, ρ1 ) + T∂i ρ1 δui , (5.50)
i=1

where we have used the condition divδu = 0.

From standard continuity results for operators T and R (see for example [7]), we have

kTδu ρ1 + R(δu, ρ1 )kH γ ≤ C kδukL∞ kρ1 kH γ .

As for the last term, since γ − 1 < 0, we infer that

T∂ ρ δui γ−1 ≤ C k∇ρ1 k γ−1 kδuk ∞ .
i 1 H H L

We eventually get
kδρkL∞ (H β−1 ) ≤ C kρ1 kL2 (H γ ) kδukL2 L∞ . (5.51)
t t t

Now we turn to bound the term δu. By using Proposition F.5, there exists T2 such that for all
t ∈ [0, T2 ],


kδukL∞ (H α ) + k∇h δukL2 (H α ) ≤ C kδρkL2 (H β ) + kδu · ∇u1 kL2 (H β )
t t t t

22
for some constant C depending only on α, β and u2 . Using again the Bony decomposition and
arguing exactly as for proving (5.51), we get

kδu · ∇u1 kH β−1 ≤ C kδukL∞ ku1 kH β .

Therefore, given that u1 ∈ L∞ β

loc (R; H ),

kδukL∞ (H α ) + k∇h δukL2 (H α ) ≤ C(kδρkL2 (H β−1 ) + kδukL2 (L∞ ) ). (5.52)

t t t t

Next, our task is to show that kδukL2 L∞ may be bounded in terms of kδukL∞ H α and of
t t
k∇h δukL2 H α .
t

According to the assumption α ∈] 21 , 1], we have (see the proof in Appendix A)

2
α− 1 2
3
−α
kδukL∞ (R3 ) ≤ C kδukH α (R 3 ) k∇h δuk H α (R3 ) . (5.53)

Combining these estimates, we can deduce that for some constant C depending only on T =
min{T1 , T2 } and on the norms of (ρ1 , u1 ) and (ρ2 , u2 ), we have
α 1 1 α 1

kδρkL∞ H β−1 ≤ Ct 2 − 4 δU (t), δU (t) ≤ C t 2 kδρkL∞ H β−1 + t 2 − 4 δU (t)
t t

with
δU (t) := kδukL∞ H α + k∇h δukL2 H α .
t t

It follows that δu ≡ 0 (and thus δρ ≡ 0) on a suitably small time interval. Finally, let us notice
that our assumptions on the solutions ensure that δρ ∈ C([0, T ]; H β−1 ) and δu ∈ C([0, T ]; H α ).
Using a classical connectivity argument, it is now easy to get the uniqueness on the whole
interval [0, ∞[.

A Appendix

In this section, we first give some useful inequalities which have been used throughout the
paper.

Lemma F.1. There exists a constants C such that

1 1
kukL∞ (R3 ) ≤ C k∇ukL2 2 (R3 ) k∇h ∇ukL2 2 (R3 ) . (6.1)

Proof. By using the interpolation theorem, we get

1 2 1
ku(xh , ·)kL∞ (Rv ) ≤ Cku(xh , ·)kL2 6 (Rv ) kΛv3 u(xh , ·)kL2 2 (Rv ) .

This together with the Minkowski inequality and the embedding theorem gives

kukL∞ (R3 ) ≤ kukL∞ (Rv ) L∞ (R2 )
h
1 2 1
≤C kukL6 (Rv ) kΛv ukL2 2 (Rv ) L∞ (R2 )
2 3
(6.2)
h
1 1 2 1
≤ kΛv ukL∞ (R2 ) L2 (Rv ) kΛv ukL∞ (R2 ) L2 2 (Rv ) .
3 2 3
h h

23
On the other hand, using again the interpolation theorem and the embedding theorem, we
have
1 1 2 5 1 1
kΛv3 u(·, z)kL∞ (R2 ) ≤CkΛv3 u(·, z)kL3 6 (R2 ) kΛh3 Λv3 u(·, z)kL3 2 (R2 )
h h h
2 1 2 5 1 1
(6.3)
3 3 3 3 3 3
≤CkΛh Λv u(·, z)k L2 (R2h )
kΛh Λv u(·, z)k L2 (R2h )

and
2 2 1 4 2 2
kΛv3 u(·, z)kL∞ (R2 ) ≤CkΛv3 u(·, z)kL3 3 (R2 ) kΛh3 Λv3 u(·, z)kL3 2 (R2 )
h h h
1 2 1 4 2 2
(6.4)
3 3 3 3 3 3
≤CkΛh Λv u(·, z)k L2 (R2h )
kΛh Λv u(·, z)k L2 (R2h )
.

Inserting (6.3) and (6.4) into (6.2), and using the Hölder inequality, we get
2 1 2 5 1 1 1 1 2 1 4 2 2 1
kukL∞ (R3 ) ≤ kΛh3 Λv3 ukL3 2 (R2 ) kΛh3 Λv3 ukL3 2 (R2 ) L2 2 (Rv ) kΛh3 Λv3 ukL3 2 (R2 ) kΛh3 Λv3 ukL3 2 (R2 ) L2 2 (Rv )
h h h h
2 1 1 5 1 1 1 2 1 4 2 1
3 3 3 3 3 6 3 3 6 3 3 3
≤kΛh Λv ukL2 (R3 ) kΛh Λv ukL2 (R3 ) kΛh Λv ukL2 (R3 ) kΛh Λv ukL2 (R3 )
1 1
≤C k∇ukL2 2 (R3 ) k∇h ∇ukL2 2 (R3 ) .

This completes the proof.

Lemma F.2. Let q ∈]2, ∞[, there holds that

Z q−1 1 q−2 1 1
q
f ghdx1 dx2 dx3 ≤ C kf kL2(q−1) k∂x1 f kLq 2 kgkL2q k∂x2 gkLq 2 k∂x3 gkLq 2 khkL2 . (6.5)
R3

In particular, if we take q = 4 in (6.5), we have

Z 3 1 1 1
f ghdx1 dx2 dx3 ≤ C kf kL4 6 k∂x3 f kL4 2 kgkL2 2 k∇h gkL2 2 khkL2 . (6.6)
R3

Proof. We only just to show the inequality for functions f, g, h ∈ C0∞ (R3 ) and then pass to
the limit by virtue of the density argument.
Using some basic inequalities, we have
Z
f ghdx1 dx2 dx3
R3
Z " Z  1 Z
2
1 #
2
2 2
≤C max |f | g dx1 h dx1 dx2 dx3
R2 x1 R R
Z  1 "Z Z  q # q−2 Z
2q 1
q q−2 2
q 2 2
≤C max |f | dx2 dx3 g dx1 dx2 dx3 h dx1 dx2 dx3
R2 x1 R2 R R3
Z Z  1 "Z Z  q # q−2
2q
q q−2
≤C |f |q−1 |∂x1 f |dx1 dx2 dx3 g 2 dx1 dx2 dx3 khkL2
R2 R R2 R
q−1 1 q−2 1 1
q
≤C kf kL2(q−1) k∂x1 f kLq 2 kgkL2q k∂x2 gkLq 2 k∂x3 gkLq 2 khkL2 .

24
Indeed, by imbedding theorem, Hölder’s inequality and Plancherel theorem, we obtain that
1 1
kgk 2q ≤C kΛ2q gkL22 (R1 ) 2q ≤ C kΛ2q gk 2q
q−2 L21 (R) L21 (R) L22 (R1 ) L21 (R)
L2,3 (R2 ) L3q−2 (R1 ) L3q−2 (R1 )
1 1 1 1
≤C kΛ3q Λ2q gkL23 (R1 ) L2 (R1 ) L2 (R) = C Λ3q Λ2q g L2
2 1
Z 2 2
1 q−2 1 1
2
=C |ξ2 | q |ξ3 | q ĝ 2 (ξ)dξ1 dξ2 dξ3 ≤ kĝkL2q kξ2 ĝkLq 2 kξ3 ĝkLq 2
R3
q−2 1 1
≤C kgkL2q k∂x2 gkLq 2 k∂x3 gkLq 2 .

This completes the proof.

Lemma F.3. [14] A constant C exists such that

Z 1 1 1 1 1 1
f ghdx1 dx2 dx3 ≤ C kf kL2 2 k∂x3 f kL2 2 kgkL2 2 k∇h gkL2 2 khkL2 2 k∇h hkL2 2 . (6.7)
R3

Proof of Inequality (5.53). As for α ∈] 12 , 32 [, by the interpolation theorem and the embedding
theorem, we get that for any z ∈ R
α+ 1
α− 12 α+ 1 2
ku(·, z)kL∞ (R2 ) ≤C ku(·, z)k Λ 2 u(·, z)
h
4 h
Lh3−2α (R2 ) L2h (R2 )
α− 1 3 −α
α− 1 2 α+ 1 2
≤C 2
Λh u(·, z) 2 2
Λ 2
h u(·, z) 2 2 (6.8)
Lh (R ) Lh (R )
α− 21 3
−α
≤C ku(·, z)k α− 1
k∇h u(·, z)k 2 α− 1 .
H 2 (R2 ) H 2 (R2 )

1
On the other hand, by the trace theorem that H α (R3 ) ֒→ L∞ (Rv ; H α− 2 (R2h )),

kuk α− 2 1 ≤ C kuk(H α ) and k∇h uk α− 21 ≤ C k∇h uk(H α ) .

v (H
L∞ ) v (H
L∞ )

Inserting these inequalities in (6.8), we get the desired result (5.53).

For the sake of completeness, we give an existence result for the anisotropic equations with
a convection term, which is similar to the case of the transport equation in [7, 26].

Proposition F.4. Let s ≥ −1, 1 ≤ p < ∞ and 1 ≤ r ≤ ∞. Assume that f0 ∈ Bp,r s ,g ∈

1 s σ −m
L ([0, T ]; Bp,r ), and that u be a divergence free vector-field satisfying u ∈ L ([0, T ]; B∞,∞ ) for
some σ > 1 and m > 0, and ∇u ∈ L1 ([0, T ]; L∞ ). Then the following equations
(
∂t f + u · ∇f − ∆h f = g,
(6.9)
f |t=0 = f0

admits a unique solution f in

s ), if r < ∞,
• the space C([0, T ]; Bp,r


s′ ) ∩ C ([0, T ]; B s ), if r = ∞.
• the space ∩s′ <s C([0, T ]; Bp,∞ w p,∞

25
Moreover, we have
Rt  Z t Rτ 
V (τ )dτ
kf (t)kBp,r
s ≤ eC 0 kf0 kBp,r
s + e−C 0 V (s)ds
kg(τ )kBp,r
s dτ . (6.10)
0

Here V (t) := k∇u(t)kL∞ .

Proof. Without loss of generality, we assume that u and g are defined on R × Rn . We first
construct the approximate solutions fn of (6.9) as follows.


∂t fn + un · ∇fn − ∆h fn = gn ,
un := ϕn ∗t Sn u, gn := ϕn ∗t Sn g, (6.11)


fn |t=0 = f0,n := Sn f0 ,

where, ϕn denotes a family of mollifiers with respect to t. Thanks to the properties of mollifier
and the operator Sj , it is clear that f0,n ∈ Bp,r∞ and u , g ∈ C([0, T ]; B ∞ ) with B ∞ :=
n n p,r p,r
s . Moreover, (f
∩s∈R Bp,r s 1 s
0,n )n∈N is bounded in Bp,r , (gn )n∈N is bounded in L ([0, T ]; Bp,r ),
σ −m 1 ∞
(un )n∈N is bounded in L ([0, T ]; B∞,∞ ), and (∇un )n∈N is bounded in L ([0, T ]; L ).
Applying ∆k to (6.11), we have
(
∂t ∆k fn + un · ∇∆k fn − ∆h ∆k fn = ∆k gn + Rnk ,
∆k fn |t=0 = ∆k f0,n ,

where Rnk := un · ∇∆k fn − ∆k (u · ∇fn ). Note that

Z
− ∆h (∆k fn )|∆k fn |p−2 ∆k fn dx ≥ 0,

it is easy to conclude that

Z t Z t
k
k∆k fn (t)kLp ≤ k∆k f0 kLp + k∆k gn (τ )kLp dτ + Rn (τ ) dτ. (6.12)
0 0 Lp

This together with the commutator estimate (see e.g. [26], Chap-2 )

k
Rn p ≤ Cck (t)2−ks Vn (t) kfn (t)kBp,r
(t) s with kck (t)klr = 1
L

leads to Z t 
kfn (t)kBp,r
s ≤ kf0 kBp,r
s + kgn (τ )kBp,r
s + CVn (τ ) kf (τ )kBp,r
s dτ,
0
where Vn (t) := k∇un (t)kL∞ . Applying the Gronwall inequality, we obtain
Rt  Z t Rτ 
Vn (τ )dτ
kfn (t)kBp,r
s ≤e C 0 kf0 kBp,r
s + e−C 0
Vn (s)ds
s dτ .
kgn (τ )kBp,r
0

In the following, we shall show that, up to an subsequence, the sequence (fn )n∈N converges in
D ′ (R+ × Rn ) to a solution f of (6.9) which has the desired regularity properties. First, one
may write
∂t fn − gn = −un · ∇fn + ∆h fn .

26
Hence un ∈ Lσ ([0, T ]; B∞,∞ −m ) enables us to conclude that ∂ f − g
t n n is bounded in
σ −M
L ([0, T ]; Bp,∞ ) for some sufficiently large M > 0. For the sake of convenience, let
Z t
f¯n (t) := fn (t) − gn (τ )dτ.
0

Thanks to the imbedding theorem, one may deduce that (f¯n )n belongs to C β ([0, T ]; Bp,∞ −M )

with β > 0 and hence uniformly equicontinuous with value in Bp,∞ −M . Now, let (χ )
l l∈N be a
∞
sequence of C0 (R ) cut-off functions supported in the ball B(0, l + 1) of Rn and equal to 1 in
n

a neighborhood of B(0, l). On the other hand, by Theorem 2.94 of [7], we know that the map
s to B −M . By using Ascoli’s theorem and the Cantor diagonal
u 7→ χl u is compact from Bp,r p,∞
process, there exists a subsequence which we still denote by (f¯n )n∈N such that, for all l ∈ N,

χl f¯n →n→∞ χl f¯ in −M
C([0, T ]; Bp,∞ ).

It follows that the sequence (f¯n )n∈N converges to some distribution f¯ in D ′ (R+ × Rn ).
The only problem is to pass to the limit in D ′ (R+ × Rn ) for the convection term. Let
ψ ∈ C0∞ (R+ × Rn ) and l ∈ N be such that suppψ ⊂ [0, T ] × B(0, l), we have the decomposition


ψun · ∇fn − ψu · ∇f = ψun · ∇(χl fn − χl f ) − ψχl (un − u) · ∇f (6.13)

Coming back to the uniform estimates of fn ∈ L∞ ([0, T ]; Bp,r s ), the Fatou properties of Besov

space ensures f¯ belong to L∞ ([0, T ]; Bp,rs ). By preceding argument, we find that χ f¯ tends to
l n
¯ s−ε
χl f in C([0, T ]; Bp,∞ ) for all ε > 0 and l ∈ N. Therefore, both two terms in the right of (6.13)
tend to zero in L∞ ([0, T ]; Bp,∞ s−1−ε ). On the other hand, the sequences (f
0,n )n∈N , (gn )n∈N and
Rt
(un )n∈N converges to f0 , g and u, respectively. So, we finally conclude that f := f¯ + 0 g(τ )dτ
is a solution of (6.9).
s ), when r < ∞. Making use of uniform estimates
It remains to prove that f ∈ C([0, T ]; Bp,r
of f¯n , one can deduce that ∂t f belongs to L1 ([0, T ]; Bp,∞
−M ). Obviously, for fixed k, ∂ ∆ f
t k
1
belongs to L ([0, T ]; L ) so that each ∆k f is continuous in time with value in Lp . This implies
p
s ) for all k ∈ N. Since
Sk f ∈ C([0, T ]; Bp,r
X
∆k′ (f − Sk f ) = ∆k′ (∆k′′ f ),
|k ′′ −k ′ |≤1,k ′′ ≥k

then we have  X 1
′ r
kf − Sk f kBp,r
s ≤C 2k sr k∆k′ f kLp .
k ′ ≥k−1

By the same argument as in proof of (6.12), one may conclude that

Z t Z t
k∆k f (t)kLp ≤ k∆k f0 kLp + k∆k g(τ )kLp dτ + C ck (t)2−ks V (t) kf (τ )kBp,r
s dτ. (6.14)
0 0

It follows that
 X
r  1r
k′ s
s ) ≤C
kf − Sk f kL∞ (Bp,r 2 k∆k′ f0 kLp
T
k ′ ≥k−1
Z T X ′
r  1r
+C 2k s k∆k′ g(τ )kLp dτ
0 k ′ ≥k−1

27
 Z T  X 1
r
+ C kf kL∞ (Bp,r
s ) crk′ (τ ) V (τ )dτ
T
0 k ′ ≥k−1

The fact f0 ∈ Bp,r s ensures that the first term tends to zero as k goes to infinity. Since

g, V ∈ L1T (Bp,r
s ), one conclude that the terms in the integrals also tends to zero for almost every

t. This together with the Lebesgue dominated convergence theorem entails kf − Sk f kL∞ (Bp,r s )
T
s
tends to zero as k goes to infinity. Thus, we can conclude that f belongs to C([0, T ]; Bp,r ).
For the case r = ∞, by using the interpolation theorem, we deduce that for any t0 ∈ [0, T ]
and s′ ∈] − M, s[, there exists a constant θ ∈]0, 1[ depending on s′ such that
ku(t) − u(t0 )kB s′ ≤ ku(t) − u(t0 )kθB −M ku(t) − u(t0 )k1−θ
s
Bp,∞
p,∞ p,∞

≤2 ku(t) − u(t0 )kθB −M kuk1−θ

L∞ (B s .
p,∞ T p,∞ )

−M ) yields f ∈ C([0, T ]; B s ) for all s′ < s. Now,′

This together with the fact f ∈ C([0, T ]; Bp,∞ p,∞
s ). Indeed, for fixed φ ∈ S(Rn ), the low-high
we only need to prove that f ∈ Cw ([0, T ]; Bp,∞
decomposition technique leads to
hf (t), φi = hSk f (t), φi + h(Id − Sk )f (t), φi = hSk f (t), φi + hf (t), (Id − Sk )φi.
′
s ) gives that the function t 7→ hS f (t), φi is continuous.
Combining this with f ∈ C([0, T ]; Bp,∞ k
As for the second term, we have
|hf (t), (Id − Sk )φi| ≤ kf kBp,∞
s kφ − Sk φkB −s .
p′ ,1

It follows that hf (t), (Id − Sj )φi tends to zero uniformly on [0, T ] as k goes to infinity. This
s ).
means that f (t) ∈ Cw ([0, T ]; Bp,∞
Now, we focus on the proof of the uniqueness. Let f1 and f2 solve (6.9) with the same
initial datum. If we define δf = f1 − f2 , then δf solves

∂t δf + u · ∇δf − ∆h δf = 0.
This together with the estimate (6.10) ensures the uniqueness of solution of (6.9).

The last part of the appendix is devoted to the proof of losing a priori estimate for (1.2)
with ∇u ∈ L1 ([0, T ]; LL), where the LogLip space LL is the set of those functions f which
belong to S ′ and satisfy
k∇Sq f kL∞
kf kLL := sup < ∞. (6.15)
2≤q<∞ q+1
This estimate is the cornerstone to the proof of uniqueness in Theorem 1.2. In the sprite
of [7, 17], we prove linear losing a priori estimates for the general anisotropic system with
convection. More precisely, we have:
Proposition F.5. Let s1 ∈ [− 12 , 1[ and assume that s ∈]s1 , 1[. Let v satisfies the following
system (
∂t v + u · ∇v − ∆h v + ∇p = f + ge3 ,
(6.16)
divv = divu = 0
with initial data v0 ∈ H s and source terms f ∈ L1 ([0, T ]; H s ), g ∈ L2 ([0, T ]; H s−1 ). Assume
in addition that, for some h(t) ∈ L1 [0, T ] satisfying
kukLL ≤ h(t). (6.17)

28
Then there exists a constant C such that for any λ > C, T > 0 and
Z t
st := s − λ h(τ )dτ,
0

the following estimate holds

√ C Z t 

kv(t)kH st + k∇h vkL2 H st ≤ C(1 + t) exp h(τ )dτ kρ0 kH s + kf kL1 H s + kgkL2 H s−1 .
t λ 0
t t

Proof. Applying the operator ∆q to the system (6.16), we find that for all q ≥ −1, the fq
solves the following equations

∂t vq + Sq−1 u · vq − ∆h vq + ∇pq = fq + gq e3 + Fq (u, v)

with Fq (u, v) = Sq−1 u · ∇vq − ∆q (u · ∇v).

Taking the L2 -inner product to the above equation with vq and using divu = 0, we see that
Z Z Z
1d 2 2 3
kvq kL2 + k∇h vq kL2 = fq vq dx + gq vq dx + Fq (u, v)vq dx. (6.18)
2 dt
Assume that q ≥ 0. Applying the Bernstein and the Young inequalities, we can deduce that
Z
1 2
gq vq3 dx ≤C2−q kgq kL2 ∇vq3 L2 ≤ ∇vq3 L2 + C2−2q kgq k2L2
4
1 2 1 2
≤ ∇h vq3 L2 + ∂3 vq3 L2 + C2−2q kgq k2L2
4 4
1
≤ k∇h vq k2L2 + C2−2q kgq k2L2 ,
2
in the last line we have used the fact divv = 0.
Integrating the both sides of (6.18) with respect to t, we get for all q ≥ 0,

kvq k2L∞ L2 + k∇h vq k2L2 L2 ≤ kvq (0)k2L2 + 2 kfq k2L1 L2 + C2−2q kgq k2L2 L2 + 2 kFq (u, v)k2L1 L2 .
t t t t t

For q = −1, we merely have

Z t

kv−1 (t)kL2 ≤ kv−1 (0)kL2 + kf−1 (τ )kL2 + kg−1 (τ )kL2 + kF−1 (u, v)(τ )kL2 dτ.
0

On the other hand, by the Bernstein inequality, we know that

1
k∇h v−1 kL2 L2 ≤ Ct 2 kv−1 kL∞ L2 .
t t

Therefore, for all q ≥ −1, we have

kvq kL∞ L2 + k∇h vq kL2 L2

t t
√   (6.19)
≤2(1 + t) kvq (0)kL2 + kfq kL1 L2 + C2−q kgq kL2 L2 + kFq (u, v)kL1 L2 .
t t t

From a standard commutator estimate (see e.g. [7], Chap. 2), we know that for all ε ∈
]0, s+1
2 [, q ≥ −1 and t ∈ [0, T ]

2q(s−ε) kFq (u, v)(t)kL2 ≤ Ccq (2 + q)h(t) kv(t)kH s−ε with cq ∈ l2 (6.20)

29
for some constant C depending only on s.
Rt
Set st := s − λ 0 h(τ )dτ for t ∈ [0, T ]. Collecting (6.19) and (6.20) yields that
√  Rt
2(2+q)st kvq kL2 ≤ 2(1 + t) 2(2+q)s kvq (0)kL2 2−η(2+q) 0 h(τ )dτ
Z t Rt ′ ′
+ 2(2+q)sτ kfq (τ )kL2 2−η(2+q) τ h(τ )dτ dτ
0
Z t Rt 1 (6.21)
22(2+q)sτ 2−2q kgq (τ )k2L2 2−2λ(2+q) τ h(τ )dτ dτ
′ ′ 2
+
0
Z t Rt 
′ ′
+ Ccq (2 + q) h(τ )2−λ(2+q) τ h(τ )dτ kf (τ )kH sτ dτ .
0

For the last term of (6.21), we observe that

Z t Rt
h(τ ′ )dτ ′
Ccq (2 + q) h(τ )2−λ(2+q) τ kf (τ )kH sτ dτ
0
Z t Rt
1 h(τ ′ )dτ ′
≤Ccq d2−λ(2+q) τ sup kf (τ )kH sτ
λ log 2 0 τ ∈[0,t]
C Rt

=cq 1 − 2−λ(2+q) 0 h(τ )dτ sup kf (τ )kH sτ .
λ log 2 τ ∈[0,t]

Thus, multiplying 2qsτ and taking the l2 -norm of both sides of (6.21) over q ≥ −1, we get
√ 
sup kf (τ )kH sτ ≤ 2(1 + t) kf0 kH s + kf (τ )kL1 H sτ + kg(τ )kL2 H sτ −1
t t
τ ∈[0,t]
C 
+ sup kf (τ )kH sτ .
λ log 2 τ ∈[0,t]
√
2C(1+ t)
Choosing λ0 such that λ0 log 2 = 21 , we get by the Gronwall inequality that for any λ > λ0 ,
√ C
Rt  
h(τ )dτ
sup kf (t)kH sτ ≤ 2(1 + t)e λ 0 kf0 kH s + kf (τ )kL1 H sτ + kg(τ )kL2 H sτ −1 .
t t
t∈[0,t]

This implies the desired result.

Acknowledgements.

The authors are partly supported by the NSF of China No. 11171033. The authors would like
to thank Dr.L. Xue for his valuable discussion.

References

[1] H. Abidi and T. Hmidi, On the global well-posedness for Boussinesq System, J. Diff.
Equa., 233, 1 (2007), 199-220.

[2] H. Abidi and T. Hmidi, K. Sahbi, On the global regularity of axisymmetric Navier-
Stokes-Boussinesq system, Discrete Contin. Dyn. Syst., 29 (2011), 737-756.

30
 [3] H. Abidi, T. Hmidi and K. Sahbi, On the global well-posedness for the axisymmetric
Euler equations, Math. Ann., 347 (1) (2010), 15-41.

[4] A. Adhikari, C. Cao and J. Wu, The 2-D Boussinesq equations with vertiacal viscosity
and vertical diffusivity, J. Differential Equations, 249 (2010). 1078-1088.

[5] A. Adhikari, C. Cao and J. Wu, Global regularity results for the 2-D Boussinesq
equations with vertical disspation, J. Differential Equations, 251 (2011). 1637-1655.

[6] J. Ben Ameur and R. Danchin, Limite non visqueuse pour les fluids incompressibles
axisymétriques. Nonlinear partial differential equations and their apllations. Collége de
France seminar (Pairs, France, 1997-1998), vol. XIV, Stud. Math. Appl., vol.31, North-
Holland, Amsterdam (2002), 29-55.

[7] H. Bahouri, J.-Y. Chemin and R. Danchin, Fourier Analysis and Nonlinear Partial
Differential Equations, Grundlehren der mathematischen Wissenschaften 343, Springer-
Verlag, (2011).

[8] M. Cannone, Harmonic Analysis Tools for Solving the Incompressible Navier–Stokes
Equations, 161–244, in: Handbook of Mathematical Fluid Dynamics, vol. III, S. J.
Friedlander, D. Serre, eds., Elsevier, (2004).

[9] C. Cao and E. S. Titi, Global regularity criterion for the 3-D Navier-Stokes equations
involving one entry of the velocity gradient tensor, to appear Arch. Ration. Mech. Anal.

[10] C. Cao and J. Wu, Global regularity for the 2-D MHD equations with mixed partial
disspation and magnetic diffusion, Advances in Math., 226 (2011), 1803-1822.

[11] C. Cao and J. Wu, Global regularity results for the 2-D anisotropic Boussinesq equa-
tions with vertical disspation, arXiv:1108.2678v1

[12] D. Chae, Global regularity for the 2-D Boussinesq equations with partial viscous terms,
Advances in Math., 203, 2 (2006), 497-513.

[13] J.-Y. Chemin, Perfect Incompressible Fluids, Oxford University Press, (1998).

[14] J.-Y. Chemin, B. Desjardins, I. gallagher and E. Grenier, Fluids with anistrophic
viscosity, Modélisation Mathématique et Analyse Numérique, 34 (2000), 315-335.

[15] J.-Y. Chemin, B. Desjardins, I. gallagher and E. Grenier, Mathematical Geo-

physics: An Introduction to Rotating Fluids and to the Navier-Stokes Equations, Oxford
Unversity Press, (2006).

[16] R. Danchin, Axisymmetric incompressible flows with bounded vorticity, Russian Math.,
Surveys 62 (2007), no 3, 73-94.

[17] R. Danchin and M. Paicu, Global existence results for the anisotropic Boussinesq
system in dimension two. Mathematical Models and Methods in Applied Sciences, 21
(2011), 421-457.

[18] R. J. Diperna and P. L. Lions, Ordinary differnential equations, transport theory and
Sobolev spaces Inventiones mathematicae, 98 (1989), 511-547.

[19] T. Hmidi, Low Mach number limit for the isentropic Euler system with axisymmetric
initial data, arXiv:1109.5339v1

31
[20] T. Hmidi and F. Rousset, Global well-posedness for the Euler-Boussinesq system with
axisymmetric data. J. Functional Analysis, 260 (2011), 745-796.

[21] T. Hmidi and F. Rousset, Global well-posedness for the Navier-Stokes-Boussinesq sys-
tem with axisymmetric data. Ann. I. H. Poincaŕe-AN., 27 (2010), 1227-1246.

[22] T. Hmidi, S. Keraani and F. Rousset, Global well-posedness for Euler-Boussinesq

system with critical disspation. Comm, Partical Differential Equations, 36 (2011), 420-
445.

[23] T. Hmidi, S. Keraani and F. Rousset, Global well-posedness for Boussinesq-Navier-

Stokes system with critical disspation. J. Differential Equations, 249 (2010), 2147-2174.

[24] T. Hou and C. Li, Global well-posedness of the viscous Boussinesq equations, Discrete
and Cont. Dyn. Syst., 12 (2005), 1-12.

[25] P. G. Lemarié, Recent Developments in the Navier-Stokes Problem, CRC Press, (2002).

[26] C. Miao, J. Wu and Z. Zhang, Littlewood-Paley Theory and Applications to Fluid

Dynamics Equations, Monographs on Modern pure mathematics, No.142. Science Press,
Beijing, (2012).

[27] C. Miao and L. Xue, On the golbal well-posedness of a class of Boussinesq-Navier-

Stokes systems, Nonlinear Differ. Equ. Appl., 18 (2011), 707-735.

[28] J. Pedlosky, Geophysical Fluid Dynsmics, Springer Verlag, New-York, (1987).

[29] T. Shirota and T. Yanagisawa, Note on global existence for axially symmetric so-
lutions of the Euler system. Proc. Japan Acad. Ser. A Math. Sci., 70 (1994), no. 10,
299-304.

[30] E. M. Stein, Singular Integrals and Differentiability Properties of Functions, Princeton

University Press, Princeton, New Jersey, (1970).

[31] M. R. Ukhovskii and V. I. Yudovich, Axially symmetric flows of ideal and viscous
fluids filling the whole space, Prikl. Mat. Meh., 32 (1968), no. 1, 59-69.

32