GLOBAL REFINED FUJITA-KATO SOLUTION OF 3−D

INHOMOGENEOUS INCOMPRESSIBLE NAVIER-STOKES EQUATIONS

WITH LARGE DENSITY

HAMMADI ABIDI, GUILONG GUI, AND PING ZHANG

arXiv:2410.09386v1 [math.AP] 12 Oct 2024

Abstract. We investigate the global unique Fujita-Kato solution to the 3-D inhomogeneous
incompressible Navier-Stokes equations, (1.1), with initial velocity u0 being sufficiently small
1
in Ḃ2,∞
2
and with initial density being bounded from above and below. We first prove
the global existence of Fujita-Kato solution to the system (1.1) if we assume in addition
1
that the initial velocity u0 ∈ Ḣ 2 . While under the additional assumptions that the initial
1 1
velocity u0 ∈ Ḃ2,1
2
and initial density ρ0 satisfying ρ−1
0 − 1 ∈ Ḃ6,1 , we prove that kρ
2 −1
−
1k 1 and kuk 1 5 are controlled by the norm of the initial data.
e ∞ (R+ ;Ḃ 2 )
L e ∞ (R+ ;Ḃ 2 )∩L1 (R+ ;Ḃ 2 )
L
6,1 2,1 2,1
Our results not only improve the smallness condition in the previous references for the initial
velocity concerning the global Fujita-Kato solution of the system (1.1) but also improve the
exponential-in-time growth estimate for the solution in [2] to be the uniform-in-time estimate
(1.11).
Keywords: Inhomogeneous Navier-Stokes equations, Littlewood-Paley theory, Critical spaces,
Lorentz spaces.
AMS Subject Classification (2000): 35Q30, 76D03

1. Introduction
In this paper, we investigate the global unique Fujita-Kato solution of the following 3-D
inhomogeneous incompressible Navier-Stokes equations:


 ∂t ρ + div(ρu) = 0 (∀ (t, x) ∈ R+ × R3 ),

ρ(∂ u + u · ∇u) − ∆u + ∇Π = 0,
t
(1.1)

 div u = 0,


(ρ, u)|t=0 = (ρ0 , u0 ),
where the unknowns ρ and u = (u1 , u2 , u3 )T stand for the density and velocity of the fluid
respectively, and Π is a scalar pressure function, which guarantees the divergence free con-
dition of the velocity field. Such a system can be used to describe the mixture of several
immiscible fluids that are incompressible and with different densities. One may check [19] for
more background of this system.
Just as the classical Navier-Stokes equations (NS), which corresponds to the case when
ρ = 1 in (1.1), the system (1.1) has the following scaling-invariant property: if (ρ, u) solves
(1.1) with initial data (ρ0 , u0 ), then for any ℓ > 0,
def
(1.2) (ρ, u)ℓ (t, x) = (ρ(ℓ2 ·, ℓ·), ℓu(ℓ2 ·, ℓ·))
is also a solution of (1.1) with initial data (ρ0 (ℓ·), ℓu0 (ℓ·)). We call such functional spaces as
critical spaces if the norms of which are invariant under the scaling transformation (1.2).
In [14], Fujita and Kato proved the following celebrated result for the classical Navier-
1
Stokes equations: given solenoidal vector field u0 ∈ Ḣ 2 (R3 ) with ku0 k 21 ≤ ε0 for ε0 > 0
Ḣ
1
being sufficiently small, (NS) has a unique global solution u ∈ C([0, +∞); Ḣ 2 )∩L4 (R+ ; Ḣ 1 )∩
1
2 H. ABIDI, G. GUI, AND P. ZHANG

3
L2 (R+ ; Ḣ 2 ). Cannone, Meyer, and Planchon [7] extended the result in [14] for initial data
−1+ 3
belonging to the general critical Besov space, namely, u0 ∈ Ḃp,∞ p .
Danchin [11] first established the global well-posedness of the system (1.1) with initial data
in the almost critical Sobolev spaces. After the works [1, 3, 10] in the critical framework,
Danchin and Mucha [12] eventually proved the global well-posedness of the system (1.1) with
−1+ pd
initial density being close enough to a positive constant in the multiplier space of Ḃp,1 (Rd )
−1+ d
and initial velocity being small enough in Ḃp,1 p (Rd ) for 1 ≤ p < 2d. The work of [2] is
the first to investigate the global well-posedness of the system (1.1) with initial data in the
critical spaces and yet without the size restriction on the initial density. Huang, Paicu and
the third author [17] studied the global existence of weak solutions to the system (1.1) with
−1+ d
small initial data ρ0 − 1 ∈ L∞ (Rd ) and u0 ∈ Ḃp,r p (Rd ) with p ∈ (1, d), r ∈ (1, +∞), and
the uniqueness of such solution was only proved under additional regularity assumptions on
the initial velocity or on the initial density. The third author of this paper [22] proved the
global existence of weak solutions to the system (1.1) with initial density being bounded from
above and below, and with initial velocity being sufficiently small in the critical Besov space
1
2
Ḃ2,1 (R3 ). This solution corresponds to the Fujita-Kato solution of the classical Navier-Stokes
equations. The uniqueness of such solution was proved lately by Danchin and Wang [13].
One may check [13] and references therein for the progresses in this direction.
Before proceeding, we recall the following result concerning the global existence and unique-
ness of Fujita-Kato solution to the system (1.1) from [13, 22].
1
2
Theorem 1.1 ([13, 22]). Let (ρ0 , u0 ) satisfy u0 ∈ Ḃ2,1 (R3 ) with div u0 = 0, and
(1.3) 0 < c0 ≤ ρ0 (x) ≤ C0 < +∞ (∀ x ∈ R3 ).
Then there exists a constant ε0 > 0 depending only on c0 , C0 such that if
(1.4) ku0 k 2
1 ≤ ε0 ,
Ḃ2,1

the system (1.1) has a unique global solution (ρ, u, ∇Π) with ρ ∈ L∞ (R+ × R3 ) and u ∈
1 3
e 2 (R+ ; Ḃ 2 ) which satisfies
C([0, +∞); Ḃ 2 ) ∩ L
2,1 2,1

(1.5) c0 ≤ ρ(t, x) ≤ C0 < +∞ (∀ (t, x) ∈ R+ × R3 ),

and
1
kuk 1 + kuk 3 + kt 2 (∇u, Π)k 1
e ∞ (R
L + 2 )
;Ḃ2,1 e 2 (R
L + 2 )
;Ḃ2,1 L2 (R+ ;Ḃ6,1
2 )

1 1 1
+ kt 2 uk 3 + kt 2 ut k 1 + kt 2 (∇2 u, ∇Π)kL2 (R+ ;L3 )
e ∞ (R+ ;Ḃ 2 )
L e 2 (R+ ;Ḃ 2 )
L
2,1 2,1

+ ktut k 1 + ktDt uk 3 ≤ Cku0 k 1 ,

e ∞ (R+ ;Ḃ 2 )
L e 2 (R+ ;Ḃ 2 )
L 2
Ḃ2,1
2,1 2,1
3
def e 2 (R+ ; Ḃ 2 ) can be
where Dt u = (∂t + u · ∇)u, and the definition of Chemin-Lerner space L 2,1
found in Definition A.2.
1
Very recently, instead of using ∇ u2 ∈ L1 (0, T ; L∞ ), Hao et al. [16] used t 2 ∇ u2 ∈
L2 (0, T ; L∞ ) to derive the uniqueness of solutions to the system (1.1). Precisely, they proved
the following interesting uniqueness result:
Theorem 1.2 (Theorem 1.3 in [16]). Let (ρ1 , u1 , Π1 ) and (ρ2 , u2 , Π2 ) be two solutions of the
system (1.1) on [0, T ] × R3 corresponding to the same initial data. Assume in addition that:
√
• ρ1 (u2 − u1 ) ∈ L∞ (0, T ; L2 (R3 ));
• ∇ (u2 − u1 ) ∈ L2 ((0, T ) × R3 );
 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 3

1
• t 2 ∇ u2 ∈ L2 (0, T ; L∞ );
• ∇u2 ∈ L4 (0, T ; L2 );
3
• t 4 Dt u2 ∈ L2 (0, T ; L2 ).
Then, (ρ1 , u1 , ∇Π1 ) ≡ (ρ2 , u2 , ∇Π2 ) on [0, T ] × R3 .
Based on Theorem 1.2 and motivated by [22], Hao et al. [16] proved the global unique
solution to the system (1.1) with initial density being bounded from above and below and
1
with initial velocity being sufficiently small in Ḣ 2 .
Compared with the result of global Fujita-Kato solution to the classical 3-D Navier-Stokes
1
2
equations, the smallness condition of the initial velocity u0 in Ḃ2,∞ seems more natural for
2
the system (1.1) in the L framework. The goal of this paper is to show the global unique
Fujita-Kato solution of (1.1) with the initial data in the critical Besov spaces but without
smallness assumption on the initial density ρ0 yet with initial velocity u0 being small enough
1
2
in Ḃ2,∞ .
Our main results state as follows.
1
Theorem 1.3. Let (ρ0 , u0 ) satisfy (1.3) and u0 ∈ Ḣ 2 (R3 ) with div u0 = 0. Then there exists
a constant c > 0 depending only on c0 , C0 such that if
(1.6) ku0 k 1
2
≤ c,
Ḃ2,∞

the system (1.1) has a global unique solution (ρ, u, ∇Π) with ρ ∈ Cw ([0, ∞); L∞ ) and u ∈
1 3
C([0, +∞); Ḣ 2 ) ∩ L2 (R+ ; Ḣ 2 ) which satisfies (1.5) and

(1.7) kuk 1 + kuk 3 . ku0 k 1 for any r ∈ [2, +∞],

e ∞ (R+ ;Ḃ 2 )
L e 2 (R+ ;Ḃ 2 )
L 2
Ḃ2,r
2,r 2,r

and
1
k∇ ukL4 (R+ ;L2 ) + kt 4 ∇ ukL∞ (R+ ;L2 ) . ku0 k 1
2
,
Ḃ2,∞
1 1
kt 2 uk 3 + kt 2 ∂t uk 1 . ku0 k 1 ∀ r ∈ [2, +∞],
e ∞ (R+ ;Ḃ 2 )
L e 2 (R+ ;Ḃ 2 )
L 2
Ḃ2,r
(1.8) 2,r 2,r
1 3 1
kt 4 ut kL2 (R+ ;L2 ) + kt 4 (∂t u, ∇2 u, Dt u)kL∞ (R+ ;L2 ) + kt 2 ∇ ukL2 (R+ ;L∞ )
3 3
+ kt 4 ∇(∂t u, Dt u)kL2 (R+ ;L2 ) + kt 4 (∇2 u, ∇Π)kL2 (R+ ;L6 ) . ku0 k 1 .
Ḣ 2

Theorem 1.4 (Regularity for Lipschitz flows). Let (ρ0 , u0 ) satisfy (1.3) and
1 1
(1.9) ρ−1 2 3 2 3
0 − 1 ∈ Ḃ6,1 (R ), u0 ∈ Ḃ2,1 (R ) with div u0 = 0.
Then there exists a constant c > 0 depending only on c0 , C0 such that if (1.6) holds, the
system (1.1) has a unique global solution (ρ, u, ∇Π) which satisfies (1.5) and
1 1 5

(1.10) (ρ−1 − 1, u) ∈ C([0, ∞); Ḃ6,1
2 2
) × C([0, ∞); Ḃ2,1 ) ∩ L1 (R+ ; Ḃ2,1
2
) ,
and
−1
(1.11) kρ − 1kLe ∞ (R+ ;Ḃ 12 + kuk 1 + kuk 5 . kρ−1
0 − 1k 1 + ku0 k 1 .
e ∞ (R+ ;Ḃ 2 ) L1 (R+ ;Ḃ2,1
2 ) 2 2
6,1 ) L 2,1 Ḃ6,1 Ḃ2,1

Remark 1.1. (1) In our previous paper [2], we proved that if (ρ0 , u0 ) satisfy (1.3), (1.4),
3
ρ−1
0
2
− 1 ∈ Ḃ2,1 and div u0 = 0, the system (1.1) has a unique global solution (ρ, u) so
that for any t > 0,
kρ−1 − 1k 3 + kuk 1 + kuk 5
e ∞ (Ḃ 2 )
L e ∞ (Ḃ 2 )
L L1t (Ḃ2,1
2 )
t 2,1 t 2,1
(1.12)
√

. kρ−1
0 − 1k 3
2
+ ku0 k 1
2
exp C t .
Ḃ2,1 Ḃ2,1
4 H. ABIDI, G. GUI, AND P. ZHANG

We emphasize that Theorem 1.4 improves not only the smallness condition from (1.4)
to (1.6) but also the exponential-in- time growth estimate (1.12) to the one (1.11)
which is uniform with respect to time.
(2) We remark that the key ingredient in the proof of Theorem 1.4 is to derive the a priori
estimate:
(1.13) k∇ukL1 (R+ ;L∞ ) . ku0 k 1
2
.
Ḃ2,1

The proof of (1.13) is motivated by [13] and yet simpler here. Once again our small-
ness condition for the velocity field (1.6) improves the one (1.4) in [13].
Scheme of the proof and organization of the paper.
In Section 2, we present some laws of product and commutator’s estimates, which will be
used in the subsequent sections.
In Section 3, we present the proof of Theorem 1.3.
In Section 4, we present the proof of Theorem 1.4.
Finally in the Appendix A, we collect some basic facts on Littlewood-Paley theory and
Lorentz spaces.

Let us complete this section with the notations that we are going to use in this context.
Notations: Let A, B be two operators, we denote [A, B] = AB − BA, the commutator
between A and B. For a . b, we mean that there is a uniform constant C, which may be
different on different lines, such that a ≤ Cb.
For X a Banach space and I an interval of R, we denote by C(I; X) the set of continuous
functions on I with values in X. For q ∈ [1, +∞], the notation Lq (I; X) stands for the set of
measurable functions on I with values in X, such that t 7−→ kf (t)kX belongs to Lq (I).
We always denote {cj,r }j∈Z a nonnegative generic element in the unit sphere of ℓr (Z) and
(cj )j∈Z (resp. (dj )j∈Z ) a nonnegative generic element in the unit sphere of ℓ2 (Z) (resp. ℓ1 (Z)).

2. Preliminaries
The proofs of Theorems 1.3 and 1.4 require Littlewood-Paley theory, which we recall in
Appendix A for the convenience of readers. Below we shall apply the basic facts in Appendix
A to study some estimates, which will be used in the subsequent sections. We first present
the following law of product, the proof of which is given for the sake of completeness.
1
2
Lemma 2.1. Let f ∈ Ḃ2,∞ (R3 ) and g ∈ Ḣ 1 (R3 ), then one has
(2.1) kf gkL2 . kf k 1
2
kgkḢ 1 .
Ḃ2,∞

Proof. By virtue of Bony’s decomposition (A.4), we decompose f g into three terms:

(2.2) f g = Tf g + Tg f + R(f, g).
We first observe from Lemma A.1 that
X X 3
(2.3) kṠk−1 f kL∞ . k∆˙ ℓ f kL∞ . ˙ ℓ f kL2 . 2k kf k
2 2 ℓ k∆ 1 ,
2
Ḃ2,∞
ℓ≤k−2 ℓ≤k−2
from which, we infer
X
k∆˙ q Tf gkL2 . ˙ k gkL2
kṠk−1 f kL∞ k∆
|q−k|≤4
X
. ck kgkḢ 1 kf k 1
2
. cq kf k 1
2
kgkḢ 1 .
Ḃ2,∞ Ḃ2,∞
|q−k|≤4

Henceforth we always denote (cq )q∈Z to be a nonnegative generic element of ℓ2 (Z) so that
P 2
q∈Z cq = 1.
 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 5

Similarly, due to
k
kṠk−1 gkL∞ . ck 2 2 kgkḢ 1 ,
one has
X
˙ q Tg f kL2 .
k∆ ˙ k f kL2 . cq kf k
kṠk−1 gkL∞ k∆ 1 kgkḢ 1 .
2
Ḃ2,∞
|q−k|≤4

Finally, we deduce from Lemma A.1 that

X ė gk 2
k∆˙ q R(f, g)kL2 .2 32 q ˙ k f kL2 k∆
k∆ k L
k≥q−3
3 X 3
.kgkḢ 1 kf k 1
2
22q ck 2− 2 k . cq kf k 1
2
kgkḢ 1 .
Ḃ2,∞ Ḃ2,∞
k≥q−3

By summarizing the above estimates, we reach (2.1). 

1 3
2
Lemma 2.2. Let f ∈ Ḃ2,∞ 2
∩ Ḃ2,∞ (R3 ), then there holds

(2.4) kf 2 k 2
1 . kf k 2
1 kf k 3
2
.
Ḃ2,∞ Ḃ2,∞ Ḃ2,∞

Proof. In view of (A.4), we write

(2.5) f 2 = 2Tf f + R(f, f ).
We first get from (2.3) that
X
˙ q Tf f kL2 .
k∆ ˙ k f kL2
kṠk−1 f kL∞ k∆
|q−k|≤4
X k q
. kf k 1
2
kf k 3
2
2− 2 . 2− 2 kf k 2
1 kf k 3
2
.
Ḃ2,∞ Ḃ2,∞ Ḃ2,∞ Ḃ2,∞
|q−k|≤4

On the other hand, we deduce from Lemma A.1 that

X ė f k 2
k∆˙ q R(f, f )kL2 . 2 32 q ˙ k f kL2 k∆
k∆ k L
k≥q−3
3 X q
. kf k 1
2
kf k 3
2
22q 2−2k . 2− 2 kf k 2
1 kf k 2
3 .
Ḃ2,∞ Ḃ2,∞ Ḃ2,∞ Ḃ2,∞
k≥q−3

By summarizing the above estimates, we obtain (2.4). 

−1+ 3 1+ 3
Lemma 2.3. Let p ∈ [2, +∞), f ∈ L∞ p 3 e1 p 3
t (Ḃp,∞ (R )) ∩ Lt (Ḃp,∞ (R )), then we have

(2.6) kf 2 k p
3 . kf k 3
−1+ p kf k 1+ 3 .
e 1 (Ḃp,∞ ) L∞ e 1 (Ḃp,∞p )
L t t (Ḃp,∞ ) L t

Proof. Observing that

X 3
kṠk−1 f kL∞ ∞ .
t (L )
2 p ℓ k∆
˙ ℓ f (t)kLp
L∞
. 2k kf k 3
−1+ p ,
t L∞
ℓ≤k−2 t (Ḃp,∞ )

so that one has

X
˙ q Tf f k 1 p .
k∆ kṠk−1 f kL∞ ˙ kf k 1 p
∞ k∆
Lt (L ) t (L ) Lt (L )
|q−k|≤4
X − p3 k
. kf k 1+ 3 kf k 3
−1+ p 2
e 1 (Ḃp,∞p ) L∞
L t t (Ḃp,∞ ) |q−k|≤4
− 3p q
.2 kf k 1+ 3 kf k 3
−1+ p .
e 1 (Ḃp,∞p ) L∞
L t t (Ḃp,∞ )
6 H. ABIDI, G. GUI, AND P. ZHANG

On the other hand, we observe from Lemma A.1 that

3 X ė f k
k∆˙ q R(f, f )k 1 p . 2 p q ˙ k f kL∞ (Lp ) k∆
k∆
Lt (L ) t k L1t (Lp )
k≥q−3
3 X 6
. kf k 3
−1+ p kf k 1+ 3 2pq 2− p k
L∞ e 1 (Ḃp,∞p )
t (Ḃp,∞ ) L t k≥q−3
− p3 q
.2 kf k 3
−1+ p kf k 1+ 3 .
e ∞ (Ḃp,∞
L ) e 1 (Ḃp,∞p )
L
t t

By combining the above estimates with (2.5), we achieve (2.6). 

3

Lemma 2.4. Let p ∈ [2, 3] and λ ∈ (3, +∞) satisfy 3p − λ3 ≤ 1. Let f ∈ L∞ ∞ 3
t Ḃλ,∞ ∩ L (R )
λ

3 3
−1+ p3

e 1t Ḃ −1+
and g ∈ L λp
p
−λ
∩ Ḃ p,∞ (R3 ) . Then one has
λ−p
,1

kf gk 3
−1+ p . kf kL∞ ∞ kgk
t (L )
3
−1+ p + kf k 3 kgk 3−3
−1+ p
.
(2.7) e 1 (Ḃp,∞
L t ) e 1 (Ḃp,∞
L t ) L∞ λ
t (Ḃλ,∞ ) L1t Ḃ λp
λ
,1
λ−p

Proof. We first get, by applying Proposition A.1, that for any q ∈ Z,

˙ q Tf gk 1 p . 2 (1− 3p )q
k∆ Lt (L ) kf kL∞ ∞ kgk
t (L )
3
−1+ p ,
e 1 (Ḃp,∞
L )
t
3
˙ q R(f, g)k 1 p . 2(1− p )q kf k
k∆ 3 kgk 3−3
.
Lt (L ) λ
−1+ p
λ
L∞
t (Ḃλ,∞ ) L1t Ḃ λp
,1
λ−p

3 3
While due to p − λ ≤ 1, we have
 
X 1− p3 + λ
3
ℓ
kṠk−1 gk λp .kgk −1+ 3
p−λ
3
2 dℓ
L1t (L λ−p ) L1t Ḃ λp
,1 ℓ≤k−2
λ−p
 
3
1− p3 + λ k
.2 kgk 3−3
−1+ p
,
λ
L1t Ḃ λp
,1
λ−p

so that one has

X
˙ q Tg f k 1 p .
k∆ kṠk−1 gk λp
˙ k f kL∞ (Lλ )
k∆
Lt (L ) t
L1t (L λ−p )
|q−k|≤4
 
X 1− p3 k
. kf k 3 kgk −1+ 3 − 3 2
L∞ λ
t (Ḃλ,∞ ) L1t (Ḃ λ p p λ )
,1 |q−k|≤4
λ−p

(1− p3 )q
. kf k 3 kgk −1+ 3 − 3 2 .
L∞ λ
t (Ḃλ,∞ ) L1t (Ḃ λ p p λ )
,1
λ−p

Henceforth we always denote (dq )q∈Z to be a nonnegative generic element of ℓ1 (Z) so that
P
q∈Z dq = 1.
By summarizing the above estimates and using (2.2), we achieve (2.7). 
1
1 (R3 )) ∩ L∞ (Ḃ 2 (R3 )) with div u = 0. Then
Lemma 2.5. Let 2 ≤ p ≤ ∞ and u ∈ L1t (Ḃ∞,1 t 2,∞
there holds
X q−1+ 3 
2 p ˙ q , u · ∇]uk 1 p . kuk 1 1 kuk
k[∆ .
(2.8) Lt (L ) Lt (Ḃ∞,1 ) 1
L∞ (Ḃ 2 )
q∈Z t 2,∞
 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 7

Proof. By applying Bony’s decomposition (A.4) and the divergence free condition of u, we
write
4
˙ ˙ j ˙ j

′ j ˙ def X (i)
(2.9) [∆q , u · ∇]u = ∆q ∂j R(u , u)) + ∆q T∂j u u − T∆˙ q ∂ u u + [∆q , Tuj ]∂j u = Iq .
j
i=1

Here repeated upper and lower indices mean summation from 1 to 3. Let us handle term by
term above.
We first deduce from Lemma A.1 that
 
q 3−3 X ė
kIq(1) kL1t (Lp ) . 2q 2 2 p k∆k ukL∞ ˙ k uk 1 ∞
2 k∆
t (L ) Lt (L )
k≥q−3
 
q 5
−3 X 3
. kukL1 (Ḃ 1 ) kuk 1
2
2 2 p
dk 2− 2 k
t ∞,1 L∞
t (Ḃ2,∞ )
k≥q−3
 
q 1− p3
. dq 2 kukL1 (Ḃ 1 kuk 1 .
t ∞,1 ) L∞ 2
t (Ḃ2,∞ )

Whereas it follows from Lemma A.1 that

X  5 − 3 ℓ  
2− p3 k
(2.10) kṠk−1 ukL∞ p . 2 2 p k∆ ˙ ℓ ukL2 . kuk 1 2 ,
t (L ) L∞ 2
t L∞
t (Ḃ2,∞ )
ℓ≤k−2

from which, we infer

X
kIq(2) kL1t (Lp ) . kṠk−1 ukL∞ ˙ k uk 1 ∞
p k∆
t (L ) Lt (L )
|q−k|≤4
 
X 1− p3 k
. kukL1 (Ḃ 1 kuk 1 dk 2
t ∞,1 ) L∞ 2
t (Ḃ2,∞ )
|q−k|≤4
 
q 1− p3
. dq 2 kukL1 (Ḃ 1 kuk 1 .
t ∞,1 ) L∞ 2
t (Ḃ2,∞ )

(3) P ˙ ˙ j
Observing that Iq =− k≥q−3 Ṡk+2 ∆q ∂j u∆k u , one has
X
˙ q ∂j uk 1 p
kIq(3) kL1t (Lp ) . k∆ ˙ k uj k 1 ∞
k∆
Lt (L ) Lt (L )
k≥q−3
 
q 2− p3 X
. kukL1 (Ḃ 1 ) kuk 1
2
2 dk 2−k
t ∞,1 L∞
t (Ḃ2,∞ )
k≥q−3
 
q 1− p3
. dq 2 kukL1 (Ḃ 1 kuk 1 .
t ∞,1 ) L∞ 2
t (Ḃ2,∞ )

For the last term in (2.9), owing to the properties of spectral localization of the Littlewood-
(4) P ˙ j ˙
Paley decomposition, we have Iq = |k−q|≤4 [∆q , Ṡk−1 u ]∆k ∂j u. Then we deduce from
Lemma A.2 and (2.10) that
X
kIq(4) kL1t (Lp ) . 2k−q k∆˙ k uk 1 ∞ kṠk−1 ∇ukL∞ (Lp )
Lt (L ) t
|k−q|≤4
 
X 2− p3 k
. kukL1 (Ḃ 1 ) kuk 1
2
2−q dk 2
t ∞,1 L∞
t (Ḃ2,∞ )
|k−q|≤4
 
q 1− p3
. dq 2 kukL1 (Ḃ 1 kuk 1 .
t ∞,1 ) L∞ 2
t (Ḃ2,∞ )

By summarizing the above estimates and using (2.9), we arrive at (2.8). 

8 H. ABIDI, G. GUI, AND P. ZHANG

3
def e∞
Lemma 2.6. Let 2 ≤ p < 3 and 3 < λ < ∞ with λ1 + 13 ≥ p1 . Let a = ρ−1 −1 ∈ L 3
t (Ḃλ,1 (R ))
λ

and f ∈ L1t (L3 (R3 )). Then we have

X q−1+ 3 
2 p ˙ q P, ρ−1 ]f k 1 p . kak
k[∆ 3 kf kL1 (L3 ) .
(2.11) Lt (L )
Le ∞ (Ḃ λ ) t
q∈Z t λ,1

Proof. Similar to (2.9), we get, by applying Bony’s decomposition (A.4), that

X 4
˙ q P, ρ−1 ]f = [∆
(2.12) [∆ ˙ q P, a]f = ∆
˙ q PR(a, f ) + ∆
˙ q PTf a − T ′˙ a − [∆˙ q P, Ta ]f def
= Jq(i) .
∆ q Pf
i=1
1 1 1
Due to λ + 3 ≥ p, we deduce from Lemma A.1 that
 
3q 1
+1−1 X ė f k
kJq(1) kL1t (Lp ) .2 3 λ p ˙ k akL∞ (Lλ ) k∆
k∆ t
k L1t (L3 )
k≥q−3
 
3q 1
+1−1 X 3
. kf kL1t (L3 ) kak 3 2 3 λ p
dk 2−k λ
e ∞ (Ḃ λ )
L t λ,1 k≥q−3
 
q 1− p3
. dq 2 kf kL1t (L3 ) kak 3 .
e ∞ (Ḃ λ )
L t λ,1

Along the same line, we infer

 
X 1
3q 31 + λ − p1 X
(3) p
kJq kL . ˙
k∆k aṠk+2 ∆q f kL1t (Lp ) . 2 ˙ qf k 1 3
k∆ ˙ k akL∞ (Lλ )
k∆
Lt (L ) t
k≥q−3 k≥q−3
 
q 1− p3
. dq 2 kf kL1t (L3 ) kak 3 .
e ∞ (Ḃ λ )
L t λ,1

Notice that kṠk−1 f kL3 ≤ kf kL3 , we deduce from Lemma A.1 that
X
kJq(2) kL1t (Lp ) . ˙ k ak
kṠk−1 f kL1t (L3 ) k∆ 3p
∞ Lt (L 3−p )
|q−k|≤4
   
X k 1− p3 q 1− p3
. kak 3 kf kL1t (L3 ) dk 2 . dq 2 kf kL1t (L3 ) kak 3 .
e ∞ (Ḃ λ )
L e ∞ (Ḃ λ )
L
t λ,1 t λ,1
|q−k|≤4

Finally due to
   
X 2− p3 ℓ 2− p3 k
k∇Ṡk−1 ak 3p . kak 3 dℓ 2 . dk 2 kak 3 ,
L∞ 3−p ) e ∞ (Ḃ λ ) e ∞ (Ḃ λ )
t (L L t λ,1 L t λ,1
ℓ≤k−2

we find
X
kJq(4) kL1t (Lp ) . ˙ k f k 1 3 k∇Ṡk−1 ak
2−q k∆ 3p
Lt (L ) 3−p )
L∞
t (L
|q−k|≤4
 
X 2− p3 k
−q
. kf kL1t (L3 ) kak 3 dk 2 2
e ∞ (Ḃ λ )
L t λ,1 |q−k|≤4
 
q 1− p3
. dq 2 kf kL1t (L3 ) kak 3 .
e ∞ (Ḃ λ )
L t λ,1

By summarizing the above estimates and using (2.12), we obtain

 
˙ q P, ρ−1 ]f k 1 p . dq 2q 1− p3
k[∆ Lt (L ) kf kL1t (L3 ) kak 3 ,
e ∞ (Ḃ λ )
L t λ,1

which ensures (2.11). This completes the proof of the lemma. 

 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 9

3. The proof of Theorems 1.3

Motivated by Section 3 of [22], we shall first derive the a priori estimates which will be
used in the proof of Theorem 1.3. Then we present the proof of Theorems 1.3 at the end
of this section. Toward this, we first present the following basic energy estimates for the
linearized equation of (1.1).
Proposition 3.1. Let (ρ, u) be a smooth solution of the system (1.1) on [0, T ∗ ), for any
j ∈ Z, (uj , ∇Πj ) solves

 + 3
ρ(∂t uj + u · ∇uj ) − (∆uj − ∇Πj ) = 0 in R × R ,
(3.1) div uj = 0,

 ˙ j u0 .
uj |t=0 = ∆
Then under the assumptions of (1.3) and (1.6), there hold (1.5) and
kuj kL∞ ˙ j u0 kL2 ,
2 + k∇ uj kL2 (L2 ) . k∆
T (L ) T
(3.2) 2 j ˙
k∇ uj kL∞ 2 + k(∂t uj , ∇ uj , ∇Πj )kL2 (L2 ) . 2 k∆
T (L )
j u0 kL2 , ∀ j ∈ Z,
T

˙ j u0 kL2 1
(3.3) kt−α ∇uj kL2 (L2 ) . 22jα k∆ (∀ α ∈ [0, ), j ∈ Z),
T 2
and
(3.4) kuk 1 + kuk 3 . ku0 k 1 .
L∞ 2 e 2 (Ḃ 2 ) 2
T (Ḃ2,∞ ) L T 2,∞ Ḃ2,∞
1
2
If in addition, u0 ∈ Ḃ2,r for 1 ≤ r ≤ +∞, then there holds

(3.5) kuk 1 + kuk 3 . ku0 k 1 .

e ∞ (Ḃ 2 )
L e 2 (Ḃ 2 )
L 2
Ḃ2,r
T 2,r T 2,r

In particular, one has

1
(3.6) kuk e ∞ 1 + kuk 3 + kt− 4 ∇ukL2 (L2 ) . ku0 k 1 .
LT (Ḣ 2) L2T (Ḣ 2 ) T Ḣ 2

Proof. We first deduce from the classical theory of transport equation and (1.3) that there
holds (1.5) for 0 ≤ t < T ∗ .
Since (uj , ∇Πj ) solves (3.1), we deduce from the classical uniqueness result of local smooth
solution to (1.1) that
X X
(3.7) u = uj and ∇ Π = ∇ Πj in Sh′ .
j∈Z j∈Z

By taking L2 inner product of the momentum equation of (3.1) with uj and using the
transport equation of (1.1), we find
Z
1 d
ρ|uj |2 dx + k∇ uj k2L2 = 0, .
2 dt R3
By integrating the above equation over [0, t] for t ∈ [0, T ∗ ), we find
1 √ 1 √ ˙
k ρ uj (t)k2L2 + k∇ uj k2L2 (L2 ) = k ρ0 ∆ 2
j u0 kL2 ,
2 t 2
from which, (1.3) and (1.5), we deduce that there exists a positive constant c1 such that
(3.8) kuj kL∞ ˙ j u0 kL2 .
2 + c1 k∇ uj kL2 (L2 ) . k∆
t (L ) t

In particular, one has

j
−
(3.9) kuj kL∞ 2 + c1 k∇ uj kL2 (L2 ) . 2 2 ku0 k
t (L ) t 2
1 .
Ḃ2,∞
10 H. ABIDI, G. GUI, AND P. ZHANG

Whereas by taking L2 inner product of the momentum equation of (3.1) with ∂t uj , we find
Z
1 d 2 √ 2 √
k∇ uj kL2 + k ρ ∂t uj kL2 = − (ρ u · ∇uj ) · ∂t uj dx . ku · ∇uj kL2 k ρ ∂t uj kL2 ,
2 dt R3
from which and (2.1), we infer
1 d √ √
(3.10) k∇ uj k2L2 + k ρ ∂t uj k2L2 . kuk 21 k∇2 uj kL2 k ρ ∂t uj kL2 .
2 dt Ḃ2,∞

While in view of (3.1), we write

(
∆uj − ∇Πj = ρ(∂t uj + u · ∇uj ),
(3.11)
div uj = 0,
from which and the classical theory on Stokes operator, we infer
√
k∆uj kL2 + k∇Πj kL2 . k ρ ∂t uj kL2 + ku · ∇uj kL2 .

By applying (2.1) once again, we obtain

√
(3.12) k(∇2 uj , ∇Πj )kL2 ≤ C1 (k ρ ∂t uj kL2 + kuk 2
1 k∇2 uj kL2 ).
Ḃ2,∞

By combining (3.10) with (3.12), we find

1 d √
k∇ uj k2L2 +2c0 k( ρ ∂t uj , ∇2 uj , ∇Πj )k2L2
2 dt  
√
≤ C2 kuk 21 k∇2 uj kL2 k ρ ∂t uj kL2 + kuk2 1 k∇2 uj k2L2 .
Ḃ2,∞ 2
Ḃ2,∞

Applying Young’s inequality yields

d √
(3.13) k∇ uj k2L2 + 2c0 k( ρ ∂t uj , ∇2 uj , ∇Πj )k2L2 ≤ C0 kuk2 1 k∇2 uj k2L2 ,
dt 2
Ḃ2,∞

Let us denote
n c0 1 o
def
(3.14) T1 = sup t ∈ (0, T ∗ ) : kuk 1 ≤( )2 .
L∞ 2
t (Ḃ2,∞ ) C0
We will claim that T1 = T ∗ .
In fact, by contradiction, if T1 < T ∗ , then for any t ∈ [0, T1 ], we deduce from (3.13) that
d √
(3.15) k∇ uj k2L2 + c0 k( ρ ∂t uj , ∇2 uj , ∇Πj )k2L2 ≤ 0.
dt
By integrating the above inequality over [0, t] for t ≤ T1 , one has
2 j ˙
(3.16) k∇ uj kL∞ 2 + k(∂t uj , ∇ uj , ∇Πj )kL2 (L2 ) . 2 k∆
t (L ) t
j u0 kL2 ,

which implies
j
2
(3.17) k∇ uj kL∞ 2 + k(∂t uj , ∇ uj , ∇Πj )kL2 (L2 ) ≤ C2 2 ku0 k
t (L ) t 2
1 .
Ḃ2,∞

On the other hand, thanks to (3.7), we deduce from Lemma A.1 that
X
k∆˙ j ukL∞ (L2 ) + k∇∆
˙ j uk 2 2 . ˙ j uj ′ kL∞ (L2 ) + k∇∆
(k∆ ˙ j uj ′ k 2 2 )
t L t (L ) t Lt (L )
j ′ ∈Z
X

. k∆ ˙ j uj ′ k 2 2
˙ j uj ′ kL∞ (L2 ) + k∇∆
t Lt (L )
j ′ ≥j
X

+ 2−j k∇∆ ˙ j uj ′ k 2 2 ,
˙ j uj ′ kL∞ (L2 ) + k∇2 ∆
t L t (L )
j ′ ≤j
 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 11

from which, we infer

X

˙ j ukL∞ (L2 ) + k∇∆
k∆ ˙ j uk 2 2 . kuj ′ kL∞ 2 + k∇uj ′ kL2 (L2 )
t Lt (L ) t (L ) t
j ′ ≥j
(3.18) X
+ 2−j (k∇ uj ′ kL∞ 2
2 + k∇ uj ′ kL2 (L2 ) ).
t (L ) t
j ′ ≤j

By substituting (3.9) and (3.17) into (3.18), we obtain

X j′ X j′
˙ j ukL∞ (L2 ) + k∇∆
k∆ ˙ j uk 2 2 . 2− 2 ku0 k 1 + 2−j 2 2 ku0 k 1
t Lt (L ) 2 2
Ḃ2,∞ Ḃ2,∞
j ′ ≥j j ′ ≤j
j
.2− 2 ku0 k 2
1 .
Ḃ2,∞

As a result, it comes out

(3.19) kuk ∞ 1 + kuk 3 ≤ C1 ku0 k 1 ∀ t ∈ [0, T1 ].
2 )
Lt (Ḃ2,∞ e 2 (Ḃ 2 )
L 2
Ḃ2,∞
t 2,∞
 1
1 2
Then if we take c in (1.6) to be so small that ku0 k 1
2
≤c≤ 2C1 min{ Cc00 , C11 }, we deduce
Ḃ2,∞
from (3.19) that
1  c 1 1
0 2
(3.20) kuk 1 ≤ C1 c ≤ min{ , },
L∞ 2
T (Ḃ2,∞ )
1
2 C0 C1
which contradicts with (3.14). This in turn shows that T1 = T ∗ , and there hold (3.2), (3.4).
Furthermore, it follows from (3.12) and (3.20) that
√
(3.21) k∇2 uj (t)k2L2 + k∇Πj (t)k2L2 ≤ Ck ρ ∂t uj (t)k2L2 ∀ t ∈ [0, T ∗ ).
For α ∈ (0, 21 ), thanks to (3.2), we have
Z t Z t0 Z t
−2α 2 −2α 2
τ k∇uj kL2 dτ ≤ τ k∇uj kL2 dτ + τ −2α k∇uj k2L2 dτ
0 0 t0
. ˙ j u0 k2 2
t0 1−2α 22j k∆ + t0 −2α ˙ j u0 k2 2 ,
k∆
L L

then for t0 ≈ 2−2j , we reach (3.3).

1
2
If in addition, u0 ∈ Ḃ2,r for 1 ≤ r ≤ +∞, then by inserting (3.8) and (3.16) into (3.18),
we obtain
X X
k∆˙ j ukL∞ (L2 ) + k∇∆
˙ j uk 2 2 . ˙ j u0 kL2 + 2−j
k∆ k∇∆˙ j u0 kL2
t Lt (L )
j ′ ≥j j ′ ≤j
X j′ X j′ j
. cj ′ ,r 2− 2 ku0 k 1
2
+ 2−j cj ′ ,r 2 2 ku0 k 1
2
. cj,r 2− 2 ku0 k 1
2
,
Ḃ2,r Ḃ2,r Ḃ2,r
j ′ ≥j j ′ ≤j

which leads to (3.5).

Furthermore, taking ε ∈ (0, 14 ), we get from (3.3) that
1 1 1
kτ −( 2 −ε) ∇uj kL2t (L2 ) . 2j( 2 −2ε) cj,2 ku0 k 1 , kτ −ε ∇uj kL2t (L2 ) . 2j(2ε− 2 ) cj,2 ku0 k 1 ,
Ḣ 2 Ḣ 2
which implies
1 X X 1
kτ − 4 ∇uk2L2 (L2 ) ≤ 2 kτ −ε ∇uk kL2t (L2 ) kτ −( 2 −ε) ∇uj kL2t (L2 )
t
k∈Z j≤k
X X
k(2ε− 21 ) 1
. ku0 k2 1 2 ck,2 2j( 2 −2ε) cj,2
Ḣ 2
k∈Z j≤k
X
. ku0 k2 1 c2k,2 . ku0 k2 1 .
Ḣ 2 Ḣ 2
k∈Z
12 H. ABIDI, G. GUI, AND P. ZHANG

This completes the proof of Proposition 3.1. 

Remark 3.1. From the proof of Proposition 3.1, we find that all the constants in the proof
of Proposition 3.1 depend only on the upper and lower bounds of the initial density ρ0 , which
ensures that the small positive constant c in (1.6) depends only on c0 and C0 , bounds of the
initial density ρ0 .
1
Proposition 3.2. Let u0 ∈ Ḣ 2 (R3 ) with div u0 = 0. Then under the assumption in Propo-
sition 3.1, for any j ∈ Z, there hold
1 1
(3.22) kt 2 ∇ uj kL∞ 2 ˙ j u0 kL2 ,
2 + kt 2 (∂t uj , ∇ uj ∇ Πj )kL2 (L2 ) . k∆
T (L ) T
1 1
2 j ˙
(3.23) kt (∂t uj , ∇ uj , ∇Πj )kL∞
2 2 + kt ∇∂t uj kL2 (L2 ) . 2 k∆
T (L )
j u0 kL2 ,
2
T
1 1 1
2 j ˙
(3.24) kt Dt uj kL∞
2 2 + kt ∇Dt uj kL2 (L2 ) + kt (∇ uj , ∇Πj )kL2 (L6 ) . 2 k∆
T (L )
2
j u0 kL2 ,
2
T T
1 1
2 2
(3.25) kukL2 (L∞ ) + kt (∇ u, ∇Π, ∂t u)kL2 (L2 ) + kt ∇ uk
4 2 1 . ku0 k 1 .
T T L2T (Ḣ 2 ) Ḣ 2

def
Henceforth we always denote Dt u = ∂t + u · ∇ to be the material derivative.
1
2
If in addition u0 ∈ Ḃ2,r with 1 ≤ r ≤ +∞, there holds
1 1
(3.26) kt 2 uk 3 + kt 2 ∂t uk 1 . ku0 k 1 .
e ∞ (Ḃ 2 )
L e 2 (Ḃ 2 )
L 2
Ḃ2,r
T 2,r T 2,r

Proof. It is easy to observe from (3.15) that

d 1 1 √
kt 2 ∇ uj k2L2 + c0 kt 2 ( ρ ∂t uj , ∇2 uj , ∇Πj )k2L2 ≤ k∇ uj k2L2 .
dt
By integrating the above inequality over [0, T ] and using (3.8), we find
1 1
(3.27) kt 2 ∇ uj kL∞ 2 ˙ j u0 kL2 ,
2 + kt 2 (∂t uj , ∇ uj , ∇ Πj )kL2 (L2 ) . k∇ uj kL2 (L2 ) . k∆
T (L ) T T

and (3.22) follows.

Whereas by applying ∂t to the uj equation of (3.1) and then taking the L2 inner product
of the resulting equation with ∂t uj , we find
Z
1 d √

k ρ∂t uj k2L2 + k∇∂t uj k2L2 = − ρt ∂t uj · ∂t uj + u · ∇uj dx
2 dt 3
(3.28) ZR
def
− ρ∂t uj · (ut · ∇uj ) dx = Ij + IIj + IIIj .
R3
For Ij , we get, by using the transport equation of (1.1), that
Z Z Z
2 2
Ij = − ρt |∂t uj | dx = ∇ · (ρ u)|∂t uj | dx = −2 ρ (u · ∇)∂t uj · ∂t uj dx,
R3 R3 R3
from which and (2.1), we infer

(3.29) |Ij | . kρkL∞ k∇∂t uj kL2 ku ∂t uj kL2 . kuk 2

1 k∇∂t uj k2L2 .
Ḃ2,∞

Similarly for IIj , there holds

Z Z
 
IIj = − ρt ∂t uj · (u · ∇uj ) dx = ∇ · (ρ u) ∂t uj · (u · ∇uj ) dx
3
R3
ZR Z
   
=− ρ u · ∇∂t uj · (u · ∇uj ) dx − ρ u · ∂t uj · (∇ u · ∇uj ) dx
R3 R3
Z 3
X
  def (k)
− ρ u · ∂t uj · (u · ∇2 uj ) dx = IIj .
R3 k=1
 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 13

It follows from Lemmas 2.1 and 2.2 that

(1)
|IIj | . kρkL∞ k∇∂t uj kL2 ku ⊗ u ⊗ ∇uj kL2
. k∇∂t uj kL2 ku ⊗ uk 1
2
k∇2 uj kL2 . kuk 1
2
kuk 2
3 k∇∂t uj kL2 k∇2 uj kL2 ,
Ḃ2,∞ Ḃ2,∞ Ḃ2,∞
(2)
|IIj | . kρkL∞ ku ∂t uj kL2 k∇ u ∇uj kL2 . kuk 2
1 kuk 2
3 k∇∂t uj kL2 k∇2 uj kL2 ,
Ḃ2,∞ Ḃ2,∞
(3)
|IIj | . kρkL∞ ku ⊗ uk 1
2
k∇∂t uj kL2 k∇2 uj kL2 . kuk 2
1 kuk 3
2
k∇∂t uj kL2 k∇2 uj kL2 ,
Ḃ2,∞ Ḃ2,∞ Ḃ2,∞

so that there holds

(3.30) |IIj | . kuk 1

2
kuk 2
3 k∇∂t uj kL2 k∇2 uj kL2 .
Ḃ2,∞ Ḃ2,∞

For IIIj , one has

1 1
(3.31) |IIIj | . kρkL∞ k∂t uj kL6 kut kL2 k∇uj kL3 . k∇∂t uj kL2 kut kL2 k∇uj kL2 2 k∇2 uj kL2 2 .

By inserting the estimates (3.29), (3.30) and (3.31) into (3.28), we obtain
1 d √
k ρ ∂t uj k2L2 + k∇∂t uj k2L2 ≤ Ckuk 12 k∇∂t uj k2L2
2 dt Ḃ2,∞
 1 1 
+ Ck∇∂t uj kL2 kuk 12 kuk 32 k∇2 uj kL2 + kut kL2 k∇uj kL2 2 k∇2 uj kL2 2 .
Ḃ2,∞ Ḃ2,∞

Thanks to (1.6) and (3.4), we get, by applying Young’s inequality and using (3.21), that
d √ √
k ρ ∂t uj k2L2 + k∇∂t uj k2L2 .kuk2 1 kuk2 3 k ρ ∂t uj k2L2
(3.32) dt 2
Ḃ2,∞ 2
Ḃ2,∞
2 √
+ kut kL2 k∇uj kL2 k ρ ∂t uj kL2 .
By multiplying the above inequality by t, we find
d 1√ 1 √
kt 2 ρ ∂t uj k2L2 + kt 2 ∇∂t uj k2L2 ≤ k ρ ∂t uj k2L2
dt
1√ 1 1√
+ Ckuk2 1 kuk2 3 kt 2 ρ ∂t uj k2L2 + Ckt 4 ut k2L2 k∇uj kL2 kt 2 ρ ∂t uj kL2 ,
2
Ḃ2,∞ 2
Ḃ2,∞

from which, we infer

d 1√ 1 1
1√
kt 2 ρ ∂t uj k2L2 + kt 2 ∇∂t uj k2L2 ≤ C kuk2 1 kuk2 3 + kt 4 ut k2L2 kt 2 ρ ∂t uj k2L2
(3.33) dt Ḃ22 2
Ḃ2,∞
1

+ C k∂t uj k2L2 + kt 4 ut k2L2 k∇uj k2L2 .
Applying Gronwall’s inequality gives rise to
1 1
kt 2 ∂t uj k2L∞ (L2 ) + kt 2 ∇∂t uj k2L2 (L2 )
t t
1

(3.34) ≤C k∂t uj k2L2 (L2 ) + kt 4 ut k2L2 (L2 ) k∇uj k2L∞ (L2 )
t t t
 1

× exp C kuk2 1 kuk2 3 + kt 4 ut k2L2 (L2 ) .
L∞ 2 L2t (Ḃ2,∞
2 ) t
t (Ḃ2,∞ )

It follows from (3.4) that

(3.35) kuk2 1 kuk2 3 . ku0 k4 1 . ku0 k2 1 .

L∞ 2 L2T (Ḃ2,∞
2 ) 2 Ḣ 2
T (Ḃ2,∞ ) Ḃ2,∞
14 H. ABIDI, G. GUI, AND P. ZHANG

1
While for kt 4 ut kL2 (L2 ) , we first computer
T
Z T Z X X
1 1
2
kt ut kL2 (L2 ) =
4 t2 ∂t uj ∂t uk dxdt
T
0 R3 j∈Z k∈Z
XXZ T Z
1 XX 1
≤2 t 2 ∂t uj ∂t uk dx dt . kt 2 ∂t uj kL2 (L2 ) k∂t uk kL2 (L2 ) ,
T T
j∈Z k≤j 0 R3 j∈Z k≤j

which together with (3.16) and (3.27) ensures that

1 XX
kt 4 ut k2 2 2 . k∆ ˙ k u0 kL2
˙ j u0 kL2 k∇∆
LT (L )
j∈Z k≤j
(3.36) X j X k X
. ku0 k2 1 cj 2− 2 ck 2 2 . ku0 k2 1 c2j . ku0 k2 1 .
Ḣ 2 Ḣ 2 Ḣ 2
j∈Z k≤j j∈Z

As a consequence, we get from the momentum equation in (1.1) and (2.1) that
1 1 1
kt 4 (∇2 u, ∇Π)kL2 (L2 ) . kt 4 ut kL2 (L2 ) + kt 4 u · ∇ukL2 (L2 )
T T T
1 1
. kt ut kL2 (L2 ) + kuk
4 1
2
kt 4 ∇2 ukL2 (L2 ) ,
T L∞ T
T (Ḃ2,∞ )

which along with (3.4) ensures

1 1
(3.37) kt 4 (∇2 u, ∇Π, ut )kL2 (L2 ) . kt 4 ut kL2 (L2 ) . ku0 k 1 .
T T Ḣ 2
By substituting the above estimates and (3.16) into (3.34), we obtain for any t ≤ T
1 1
kt 2 ∂t uj k2L∞ (L2 ) + kt 2 ∇∂t uj k2L2 (L2 ) ≤ C(ku0 k )k∇∆˙ j u0 k2 2 with 1
t t L Ḣ 2

 
def
C(ku0 k 21 ) = C 1 + ku0 k2 1 exp Cku0 k2 1 .
Ḣ Ḣ 2 Ḣ 2
So that the inequality (3.23) holds according to (3.21).
Thanks to (3.27) and (3.23), we deduce that
1 1
kt 2 Dt uj kL∞ 2 + kt 2 ∇Dt uj kL2 (L2 )
T (L ) T
1 1 1 1
. kt ∂t uj kL∞
2 2 + kt u · ∇uj kL∞ (L2 ) + kt 2 ∇∂t uj kL2 (L2 ) + kt 2 ∇(u · ∇)uj kL2 (L2 )
T (L ) T
2
T T

˙ j u0 kL2 + kukL∞ (L3,∞ ) kt 12 ∇uj kL∞ (L6,2 ) + kukL∞ (L3 ) kt 12 ∇2 uj k 2 6

. 2 k∆ j
T T T L (L ) T
1
+ k∇ ukL2 (L3,∞ ) kt ∇uj kL∞
T (L
6,2 ) 2
T
1 1
˙ j u0 kL2 + kuk
. 2 k∆ j
1 kt 2 ∇2 uj kL∞ 2 + kuk 1 kt 2 ∇2 uj kL2 (L6 )
T (L ) L∞
T (Ḣ )
L∞ 2 2 T
t (Ḃ2,∞ )
1
+ k∇ uk 1 kt ∇2 uj kL∞
2 2 ,
T (L )
L2T (Ḃ2,∞
2 )

where we used Proposition A.3 in the last step. We thus obtain

1 1
kt 2 Dt uj kL∞ 2 + kt 2 ∇Dt uj kL2 (L2 )
T (L ) T
(3.38) 1
˙ j u0 kL2 + Cku0 k
≤ C(ku0 k 21 ) 2 k∆ j
1 kt 2 ∇2 uj kL2 (L6 ) .
Ḣ Ḣ 2 T

Whereas we deduce from (3.11) and the classical theory on Stokes operator that
1 1
kt 2 (∇2 uj , ∇Πj )kL2 (L6 ) . kt 2 ρ (∂t uj + u · ∇uj )kL2 (L6 )
T T
1 1
(3.39) . kt ∂t uj kL2 (L6 ) + kukL2 (L∞ ) kt ∇ uj kL∞
2 6
T (L )
2
T T
1 1
. kt ∂t ∇uj kL2 (L2 ) + kukL2 (L∞ ) kt 2 ∇2 uj kL∞
2 2 .
T (L )
T T
 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 15

1 1
Notice that kuk2L∞ . k∇ukL2 k∇2 ukL2 = kt− 4 ∇ukL2 kt 4 ∇2 ukL2 , we deduce from (3.6) and
(3.37) that
Z T
1 1
2
kukL2 (L∞ ) . kt− 4 ∇ukL2 kt 4 ∇2 ukL2 dt
(3.40) T
0
1 1
. kt− 4 ∇ukL2 (L2 ) kt 4 ∇2 ukL2 (L2 ) . ku0 k 21 .
T T Ḣ

By substituting (3.23) and (3.40) into (3.39), we find

1 1 1
kt 2 (∇2 uj , ∇Πj )kL2 (L6 ) . kt 2 ∂t ∇uj kL2 (L2 ) + ku0 k 1 kt 2 ∇2 uj kL∞ 2
T (L )
(3.41) T T Ḣ 2

˙ j u0 kL2 .
. 2j k∆

By inserting (3.41) into (3.38) we obtain (3.24).

To prove (3.25), we first get, by using (3.21), that
1 X 1√ X 1
˙ j ∇2 uk 2 2 .
kt 2 ∆ kt 2 ρ∂ t uj ′k 2 2 + 2−j
kt 2 ∇ut kL2 (L2 )
L (L ) T L (L ) T T
j ′ ≥j j ′ ≤j

from which, (3.22) and (3.23), we infer

1 X X ′ j
˙ j uk 2 2 .
kt 2 ∆ k∆˙ j ′ u0 kL2 + 2−j ˙ j ′ u0 kL2 . cj 2− 2 ku0 k
2j k∆ 1 ,
L (L )
T Ḣ 2
j ′ ≥j j ′ ≤j

which implies that

1
(3.42) kt 2 ∇2 uk 1 . ku0 k 1 .
L2T (Ḣ 2 ) Ḣ 2

Combining (3.37) with (3.42) and (3.40) ensures (3.25).

1
2
If we assume in addition that u0 ∈ Ḃ2,r , we deduce from (3.27) that
1 1 j
2 −
(3.43) kt 2 ∇ uj kL∞ 2 + kt 2 (∂t uj , ∇ uj , ∇ Πj )kL2 (L2 ) . cj,r 2 2 ku0 k
T (L ) 2
1 ,
T Ḃ2,r

Along the same line to the derivation of (3.19), we get, by using (3.23), that
1
˙ j ukL∞ (L2 ) + kt 12 ∆
kt 2 ∇∆ ˙ j ∂t uk 2 2
T LT (L )
X 1
= kt 2 ∇∆˙ j uj ′ kL∞ (L2 ) + kt 12 ∆
˙ j ∂t uj ′ k 2 2 )
T L (L ) T
j ′ ≥j
X 1

+ ˙ j uj ′ kL∞ (L2 ) + kt 21 ∆
kt 2 ∇∆ ˙ j ∂t uj ′ k 2 2
T L (L ) T
j ′ ≤j−1
X X
. ˙ j u0 kL2 + 2−j
k∆ ˙ j u0 kL2 ,
k∇∆
j ′ ≥j j ′ ≤j−1

from which, we infer

1
˙ j ukL∞ (L2 ) + kt 21 ∆
kt 2 ∇∆ ˙ j ∂t uk 2 2
T LT (L )
X j′ X j′ j
. cj ′ ,r 2− 2 ku0 k 12 + 2−j cj ′ ,r 2 2 ku0 k 1
2
. cj,r 2− 2 ku0 k 2
1 .
Ḃ2,r Ḃ2,r Ḃ2,r
j ′ ≥j j ′ ≤j−1

This leads to (3.26). This completes the proof of Proposition 3.2. 

16 H. ABIDI, G. GUI, AND P. ZHANG

Proposition 3.3. Under the assumption in Proposition 3.2, one has

1
(3.44) k∇ ukL4 (L2 ) + kt 4 ∇ ukL∞ 2 . ku0 k
T (L ) 2
1 ,
T Ḃ2,∞
3 3 3
(3.45) kt 4 (∂t u, ∇2 u)kL∞ 2
2 + kt 4 ∇∂t ukL2 (L2 ) + kt 4 (∇ u, ∇Π)kL2 (L6 ) . ku0 k
T (L )
1 ,
T T Ḣ 2
3 3
(3.46) kt 4 Dt ukL∞ 2 + kt 4 ∇Dt ukL2 (L2 ) . ku0 k
T (L )
1 ,
T Ḣ 2
1
(3.47) kt ∇ ukL2 (L∞ ) . ku0 k
2 1 .
T Ḣ 2

Proof. Thanks to (3.4) and (3.26), we get

k∇ uk2L4 (L2 ) . kuk 1 kuk 3 . ku0 k2 1 ,
L∞ 2 e 2 (Ḃ 2 ) 2
T (Ḃ2,∞ ) L
T Ḃ2,∞
T 2,∞
1 1
kt 4 uk2L∞ (Ḣ 1 ) . kuk 1
2
kt 2 uk 3
2
. ku0 k2 1 ,
T L∞
T (Ḃ2,∞ ) L∞
T (Ḃ2,∞ )
2
Ḃ2,∞

which follows (3.44).

Whereas by applying ∂t to the momentum equation of (1.1) and then taking L2 inner
product of the resulting equation with ∂t u, we find
1d √
k ρ∂t uk2L2 + k∇∂t uk2L2
2 dt Z Z
(3.48)
def
=− ρt ∂t u · (∂t u + u · ∇u) dx − ρ∂t u · (ut · ∇u) dx = I + II + III.
3 3
R R

It follows from a similar derivation of (3.29)-(3.31) that

|I| + |II| . kuk 1
2
k∇∂t uk2L2 + kuk 2
1 kuk 3
2
k∇∂t ukL2 k∇2 ukL2 ,
Ḃ2,∞ Ḃ2,∞ Ḃ2,∞

and
√ √ √
|III| . k ρkL∞ k ρ∂t ukL2 k∂t ukL6 k∇ukL3 . k ρ∂t ukL2 k∇∂t ukL2 k∇uk 1 .
Ḣ 2

By substituting the above inequalities into (3.48) and using the smallness condition (1.6), we
find
d √ √
(3.49) k ρ ∂t uk2L2 + k∇∂t uk2L2 .kuk2 1 kuk2 3 k∇2 uk2L2 + k ρ∂t uk2L2 k∇uk2 1 .
dt 2
Ḃ2,∞ 2
Ḃ2,∞ Ḣ 2

While it follows from (1.1) and (2.1) that

√ √

k(∇2 u, ∇Π)k2L2 ≤ C(k ρ ∂t uk2L2 + ku · ∇uk2L2 ) ≤ C k ρ ∂t uk2L2 + kuk2 1 k∇2 uk2L2 .
2
Ḃ2,∞

Then under the assumption of (1.6), we deduce from (3.4) that for any t ≤ T,
√
(3.50) k(∇2 u(t), ∇Π(t))k2L2 . k ρ ∂t u(t)k2L2 .

By inserting (3.50) into (3.49), we find

d √
√
k ρ ∂t uk2L2 + k∇∂t uk2L2 ≤C kuk2 1 kuk2 3 + k∇uk2 1 k ρ ∂t uk2L2 ,
dt 2
Ḃ2,∞ 2
Ḃ2,∞ Ḣ 2

from which, we infer

d 3√ 3
kt 4 ρ ∂t uk2L2 + kt 4 ∇∂t uk2L2
dt
3 1
3√
≤ kt 4 ∂t uk2L2 + C kuk2 1 kuk2 3 + k∇uk2 1 kt 4 ρ ∂t uk2L2 .
2 2
Ḃ2,∞ 2
Ḃ2,∞ Ḣ 2
 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 17

Applying Gronwall’s inequality gives rise to

3 3
kt 4 ∂t uk2L∞ (L2 ) + kt 4 ∇∂t uk2L2 (L2 )
t t
3 1 

2
≤ kt 4 ut kL2 (L2 ) exp C kuk2 1 kuk2
3 + kuk2
3 ,
2 t L∞ 2
t (Ḃ2,∞ ) L2T (Ḃ2,∞
2 ) L2t (Ḣ 2 )

which together with (3.6), (3.35), (3.36) and (3.50) ensures that
3 3
kt 4 (∂t u, ∇2 u)k2L∞ (L2 ) + kt 4 ∇∂t uk2L2 (L2 ) ≤ C(ku0 k 21 ) ku0 k2 1 with
T T Ḣ Ḣ 2
(3.51)
 
def
C(ku0 k 12 ) = C 1 + ku0 k2 1 exp Cku0 k2 1 .
Ḣ Ḣ 2 Ḣ 2
Once again we deduce from (1.1) and classical theory on Stokes operator that
3 3
kt 4 (∇2 u, ∇Π)kL2 (L6 ) . kt 4 ρ (∂t u + u · ∇u)kL2 (L6 )
T T
3 1 1
. kt ∂t ukL2 (L6 ) + kt ukL∞
4 6 kt 2 ∇ ukL2 (L∞ )
T (L )
4
T T
3 1 1 1 3 1
. kt ∇∂t ukL2 (L2 ) + kt ∇ukL∞
4
T (L )
2
2 kt ∇ uk 2
L (L6 )
kt 4 ∇2 ukL2 2 (L6 ) ,
4 4
T T T

from which, we infer

3 3 1 1
2 2
(3.52) kt 4 (∇ u, ∇Π)kL2T (L6 ) . kt 4 ∇∂t ukL2T (L2 ) + kukL∞ (Ḣ 12 ) kt 2 ukL∞ (Ḣ 23 ) kt 4 ∇ ukL2T (L2 ) .
T T

Thanks to (3.6), (3.25), (3.26) and (3.51), we conclude that

3
(3.53) kt 4 (∇2 u, ∇Π)kL2 (L6 ) . ku0 k 1 ,
T Ḣ 2
which together with (3.51) ensures (3.45).
To prove (3.46), we get, by using Proposition A.3, that
3 3
kt 4 Dt ukL∞ 2 + kt 4 ∇Dt ukL2 (L2 )
T (L ) T
3 3 3 3
. kt ∂t ukL∞
4 2 + kt u · ∇ukL∞ (L2 ) + kt 4 ∇∂t ukL2 (L2 ) + kt 4 ∇(u · ∇)ukL2 (L2 )
T (L ) T
4
T T
3 3
2
. C(ku0 k 1 ) ku0 k 1 + kukL∞
T (L
3,∞ ) kt ∇ukL∞ (L6,2 ) + kukL∞ (L3 ) kt ∇ ukL2 (L6 )
4 4
Ḣ 2 Ḣ 2 T T T
3
2
+ k∇ uk 1 kt ∇ ukL∞
4 2
T (L )
L2T (Ḣ 2 )
3 3
. C(ku0 k 1 ) ku0 k 1 + kuk 1 kt 4 ∇2 ukL∞ 2 + kuk
T (L )
1 kt 4 ∇2 ukL2 (L6 )
Ḣ 2 Ḣ 2 L∞ 2 L∞
T (Ḣ )
2 T
T (Ḃ2,∞ )
3
+ k∇ uk 1 kt 4 ∇2 ukL∞ 2 ,
T (L )
L2T (Ḣ 2 )

which along with (3.45) implies (3.46).

On the other hand, we get, by using (3.25), (3.53), and interpolating the inequality:
1 1
kf kL∞ . k∇f kL2 2 k∇f kL2 6 , that
Z T Z T
1 1 3
kt 2 ∇ uk2L2 (L∞ ) . t k∇ u(t)k2L∞ dt . kt 4 ∇2 u(t)kL2 kt 4 ∇2 u(t)kL6 dt
T
0 0
1 3
. kt ∇ ukL2 (L2 ) kt ∇ ukL2 (L6 ) . ku0 k2 1 ,
4
2 4
2
T T Ḣ 2
which leads to (3.47) This completes the proof of Proposition 3.3. 
Now we are in a position to complete the proof of Theorem 1.3.
Proof of Theorem 1.3. By mollifying the initial data (ρ0 , u0 ) to be (ρ0ǫ , u0ǫ ) , we deduce
from the classical theory of inhomogeneous incompressible Navier-Stokes system (see [2] for
instance) that (1.1) has a unique local solution (ρǫ , uǫ ) on [0, Tǫ∗ ). If c is sufficiently small
18 H. ABIDI, G. GUI, AND P. ZHANG

in (1.6), we can show that (3.4), (3.6), (3.47) hold for (ρǫ , uǫ ). Then a continuous argument
1 1
shows that Tǫ∗ = +∞. Due to u0 ∈ Ḃ2,∞
2
∩ Ḣ 2 , we deduce from (3.5) that for any r ∈ [2, ∞)

(3.54) kuǫ k 1 + kuǫ k 3 ≤ Cku0 k 1 .

e ∞ (R+ ;Ḃ 2 )
L e 2 (R+ ;Ḃ 2 )
L 2
Ḃ2,r
2,r 2,r

The derivation of (3.36) ensures that

1
(3.55) kt 4 ∂t uǫ kL2 (R+ ;L2 ) ≤ Cku0 k 1 .
Ḣ 2

And it follows from the proof of (3.26) that

1 1
(3.56) kt 2 uǫ k 3 + kt 2 ∂t uǫ k 1 . ku0 k 1 ∀ 2 ≤ r ≤ +∞.
e ∞ (R+ ;Ḃ 2 )
L e 2 (R+ ;Ḃ 2 )
L 2
Ḃ2,r
2,r 2,r

Furthermore, we deduce from the proof of Proposition 3.3 that

1
k∇ uǫ kL4 (R+ ;L2 ) + kt 4 ∇ uǫ kL∞ (R+ ;L2 ) ≤ Cku0 k 2
1 ,
Ḃ2,∞
3 3
(3.57) kt 4 (∂t uǫ , ∇2 uǫ , Dtǫ uǫ )kL∞ (R+ ;L2 ) + kt 4 (∇∂t uǫ , ∇Dtǫ uǫ )kL2 (R+ ;L2 )
3 1
+ kt 4 (∇2 uǫ , ∇Πǫ )kL2 (R+ ;L6 ) + kt 2 ∇ uǫ kL2 (R+ ;L∞ ) ≤ Cku0 k 1 ,
Ḣ 2

def
where Dtǫ = ∂t + uǫ · ∇.
Then exactly along the same line to the proof of Theorem 1.1 in [17] (one may check page
644-645 of [17]), we can complete the existence part of Theorem 1.3 by using a standard
compactness argument, which we omit details here. Moreover, we deduce (1.7) and (1.8)
from (3.54-3.57). The uniqueness part follows from Theorem 1.2. Finally let’s prove u ∈
1
C([0, ∞); Ḣ 2 ). Indeed it follows from (1.8) that
kuk e ∞ 1 ≤ Cku0 k 1 .
L (R+ ;Ḣ 2 ) Ḣ 2

Then for any ε > 0, there is a positive j0 = j0 (ε) ∈ N so that

X
4 ˙ j uk2 ∞ + 2 < ε.
2j k∆ L (R ;L )
|j|≥j0

So that for any t ∈ [0, +∞), h > 0, we have

X
ku(t + h) − u(t)k2 1 = ˙ j (u(t + h) − u(t))k2 2
2j k∆
Ḣ 2 L
j∈Z
X X
≤ j˙ j (u(t + h) − u(t))k2 2 + 4
2 k∆ ˙ j uk2 ∞ 2
2j k∆
L L (L )T
|j|≤j0−1 |j|≥j0

≤ 2j0 ku(t + h) − u(t)k2L2 + ε,

from which, we infer
Z t+h 1 1
ku(t + h) − u(t)k2 1 ≤ 2j0 k τ − 4 τ 4 ∂τ u(τ ) dτ k2L2 + ε
Ḣ 2
t
(3.58) ≤2 j0 − 14 2
kτ kL2 ([t,t+h]) kτ
1
4 ∂τ ukL2 ([t,t+h]) k2L2 + ε
1
≤ C 2j0 +2 ku0 k 1 h 2 + ε,
Ḣ 2

1
where we used the fact kt ∂t ukL2 (R+ ;L2 ) ≤ Cku0 k
4 1 of (1.8). This shows that u ∈ C([0, ∞);
Ḣ 2
1
Ḣ ), and we completes the proof of Theorem 1.3.
2 
 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 19

4. The proof of Theorem 1.4

The goal of this section is to present the proof of Theorem 1.4. With the a priori esti-
mates derived in the previous section, here we shall first derive the globally-in-time Lipschitz
1
2
estimate (4.15) below for the velocity field if we assume in addition that u0 ∈ Ḃ2,1 . Toward
this, we first derive the following estimate via energy method:
Proposition 4.1. Under the assumptions of Proposition 3.2, we have
(4.1) kt Dt uj kL∞ ˙ j u0 kL2 .
2 + kt ∇Dt uj kL2 (L2 ) . k∆
T (L ) T
1
2
If in addition u0 ∈ Ḃ2,1 , we have
1
(4.2) kt 2 (∂t u, ∇2 u, ∇Π)kL4,1 (L2 ) . ku0 k 2
1 ,
T Ḃ2,1

and
(4.3) kt (∇2 u, ∂t u, ∇Π)kL4,1 (L6 ) . ku0 k 2
1 ,
T Ḃ2,1

where the Lorentz norm of L4,1

T is defined by Definition A.3.

Proof. We first get, by multiplying (3.32) by t2 , that

d √ 1√
kt ρ ∂t uj k2L2 + kt ∇∂t uj k2L2 ≤ 2kt 2 ρ ∂t uj k2L2
dt  
√ 1 1
+ C kuk2 1 kuk2 3 kt ρ ∂t uj k2L2 + kt 4 ut k2L2 kt 2 ∇uj kL2 kt ∂t uj kL2 ,
2
Ḃ2,∞ 2
Ḃ2,∞

which together with (1.5) implies

d √ 1√
kt ρ ∂t uj k2L2 + kt ∇∂t uj k2L2 ≤ 2kt 2 ρ ∂t uj k2L2
dt  
1
√ 1 1
+ C kuk2 1 kuk2 3 + kt 4 ut k2L2 kt ρ ∂t uj k2L2 + kt 4 ut k2L2 kt 2 ∇uj k2L2 .
2
Ḃ2,∞ 2
Ḃ2,∞

Applying Gronwall’s inequality gives rise to

kt ∂t uj k2L∞ (L2 ) + kt ∇∂t uj k2L2 (L2 )
t t
1 1 1

≤ C kt 2 ∂t uj k2L2 (L2 ) + kt ut k2L2 (L2 ) kt 2 ∇uj k2L∞ (L2 )
4
t t T
 1

2 2
× exp C kuk 1 kuk 3 + kt 4 ut k2L2 (L2 ) ∀ t ∈ [0, T ].
L∞ 2 L2t (Ḃ2,∞
2 ) t
t (Ḃ2,∞ )

By inserting (3.4), (3.27) and (3.36) into the above inequality and using (3.21), we obtain
(4.4) kt (∂t uj , ∇2 uj )kL∞ 2 + kt ∇∂t uj kL2 (L2 ) ≤ C(ku0 k 1 ˙ j u0 kL2 .
) k∆
T (L ) T Ḣ 2
While we deduce from (3.11) and the classical theory on Stokes operator that
kt (∇2 uj , ∇Πj )kL6 . kt ∂t uj kL6 + kt u · ∇ uj kL6
. kt ∂t uj kL6 + t kukL6 k∇ uj kL∞
1 1
. kt ∂t uj kL6 + k∇ukL2 kt ∇ uj kL2 6 kt ∇2 uj kL2 6 .
Applying Young’s inequality yields
(4.5) kt (∇2 uj , ∇Πj )kL6 . kt ∂t uj kL6 + k∇uk2L2 kt ∇2 uj kL2 .
In particular, we have
kt (∇2 uj , ∇Πj )kL2 (L6 ) . kt ∇∂t uj kL2 (L2 ) + k∇uk2L4 (L2 ) kt ∇2 uj kL∞ 2 ,
T (L )
T T T
20 H. ABIDI, G. GUI, AND P. ZHANG

which along with (3.40) and (4.4) ensures that

(4.6) kt (∇2 uj , ∇Πj )kL2 (L6 ) ≤ C(ku0 k 1 ) 1 + ku0 k2 1
˙ j u0 kL2 .
k∆
T Ḣ 2 2
Ḃ2,∞

On the other hand, we get, by applying Propositions A.2 and A.3, that
kt Dt uj kL∞ 2 + kt ∇Dt uj kL2 (L2 )
T (L ) T

. kt∂t uj kL∞ 2 + kukL∞ (L3,∞ ) kt∇uj kL∞ (L6,2 ) + kt ∇∂t uj kL2 (L2 )
T (L ) T T T
1 1
2
+ kt ∇ ukL∞
2
T (L
3,∞ ) kt ∇uj kL2 (L6,2 ) + kukL∞ (L3 ) kt∇ uj kL2 (L6 )
2
T T T

. kt∂t uj kL∞ 2 + kuk

T (L ) 2
1 kt∇2 uj kL∞ 2 + kt ∇∂t uj kL2 (L2 )
T (L )
L∞ T
T (Ḃ2,∞ )
1 1
+ kt ∇ uk 2 1 kt ∇2 uj kL2 (L2 ) + kuk
2 1 kt∇2 uj kL2 (L6 ) ,
L∞ 2 T L∞
T (Ḣ )
2 T
T (Ḃ2,∞ )

which along with (3.4), (3.23), (4.4) and (4.6) ensures (4.1).
Whereas it follows from (3.22) and (3.23) that
1 j
kt 2 (∂t uj , ∇2 uj )kL2 (L2 ) . dj 2− 2 ku0 k 2
1 ,
T Ḃ2,1
1 j
kt 2 (∂t uj , ∇2 uj )kL∞ 2 . dj 2 2 ku0 k
T (L ) 2
1 ,
Ḃ2,1


from which and the interpolation theorem: L4,1 2 2 2 ∞ 2
T (L ) = LT (L ), LT (L ) 1 ,1 , we infer 2
1
2
kt (∂t uj , ∇ uj )kL4,1 (L2 ) . dj ku0 k
2
2
1 .
T Ḃ2,1

Then we deduce that

1 X 1
kt 2 (∂t u, ∇2 u)kL4,1 (L2 ) = k t 2 (∂t uj , ∇2 uj )kL4,1 (L2 )
T T
j∈Z
(4.7) X 1
. kt 2 (∂t uj , ∇2 uj )kL4,1 (L2 ) . ku0 k 2
1 .
T Ḃ2,1
j∈Z

In view of (2.1), we get, by applying Propositions A.2 and A.3, that

1 1
kt 2 ∇ΠkL4,1 (L2 ) = kt 2 (∆ u − ρ(∂t u + u · ∇ u))kL4,1 (L2 )
T T
1 1
2
. kt (∂t u, ∇ u)kL4,1 (L2 ) + kukL∞
2
T (L
3,∞ ) kt 2 ∇ uk 4,1
L (L6,2 )
T T
1 1
2 2
. kt (∂t u, ∇ u)kL4,1 (L2 ) + kuk
2
2
1 kt ∇ ukL4,1 (L2 ) ,
2
T L∞
T (Ḃ2,∞ )
T

which together with (3.4) and (4.7) implies

1
kt 2 ∇ΠkL4,1 (L2 ) . ku0 k 2
1 .
T Ḃ2,1

This together with (4.7) ensures (4.2).

Finally let us turn to the proof of (4.3). Indeed again thanks to (4.5), one has
1 1
kt (∇2 uj , ∇Πj )kL∞
T (L ) T
2
6 . kt∂t uj kL∞ (L6 ) + kt 4 ∇uk ∞ 2
2
L (L2 ) kt ∇ uj kL∞ 2 ,
T (L ) T

which together with (3.23) and (3.44) implies

1
kt (∇2 uj , ∇Πj )kL∞ 2
6 . kt ∂t uj kL∞ (L6 ) + kt 2 ∇ uj kL∞ (L2 )
T (L )
(4.8) T T
j
. kt ∂t uj kL∞ 6 + 2 k∆j u0 kL2 .
T (L )

To complete the proof of (4.3), we need the following lemma, the proof of which will be
postponed after we finish the proposition.
 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 21

1
2
Lemma 4.1. If u0 ∈ Ḃ2,1 , one has
j
(4.9) kt ∂t uj kL∞ 6 . 2 k∆j u0 kL2 .
T (L )

By combining (4.8) with (4.9), we obtain

kt (∇2 uj , ∂t uj , ∇Πj )kL∞ j
6 . 2 k∆j u0 kL2 ,
T (L )


from which, (4.6) and the interpolation theorem: L4,1 2 2 2 ∞ 2
T (L ) = LT (L ), LT (L ) 1 ,1 , we infer
2
j
(4.10) kt (∇2 uj , ∇Πj )kL4,1 (L6 ) . 2 k∆j u0 kL2 .
2
T

Due to the momentum equations in (3.1), we find

kt ∂t uj kL4,1 (L6 ) . kt (∇2 uj , ∇Πj )kL4,1 (L6 ) + kt u · ∇uj kL4,1 (L6 )
T T T
(4.11) 1 1
. kt (∇2 uj , ∇Πj )kL4,1 (L6 ) + kt 2 ukL∞ ∞ kt 2 ∇uj k 4,1
T (L ) L (L6 ) .
T T
1 2
Whereas we get, by using the interpolation inequality: kakL∞ . kak 3
1 k∇2 akL3 2 , (3.5) and
2
Ḃ2,∞
(3.45), that
1 1 3 2
kt 2 ukL∞ ∞ . kuk
T (L )
3
1 kt 4 ∇2 ukL3 ∞ (L2 ) . ku0 k 1 ,
L∞ 2 T Ḣ 2
T (Ḃ2,∞ )

from which, (4.10) and (4.11), we infer

1 j
kt ∂t uj kL4,1 (L6 ) . kt (∇2 uj , ∇Πj )kL4,1 (L6 ) + ku0 k 1
2
kt 2 ∇2 uj kL4,1 (L2 ) . 2 2 k∆j u0 kL2 .
T T Ḃ2,1 T

We thus deduce that

j
kt (∇2 uj , ∂t uj , ∇Πj )kL4,1 (L6 ) . 2 2 k∆j u0 kL2 . dj ku0 k 2
1 ,
T Ḃ2,1

from which, we arrive at

X
kt (∇2 u, ∂t u, ∇Π)kL4,1 (L6 ) . kt (∇2 uj , ∂t uj , ∇Πj )kL4,1 (L6 ) . ku0 k 2
1 ,
T T Ḃ2,1
j∈Z

and (4.3) follows. This completes the proof of Proposition 4.1. 

Let us complete the proof of Lemma 4.1.
Proof of Lemma 4.1. We get, by first applying ∂t to the uj equation of (3.1) and then taking
the L2 inner product of the resulting equation with |∂t uj |4 ∂t uj , that
1 d 1
kρ 6 ∂t uj k6L6 + k∇|∂t uj |3 k2L2
6 dt Z Z
(4.12) =− ρt |∂t uj |4 ∂t uj · (∂t uj + u · ∇uj ) dx − ρ|∂t uj |4 ∂t uj · (ut · ∇uj ) dx
3 3
R R
def
= I + II + III.
Next we handle term by term above.
By using the transport equation of (1.1) and integrating by parts, we have
Z Z Z
6 6
I=− ρt |∂t uj | dx = ∇ · (ρ u)|∂t uj | dx = −2 ρ |∂t uj |3 (u · ∇)|∂t uj |3 dx,
R3 R3 R3
1
from which and the embedding: Ḃ2,∞ (R3 ) ֒→ L3,∞ (R3 ) and Ḣ 1 (R3 ) ֒→ L6,2 (R3 ) (see Propo-
2

sition A.3), we infer

I . kρkL∞ kukL3,∞ k∇|∂t uj |3 kL2 k|∂t uj |3 kL6,2 . kuk 1
2
k∇|∂t uj |3 k2L2 .
Ḃ2,∞
22 H. ABIDI, G. GUI, AND P. ZHANG

Similarly for II, we have

Z Z
4
 
II = − ρt |∂t uj | ∂t uj · (u · ∇uj ) dx = ∇ · (ρ u) |∂t uj |4 ∂t uj · (u · ∇uj ) dx
3
R3
ZR Z
 4
  
=− ρ u · ∇(|∂t uj | ∂t uj ) · (u · ∇uj ) dx − ρ u · |∂t uj |4 ∂t uj · (∇ u · ∇uj ) dx
R3 R3
Z 3
X
  def
− ρ u · |∂t uj |4 ∂t uj · (u · ∇2 uj ) dx = IIk .
R3 k=1
By applying Hölder inequality and Proposition A.3, we find
|II1 | . kρkL∞ kuk2L∞ k∇|∂t uj |3 kL2 k|∂t uj |2 kL3 k∇uj kL6
. kuk2L∞ k∇|∂t uj |3 kL2 k∂t uj k2L6 k∇2 uj kL2 ,
|II2 | . kρkL∞ kukL∞ k∇ ukL3,∞ k|∂t uj |2 kL3 k|∂t uj |3 kL6,2 k∇uj kL6,2
. kukL∞ k∇ ukL3,∞ k∇|∂t uj |3 kL2 k∂t uj k2L6 k∇2 uj kL2 ,
|II3 | . kρkL∞ kuk2L∞ k|∂t uj |2 kL3 k|∂t uj |3 kL6,2 k∇2 uj kL2
. kuk2L∞ k∇|∂t uj |3 kL2 k∂t uj k2L6 k∇2 uj kL2 .
Hence, we obtain


|II| . kuk2L∞ + kukL∞ k∇ ukL3,∞ k∇|∂t uj |3 kL2 k∂t uj k2L6 k∇2 uj kL2 .
Finally it is easy to observe that
|III| . kρkL∞ kuk2L∞ k|∂t uj |3 kL6 k|∂t uj |2 kL3 k∇2 uj kL2
. kuk2L∞ k∇|∂t uj |3 kL2 k∂t uj k2L6 k∇2 uj kL2 .
By substituting the above estimates into (4.12), we arrive at
1d 1
kρ 6 ∂t uj k6L6 + k∇|∂t uj |3 k2L2
6 dt

≤ C kuk2L∞ + kukL∞ k∇ ukL3,∞ k∇|∂t uj |3 kL2 k∂t uj k2L6 k∇2 uj kL2 ,
applying Young’s inequality yields
d 1
kρ 6 ∂t uj k6L6 + k∇|∂t uj |3 k2L2 ≤ C(kuk4L∞ + kuk2L∞ k∇ uk2L3,∞ )k∂t uj k4L6 k∇2 uj k2L2 .
dt
Then we get, by multiplying t6 to the above inequality, that
d 1 1 1 1
kt ρ 6 ∂t uj k6L6 +kt3 ∇|∂t uj |3 k2L2 ≤ 6kt 2 ρ 6 ∂t uj k2L6 kt ρ 6 ∂t uj k4L6
dt
1 1 1
1
+ C kt 2 uk4L∞ + kt 2 uk2L∞ kt 2 ∇ uk2L3,∞ kt ρ 6 ∂t uj k4L6 k∇2 uj k2L2 ,
from which, we deduce that
1
kt ∂t uj k2L∞ (L6 ) ≤ Ckt 2 ∂t uj k2L2 (L6 )
T T
(4.13) 1 1 1
+ Ckt 2 uk2L∞ (L∞ ) (kt 2 uk2L∞ (L∞ ) + kt 2 ∇ uk2L∞ (L3,∞ ) )k∇2 uj k2L2 (L2 ) .
T T T T

Yet it follows from (3.23), (3.26) and (3.16) that

1 1
kt 2 ∂t uj kL2 (L6 ) . kt 2 ∇∂t uj kL2 (L2 ) . 2j k∆j u0 kL2 ,
T T
1 1 1
kt ukL∞
2 ∞ + kt ∇ ukL∞ (L3,∞ ) . kt 2 uk
T (L ) T
2
2
3 . ku0 k 2
1 ,
L∞
T (Ḃ2,1 ) Ḃ2,1

k∇2 uj kL2 (L2 ) ) . 2j k∆j u0 kL2 .

T

By inserting the above estimates into (4.13), we obtain (4.9). This completes the proof of
Lemma 4.1. 
 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 23

1
2
Corollary 4.1. Let u0 ∈ Ḃ2,1 . Then under the assumptions in Proposition 3.1, we have

(4.14) k(∆ u, ∇ Π, ∂t u)kL1 (L3 ) . ku0 k 2

1
T Ḃ2,1

and
(4.15) k∇ukL1 (L∞ ) . ku0 k 1
2
.
T Ḃ2,1

Proof. We first deduce from (4.2) and (4.3) that

3 1 1 1
kt 4 (∆ u, ∇ Π)kL4,1 (L3 ) . kt 2 (∆ u, ∇ Π)k 2 4,1 kt (∆ u, ∇ Π)k 2 4,1 . ku0 k 1 ,
T LT (L2 ) LT (L6 ) 2
Ḃ2,1

from which and Proposition A.2, we infer

3 3
k(∆ u, ∇ Π)kL1 (L3 ) . kt− 4 k 4 kt 4 (∆ u, ∇ Π)kL4,1 (L3 )
T L 3 ,∞ (R+ ) T
3
(4.16) . kt (∆ u, ∇ Π)kL4,1 (L3 )
4
T

. ku0 k 1
2
.
Ḃ2,1

To control k∇ukL1 (L∞ ) , we get, by using the interpolation inequality, that

T
Z T Z T 1 1
Z T 1 1
3 1
2 2
k∇ukL∞ dt . 2 2
k∇ ukL2 k∇ ukL6 dt . t− 4 kt 2 ∇2 ukL2 2 kt ∇2 ukL2 6 dt
0 0 0
1 1
(4.17) − 34 1
. kt k 4 kt ∇2 uk
2 2
kt ∇2 uk 2 4,1
L 3 ,∞ (R+ ) L4,1 2
T (L ) LT (L6 )
. ku0 k 2
1 ,
Ḃ2,1

where we used (4.2) and (4.3) in the last inequality.

Finally, we deduce from the momentum equations in (1.1), (3.5), (4.16), and (4.17), that
k∂t ukL1 (L3 ) . k(∆ u, ∇ Π)kL1 (L3 ) + ku · ∇ ukL1 (L3 )
T T T

. k(∆ u, ∇ Π)kL1 (L3 ) + kuk 2

1 k∇ ukL1 (L∞ )
T L∞ T
T (Ḃ2,1 )

. ku0 k 2
1 .
Ḃ2,1

This together with (4.16) and (4.17) finishes the proof of Corollary 4.1. 
Lemma 4.2. Under the assumptions of Corollary 4.1, one has


kak 3 ≤ ka0 k 3 exp Cku0 k 12 if 3 < λ < +∞,
e ∞ (Ḃ λ )
L λ
Ḃλ,1 Ḃ2,1
T λ,1
(4.18)

kakLe ∞ (Ḃ 0 ≤ ka0 kḂ 0 1 + Cku0 k 1 .
T ∞,1 ) ∞,1 2
Ḃ2,1

Proof. It follows from Lemma 2.100 of [4] that

kak 3 ≤ ka0 k 3 exp Ck∇ukL1 (L∞ ) if 3 < λ < +∞,
e ∞ (Ḃ λ )
L λ
Ḃλ,1 T
T λ,1


kakLe ∞ (Ḃ 0 ≤ ka0 kḂ 0 1 + Ck∇ukL1t (L∞ ) .
t ∞,1 ) ∞,1

By plugging (4.17) into the above inequalities, we obtain (4.18). 

Remark 4.1. By applying Lemmas 2.6 and 4.2 and Corollary 4.1, we find
X j −1+ 3 
2 p ˙ j P, ρ−1 ](∆ u − ∇ Π)k 1 p ≤ Ckak
k[∆ 3 k(∆ u − ∇ Π)kL1 (L3 )
Lt (L ) e ∞ (Ḃ λ ) t
L t λ,1
(4.19) j∈Z


≤ Cka0 k 3 ku0 k 21 exp Cku0 k 12 ,
λ
Ḃλ,1 Ḃ2,1 Ḃ2,1
24 H. ABIDI, G. GUI, AND P. ZHANG

1 1
when 2 ≤ p < 3 and 3 < λ < ∞ such that p ≤ 3 + λ1 .
Proposition 4.2. Under the assumptions of Corollary 4.1, we have


(4.20) kuk 1 + kuk 5 . ku 0 k 1 1 + ka0 k 12 exp Cku0 k 1 .
e∞ 2 1
LT (Ḃ2,1 ) 2 2
LT (Ḃ2,1 ) Ḃ2,1 Ḃ6,1 2
Ḃ2,1

def
Proof. We first get, by applying Leray projection operator P = Id + ∇(−∆)−1 div to the
equation ∂t u + u · ∇u + ρ−1 (∇Π − ∆u) = 0, that
∂t u + P(u · ∇u) + P[ρ−1 (∇Π − ∆u)] = 0.
Applying dyadic operator to the above equation yields


˙ ju + ∆
∂t ∆ ˙ j P(u · ∇u) + ∆
˙ j P ρ−1 (∇Π − ∆u) = 0.
By multiplying the above equation by ρ and using a standard commutator’s argument, we
write
˙ j u + ρu · ∇∆
ρ∂t ∆ ˙ j u − ∆∆
˙ j u + ρ[∆
˙ j P, u] · ∇u + ρ[∆
˙ j P, ρ−1 ](∇Π − ∆u) = 0.
˙ j u, that
Then we get, by taking L2 inner product of the above equation with ∆
d ˙ ˙ j uk2 2
k∆j uk2L2 +c1 22j k∆ L
dt  
. k[∆ ˙ j P, ρ−1 ](∇Π − ∆u)kL2 k∆
˙ j P, u] · ∇ukL2 + k[∆ ˙ j ukL2 ,

from which, we infer

k∆˙ j ukL∞ (L2 ) + 22j k∆
˙ j uk
t L1t (L2 )

. k∆ ˙ j P, u] · ∇uk 1 2 + k[∆
˙ j u0 kL2 + k[∆ ˙ j P, ρ−1 ](∇Π − ∆u)k 1 2 .
Lt (L ) Lt (L )

So that we have
X j
kuk 1 + kuk 5 .ku0 k 1 + ˙ j P, u] · ∇uk 1 2
2 2 k[∆
e ∞ (Ḃ 2 ) e 1 (Ḃ 2 ) 2 Lt (L )
L t 2,1 L t 2,1 Ḃ2,1
j∈Z
X j
+ ˙ j P, ρ−1 ](∇Π − ∆u)k 1 2 ,
2 k[∆
2
Lt (L )
j∈Z

which together with (2.8) and (4.19) ensures that

kuk 1 + kuk 5 .ku0 k 1 + kukL1 (Ḃ 1 kuk 1
e ∞ (Ḃ 2 ) L1t (Ḃ2,1
2 ) 2 t ∞,1 ) L∞ 2
L t 2,1 Ḃ2,1 t (Ḃ2,∞ )


+ ka0 k 1
2
exp Cku0 k 1
2
ku0 k 2
1 .
Ḃ6,1 Ḃ2,1 Ḃ2,1

Due to (1.6), ku0 k 2

1 is sufficiently small, we deduce (4.20) from the above inequality. This
Ḃ2,∞
completes the proof of the proposition. 
Let us turn to the proof of Theorem 1.4.
Proof of Theorem 1.4. Under the assumptions of Theorem 1.4, we deduce from the classical
theory of inhomogeneous Navier-Stokes equations (see [2, 13] for instance), Lemma 4.2 and
Proposition 4.2 that the system (1.1) has a unique local solution (ρ, u) on [0, T ⋆ ) so that
there holds (1.5) and
1 1 5

(ρ−1 − 1, u) ∈ C([0, T ⋆ ); Ḃ6,1
2
) × C([0, T ⋆ ); Ḃ2,1
2
) ∩ L1 ([0, T ⋆ ); Ḃ2,1
2
) ,
where T ⋆ denotes the lifespan of the solution (ρ, u).
Furthermore due to (4.15), we get, by a continuous argument, that T ⋆ = ∞. The estimate
(1.11) follows from (4.18) for λ = 6 and (4.20). This completes the proof of Theorem 1.4. 
 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 25

Appendix A. Tools box on Littlewood-Paley analysis and Lorentz spaces

In this section, we recall some basic facts on Littlewood-Paley theory from [4] and Lorentz
spaces from [15]. Let us briefly explain how it may be built in the case x ∈ R3 .
Let χ(τ ) and ϕ(τ ) be smooth functions such that
n 3 8o X
Supp ϕ ⊂ τ ∈ R : < τ < and ∀τ > 0 , ϕ(2−q τ ) = 1,
4 3
q∈Z

def 
we define the dyadic operators as follows: for u ∈ Sh′ = u ∈ S ′, lim kθ(λ D)ukL∞ =
λ→+∞
0 for any θ ∈ D(R3 ) ,
X
˙ q u def
∆ = ϕ(2−q |D|)u ∀q ∈ Z, and Ṡq u =
def ˙ j u,
∆
(A.1)
j≤q−1

The dyadic operators satisfy the property of almost orthogonality:

˙ k∆
∆ ˙ q u ≡ 0 if |k − q| ≥ 2 and ˙ k (Ṡq−1 u∆
∆ ˙ q u) ≡ 0 if |k − q| ≥ 5,
s
Definition A.1. Let s ∈ R, 1 ≤ p, r ≤ +∞, we define the homogeneous Besov space Ḃp,r
def
to be the set of distributions u in Sh′ (Sh′ = {u ∈ S ′ , lim kθ(λ D)ukL∞ = 0 for any θ ∈
λ→+∞
D(R3 )}) so that
def ˙ q ukLp
kukḂ s = 2qs k∆ < ∞.
p,r ℓr (Z)

Remark A.1. s =
(1) We point out that if s > 0 then the inhomogeneous Besov space Bp,r
s ∩ Lp and
Ḃp,r
kukBp,r
s ≈ kukḂ s + kukLp .
p,r

s s̃ θs+(1−θ)s̃
(2) If u ∈ Ḃp,∞ ∩ Ḃp,∞ with s < s̃, 1 ≤ p ≤ ∞, then for any θ ∈ (0, 1), u ∈ Ḃp,1
and
C 1 1

(A.2) kukḂ θs+(1−θ)s̃ ≤ + kukθḂ s kuk1−θ
s̃
.
p,1 s̃ − s θ 1 − θ p,∞ Ḃp,∞

(3) Let s ∈ R, 1 ≤ p, r ≤ +∞, and u ∈ Sh′ . Then u belongs to Ḃp,r

s if and only if there

exists some positive constant C and some nonnegative generic element {cq,r }q∈Z of
ℓr (Z) such that k{cq,r }q∈Z kℓr (Z) = 1 and for any q ∈ Z
˙ q ukLp ≤ Ccq,r 2−qs kuk s .
k∆ Ḃ p,r

We also recall Bernstein’s lemma from [4]:

def def
Lemma A.1. Let B = {ξ ∈ R3 , |ξ| ≤ 43 } be a ball and C = {ξ ∈ R3 , 43 ≤ |ξ| ≤ 83 } be an
annulus. A constant C exists so that for any positive real number λ, any nonnegative integer
k, any smooth homogeneous function σ of degree m, any couple of real numbers (a, b) with
b ≥ a ≥ 1, and any function u in La , there hold
1 1
Supp û ⊂ λB ⇒ sup k∂ α ukLb ≤ C k+1 λk+3( a − b ) kukLa ,
|α|=k

(A.3) Supp û ⊂ λC ⇒ C −1−k λk kukLa ≤ sup k∂ α ukLa ≤ C 1+k λk kukLa ,

|α|=k
1 1
Supp û ⊂ λC ⇒ kσ(D)ukLb ≤ Cσ,m λm+3( a − b ) kukLa ,
def
with σ(D)u = F −1 (σ û).
26 H. ABIDI, G. GUI, AND P. ZHANG

In what follows, we shall frequently use Bony’s decomposition [6] in the homogeneous
context, which reads
uv =Tu v + Tv′ u = Tu v + Tv u + R(u, v) with
def X ˙ q v, T ′ u def
X
˙ q u Ṡq+2 v,
Tu v = Ṡq−1 u∆ v = ∆ and
(A.4) q∈Z q∈Z
def X ė v with ∆
ė v def X
R(u, v) = ˙ q u∆
∆ q q =
˙ q′ v.
∆
q∈Z |q ′ −q|≤1

Proposition A.1 (Theorems 2.47 and 2.52 in [4]). (1) There exits a constant C so that
for s ∈ R, t < 0, p, p1 , p2 , r, r1 , r2 ∈ [1, +∞],
kTu vkḂ s ≤ C |s|+1kukL∞ kvkḂ s ,
p,r p,r

C |s+t|+1 1 def 1 1 1 def  1 1

kTu vkḂp,r
s+t ≤ kukḂ t kvkḂ s with = + , = min 1, + .
−t p1 ,r1 p2 ,r2 p p1 p2 r r1 r2
1 def 1
(2) Let (s1 , s2 ) be in R2 and (p1 , p2 , r1 , r2 ) be in [1, +∞]4 . We assume that p = p1 + p12 ≤
1 def 1 1
1 and r = r1 + r2 ≤ 1. Then there exits a constant C so that

C s1+s2 +1
kR(u, v)kḂ s1 +s2 ≤ kukḂps1 ,r kvkḂps2 ,r if s1 + s2 > 0,
p,r s1 + s2 1 1 2 2

kR(u, v)kḂ 0 ≤ CkukḂps1 ,r kvkḂps2 ,r if r = 1 and s1 + s2 = 0.

p,∞ 1 1 2 2

In order to obtain a better description of the regularizing effect of the transport-diffusion

equation, we shall use Chemin-Lerner type norm from [8].
Definition A.2. Let s ∈ R, r, λ, p ∈ [1, +∞] and T > 0. we define
def ˙ q uk λ p
kukLe λ (Ḃ s = 2qs k∆ L (L ) .
T p,r ) T ℓr (Z)

Next we recall the following commutator’s estimate which will be frequently used through-
out this paper.
1 1
Lemma A.2 ( Lemma 2.97 in [4]). Let (p, q, r) ∈ [1, +∞]3 satisfy r = p + 1q , θ be a C 1
function on Rd such that (1+ |·|)θ̂ ∈ L1 . There exists a constant C such that for any function
a with gradient in Lp and any function b in Lq , we have, for any positive λ,
(A.5) k[θ(λ−1 D), a]bkLr ≤ Cλ−1 k∇akLp kbkLs .
Before recalling the definition of the Lorentz space, we begin by introducing the rearrange-
ment of a function (see [15] for instance). For a measurable function f, we define its non-
increasing rearrangement by f ∗ : R+ → R+ by
n o
def
f ∗ (λ) = inf s ≥ 0; { x ∈ R3 : |f (x)| > s } ≤ λ ,

where { x ∈ R3 : |f (x)| > s } denotes the Lebesgue measure of the set { x ∈ R3 : |f (x)| >
s }.
Definition A.3. (Lorentz spaces) Let f a mesurable function and 1 ≤ p, q ≤ ∞. Then f
belongs to the Lorentz space Lp,q if
 1
 R 1
 ∞ (t p f ∗ (t))q dt q < ∞ if q < ∞
def 0 t
kf kLp,q = 1

 ∗
sup t f (t) < ∞
p if q = ∞.
t>0
 FUJITA-KATO SOLUTION OF 3−D INHOMOGENEOUS NS EQUATIONS 27

Alternatively, we can define the Lorentz spaces by the real interpolation, as the interpola-
tion between the Lebesgue space :
def
Lp,q = (Lp0 , Lp1 )(θ,q) ,
with 1 ≤ p0 < p < p1 ≤ ∞, 0 < θ < 1 satisfying p1 = 1−θ θ
p0 + p1 and 1 ≤ q ≤ ∞, also f ∈ L
p,q

if the following quantity

Z ∞
q dt  1q
def
kf kLp,q = t−θ K(t, f )
0 t
is finite with
def 
K(f, t) = inf kf0 kLp0 + tkf1 kLp1 f 0 ∈ Lp 0 , f 1 ∈ Lp 1 .
f =f0 +f1

The Lorentz spaces verify the following properties (see [18, 20] for more details) :
Proposition A.2. Let f ∈ Lp1 ,q1 , g ∈ Lp2 ,q2 and 1 ≤ p, q, pj , qj ≤ ∞, for 1 ≤ j ≤ 2.

(1) If 1 < p < ∞ and 1 ≤ q ≤ ∞, then

kf gkLp,q . kf kLp,q kgkL∞ .
1 1 1 1 1 1
(2) If p = p1 + p2 and q = q1 + q2 , then
kf gkLp,q . kf kLp1 ,q1 kgkLp2 ,q2 .
(3) For 1 ≤ p ≤ ∞ and 1 ≤ q1 ≤ q2 ≤ ∞, we have
Lp,q1 ֒→ Lp,q2 and Lp,p = Lp .
Proposition A.3. The following imbedding relations hold
(1) Ḣ 1 (R3 ) ֒→ L6,2 (R3 ).
3
−1
p
(2) Ḃp,r (R3 ) ֒→ L3,r (R3 ), for 1 ≤ p < 3 and r ∈ [1, ∞].
Proof. The first part can be found in [21]. We focus on the proof of the second part.
Let (p, p1 ) be a fixed exponents pair satisfying 1 ≤ p < 3 < p1 ≤ ∞ and r ∈ [1, ∞]. By
definition, we have
1 1−θ θ
(A.6) (Lp , Lp1 )(θ,r) = L3,r , with = + ·
3 p p1
While it follows from Lemma A.1 that
3
− p3
֒→ Ḃp01 ,1 ֒→ Lp1 0
֒→ Lp ,
p 1
Ḃp,1 and Ḃp,1
from which and (A.6), we infer
0
3
− p3

֒→ L3,r .
p 1
(A.7) Ḃp,1 , Bp,1 (θ,r)

On the other hand, we have (see for instance [5] page 152),
0
3
p
− p3
1

θ( p3 − p3 )
1 p
3
−1
Ḃp,1 , Ḃp,1 (θ,r)
= Ḃp,r = Ḃp,r ,

which together with (A.7) ensures the second part of the lemma. 
Acknowledgments. G. Gui is supported in part by National Natural Science Foundation
of China under Grants 12371211 and 12126359. P. Zhang is supported by National Key R&D
Program of China under grant 2021YFA1000800 and National Natural Science Foundation
of China under Grants 12288201 and 12031006.
28 H. ABIDI, G. GUI, AND P. ZHANG

References
[1] H. Abidi, Équation de Navier-Stokes avec densité et viscosité variables dans l’espace critique, Rev. Mat.
Iberoam., 23 (2) (2007), 537–586.
[2] H. Abidi, G. Gui, and P. Zhang, On the wellposedness of 3−D inhomogeneous Navier-Stokes equations
in the critical spaces, Arch. Rational Mech. Anal., 204 (2012), 189–230.
[3] H. Abidi and M. Paicu, Existence globale pour un fluide inhomogéne, Ann. Inst. Fourier (Grenoble), 57
(2007), 883–917.
[4] H. Bahouri, J. Y. Chemin, and R. Danchin, Fourier Analysis and Nonlinear Partial Differential Equations,
Grundlehren der Mathematischen Wissenschaften, Springer, 2011.
[5] J. Bergh and J. Löfström, Interpolation spaces. An introduction, Springer-Verlag, 1976.
[6] J. M. Bony, Calcul symbolique et propagation des singularités pour les équations aux dérivées partielles
non linéaires, Ann. Sci. École Norm. Sup., 14 (1981), 209–246.
[7] M. Cannone, Y. Meyer, and F. Planchon, Solutions autosimilaires des équations de Navier-Stokes,
Séminairé Équations aux Dérivées Partielles de l’École Polytechnique, 1993-1994.
[8] J.-Y. Chemin and N. Lerner, Flot de champs de vecteurs non lipschitziens et équations de Navier-Stokes,
J. Differential Equations, 121 (1995), 314–328.
[9] D. Chen, Z. Zhang, and W. Zhao, Fujita-Kato theorem for the 3-D inhomogeneous Navier-Stokes equa-
tions, J. Differential Equations, 261 (2016), 738–761.
[10] R. Danchin, Density-dependent incompressible viscous fluids in critical spaces, Proc. R. Soc. Edin. Sect.
A, 133 (2003), 1311–1334.
[11] R. Danchin, Local and global well-posedness resultats for flows of inhomogenenous viscous fluids, Adv.
differential equations, 9 (2004), 353–386.
[12] R. Danchin and P. B. Mucha, A Lagrangian approach for the incompressible Navier-Stokes equations
with variable density, Comm. Pure. Appl. Math., 65 (2012), 1458–1480.
[13] R. Danchin and S. Wang, Global unique solutions for the inhomogeneous Navier-Stokes equations with
only bounded density in critical regularity spaces, Comm. Math. Phy., 399 (2023), 1647–1688.
[14] H. Fujita and T. Kato, On the Navier-Stokes initial value problem. I, Arch. Ration. Mech. Anal., 16
(1964), 269–315.
[15] L. Grafakos, Classical Fourier analysis, Springer, New York, third edition, 2014.
[16] T. Hao, F. Shao, D. Wei, and Z. Zhang, Global well-posedness of inhomogeneous Navier-Stokes equations
with bounded density, arXiv:2406.19907, 2024.
[17] J. Huang, M. Paicu, and P. Zhang, Global wellposedness to incompressible inhomogeneous fluid system
with bounded density and non-Lipschitz velocity, Arch. Ration. Mech. Anal., 209 (2013), 631-682.
[18] P. G. Lemarié-Rieusset, Recent Developments in the Navier-Stokes Problem, Chapman & Hall/CRC Re-
search Notes in Mathematics, 431 Chapman & Hall/CRC, Boca Raton, FL, 2002.
[19] P. L. Lions, Mathematical Topics in Fluid Mechanics. Vol. 1. Incompressible Models, Oxford Lecture
Series in Mathematics and its Applications, 3. Oxford Science Publications. The Clarendon Press, Oxford
University Press, New York, 1996.
[20] R. O’Neil, Convolution operators and L(p,q) spaces, Duke Math. J., 30 (1963), 129–142.
[21] L. Tartar: Imbedding theorems of Sobolev spaces into Lorentz spaces, Bollettino dell’Unione Matematica
Italiana, Serie 8, 1-B (3) (1998), 479–500.
[22] P. Zhang, Global Fujita-Kato solution of 3-D inhomogeneous incompressible Navier-Stokes system, Adv.
Math., 363 (2020), 107007, 43 pp.

(H. Abidi) Département de Mathématiques Faculté des Sciences de Tunis Université de Tunis
EI Manar 2092 Tunis Tunisia
Email address: [email protected]

(G. Gui) School of Mathematics and Computational Science, Xiangtan University, Xiangtan
411105, China
Email address: [email protected]

(P. Zhang) Academy of Mathematics & Systems Science, The Chinese Academy of Sciences,
Beijing 100190, China;, School of Mathematical Sciences, University of Chinese Academy of
Sciences, Beijing 100049, China
Email address: [email protected]