Skip to content
BY 4.0 license Open Access Published by De Gruyter Open Access March 14, 2024

Local and global solvability for the Boussinesq system in Besov spaces

  • Shuokai Yan , Lu Wang and Qinghua Zhang EMAIL logo
From the journal Open Mathematics

Abstract

This article focuses on local and global existence and uniqueness for the strong solution to the Boussinesq system in R n ( n 3 ) with full viscosity in Besov spaces. Under the hypotheses 1 < p < and min { n p , 2 n p } < s n p , and the initial condition ( θ 0 , u 0 ) B ˙ p , 1 s 1 × B ˙ p , 1 n p 1 , the Boussinesq system is proved to have a unique local strong solution. Under the hypotheses n p < and n p < s n p , or especially n p < 2 n and n p < s < n p 1 , and the initial condition ( θ 0 , u 0 ) ( B ˙ p , 1 s 1 L n 3 ) × ( B ˙ p , 1 n p 1 L n ) with sufficiently small norms θ 0 L n 3 and u 0 L n , the Boussinesq system is proved to have a unique global strong solution.

MSC 2010: 35Q30; 76D05

1 Introduction

Motion of the ocean or the atmosphere can be simulated by the Boussinesq system. By Boussinesq approximation, we can neglect variation of the density of the fluid and derive a simplified model as follows:

(1.1) t θ ν Δ θ + u θ = 0 , t > 0 , x R n ; t u μ Δ u + u u + π = κ θ e n , t > 0 , x R n ; u = 0 , t > 0 , x R n ; θ ( 0 , x ) = θ 0 ( x ) , u ( 0 , x ) = u 0 ( x ) , x R n .

Here u ( x , t ) = ( u 1 ( x , t ) , u 2 ( x , t ) , , u n ( x , t ) ) , θ ( x , t ) , and π ( x , t ) represent the velocity, temperature (or salinity), and inner pressure of the fluid at the point x R n ( n 3 ) and the time t > 0 , respectively. However, u 0 ( x ) and θ 0 ( x ) represent initial distributions of the velocity and temperature (or salinity) in the fluid. Because of the variation of the temperature (or salinity), a vortex buoyancy force κ θ e n arises, where κ > 0 is the proportion coefficient, and e n = ( 0 , , 0 , 1 ) is the unit vertical vector. For the sake of simplicity, by re-scaling of the unknowns, we always assume that κ = 1 . Moreover, the viscosity coefficients ν , μ appearing in the diffusion terms are both assumed to be larger than 0.

Danchin and Paicu [1] and Cannon and DiBenedetto [2] addressed global existence of the weak solution of (1.1) under the initial condition ( θ 0 , u 0 ) L 2 ( R n ) × L σ 2 ( R n ) , where L σ 2 ( R n ) denotes the collection of all solenoidal vector fields with L 2 components. Brandolese and Schonbek [3] and Han [4] investigated time-decaying property of the weak solutions in energy space under the additional assumption θ 0 L 1 . Danchin and Paicu [5] readdressed existence and uniqueness of the global weak solution by assuming that ( θ 0 , u 0 ) ( L n 3 , L p 0 , ) × L n , for some p 0 > 1 , and

(1.2) μ 1 θ 0 L n 3 , + u L n , c μ ,

for sufficiently small number c > 0 . Here L r , denotes the Lorentz space, which is equivalent to the weak L r space for r > 1 (refer to [6, §7.24]). If n = 3 , then L n 3 , is replaced by L 1 .

We are much concerned with strong solvability of the Boussinesq system. Under the initial assumption θ 0 B ˙ n , 1 0 L n 3 , and u 0 B ˙ p , 1 n p 1 L n , for some p n , and condition (1.2), where θ 0 L n 3 , is replaced by θ 0 L n 3 , [1] proved that system (1.1) with partial viscosity ( ν = 0 ) admits a unique global strong solution ( θ , u ) in the class

C ( [ 0 , ) ; B ˙ n , 1 0 ) × ( C ( [ 0 , ) ; B ˙ p , 1 n p 1 ) L loc 1 ( 0 , ; B ˙ n , 1 n p + 1 ) ) .

Under the initial assumption

θ 0 L 1 + θ 0 L ( R 3 , x 3 d x ) + u 0 L ( R 3 , x d x ) ε

for some sufficiently small number ε > 0 , [3] proved that 3 D Boussinesq system (1.1) with full viscosity has a unique global strong solution ( θ , u ) in the class

( L ( 0 , ; L 1 ) L ( ( 0 , ) × R 3 ; ( t + x ) 3 d t d x ) ) × L ( ( 0 , ) × R 3 ; ( t + x ) d t d x ) .

Here the initial spaces L n 3 , L ( R 3 , x 3 d x ) , and L n , B ˙ p , 1 n p 1 , L ( R 3 , x d x ) are critical, in other words, they are invariant under the following scaling transformation:

θ ( x , t ) λ 3 θ ( λ x , λ 2 t ) , u ( x , t ) λ u ( λ x , λ 2 t ) , θ 0 ( x ) λ 3 θ 0 ( λ x ) , u 0 ( x ) λ u 0 ( λ x ) .

So [1,3] can be viewed as partial extensions of [79], where the critical space H ˙ 1 2 or L n was employed to deal with the global and strong solvability for the classical Navier-Stokes equations.

This article focuses on the local and global strong solvability of the Boussinesq system (1.1) with full viscosity in Besov spaces. Because of the existence of the diffusion term ν Δ θ , the first equation of (1.1) exhibits different mechanical characteristics to the transport equation. This means that we could not make a time-independent estimate for the function θ under the single assumption u L 1 ( ( 0 , T ) ; L ) anymore. In fact, the two equations in (1.1) together comprise a mixed evolution system, where asymptotic actions of θ and u are influenced by the convective terms u θ and u u simultaneously.

By reducing (1.1) into a sequence of approximate systems and making estimates for the approximate solutions, we will prove that if 1 < p < and min { n p , 2 n p } < s n p , then for any ( θ 0 , u 0 ) B ˙ p , 1 s 1 × B ˙ p , 1 n p 1 , system (1.1) possesses a unique strong solution ( θ , u , π ) on a short interval [ 0 , T ] in the class

( C ( [ 0 , T ] ; B ˙ p , 1 s 1 ) L T 1 ( B ˙ p , 1 s + 1 ) ) × ( C ( [ 0 , T ] ; B ˙ p , 1 n p 1 ) L T 1 ( B ˙ p , 1 n p + 1 ) ) × L T 1 ( B ˙ p , 1 n p 1 ) ,

where the existing time T is uniform on a neighbourhood of ( θ 0 , u 0 ) in B ˙ p , 1 s 1 × B ˙ p , 1 n p 1 (see Theorem 2.4 and Remark 3.1).

We also pursuit global existence and uniqueness of the strong solution under the assumption n p < and n p < s n p , or especially n p < 2 n and n p < s < n p 1 . By making investigations on the global boundedness and convergence of the approximate solutions in L p spaces, we will prove that if ( θ 0 , u 0 ) ( B ˙ p , 1 s 1 L n 3 ) × ( B ˙ p , 1 n p 1 L n ) and

θ 0 L n 3 ε min { μ , ν } 2 , u 0 L n ε min { μ , ν }

for some sufficiently small number ε > 0 , then the local strong solution of (1.1) can be extended to the whole interval [ 0 , ) , and the norms

θ L ˜ t ( B ˙ p , 1 s 1 ) + ν θ L t 1 ( B ˙ p , 1 s + 1 ) and u L ˜ t ( B ˙ p , 1 n p 1 ) + μ u L t 1 ( B ˙ p , 1 n p + 1 ) + π L t 1 ( B ˙ p , 1 n p 1 )

can be controlled by the initial data θ 0 B ˙ p , 1 s 1 + θ 0 L n 3 (or θ 0 B ˙ p , 1 s 1 singly) and u 0 B ˙ p , 1 n p 1 uniformly on any bounded interval (see Theorem 2.5).

This is a beneficial attempt to deal with strong solvability of the Boussinesq system (1.1) with full viscosity in Besov spaces. Compared with [1,5], treatment of the first equation of (1.1) here is much different. In order to make suitable estimates for θ from the transport-diffusion equation in general case n p < and n p < s n p , another function space L t ( B ˙ p , σ 1 ) L ˜ t 1 ( B ˙ p , σ + 1 ) with lower index σ ( 1 , min { s , n p 1 } ) is applied. We think here the value of σ is taken reasonable. In fact, it fits the treatment of θ u L ˜ t 1 ( B ˙ p , 1 σ ) through Bony’s decomposition well. Since L n 3 × L n is a pair of critical spaces for the initial data, and L n 3 = L 1 for n = 3 , initial assumption employed here is optimal.

2 Preliminaries and main results

We first make a brief review on the theory of Littlewood-Paley decomposition and homogeneous Besov spaces, for the detailed discussions, please refer to [10, §2.3]. Let C 0 ( R n ) be the collection of all smooth functions with compact supports and S ( R n ) be the set of all temperate distributions. For each 1 p , let L p ( R n ) be the common Lebesgue space, scalar or vector-valued type, with the norm denoted by L p .

Suppose that φ C 0 ( R n ) with 0 φ 1 and supp φ { ξ R n : 3 4 ξ 8 3 } such that

q Z φ ( 2 q ξ ) = 1 for all ξ R n \ { 0 } .

For each q Z and f S ( R n ) , define

Δ ˙ q f = 1 ( φ ( 2 q ξ ) f ) and S ˙ q f = 1 ( χ ( 2 q ξ ) f ) ,

where χ ( ξ ) = 1 q 0 φ ( 2 q ξ ) for all ξ R n . It is easy to check that Δ ˙ q f L p f L p and S ˙ q f L p f L p for all 1 p , where A B means that A C B for some C > 0 .

Define

S h ( R n ) = { f S ( R n ) : lim λ 1 ( θ ( λ ξ ) f ) L = 0 for all θ C 0 ( R n ) } .

By this definition, it is easy to check that

q Z Δ ˙ q f = f and S ˙ q f = q q 1 Δ ˙ q f in S ( R n )

for all f S h ( R n ) .

For each triple of exponents ( s , p , r ) R × [ 1 , ] × [ 1 , ] , define the homogeneous Besov space B ˙ p , r s as follows:

B ˙ p , r s = { f S h ( R n ) : f B ˙ p , r s = ( 2 q s Δ ˙ q f L p ) l r < } ,

where ( a q ) l r = ( q Z a q r ) 1 r if 1 r < and ( a q ) l r = sup q Z a q if r = . Evidently, endowed with the norm B ˙ p , r s defined above, B ˙ p , r s is a normed space. Additionally, if s < n p or s = n p and r = 1 , then B ˙ p , r s turns to be a Banach space. In this case, f S h ( R n ) for all f S ( R n ) verifying f B ˙ p , r s < . There are some elementary properties of Besov spaces. If 1 p 1 p 2 and 1 r 1 r 2 , then B ˙ p 1 , r 1 s B ˙ p 2 , r 2 s n ( 1 p 1 1 p 2 ) and L p 1 s B ˙ p 2 , n ( 1 p 1 1 p 2 ) . For any f S h ( R n ) , it holds that f S h ( R n ) . Moreover, f B ˙ p , r s + 1 if and only if f B ˙ p , r s , and the norms f B ˙ p , r s + 1 and f B ˙ p , r s are equivalent.

Lemma 2.1

[11] Let s R , p , r [ 1 , ] such that

(2.1) min { n p , n p } < s < n p ,

then we have

(2.2) u v B ˙ p , r s u B ˙ p , r s v B ˙ p , n p L .

If s = n p and r = 1 , then B ˙ p , r s is an algebra, and

(2.3) u v B ˙ p , 1 n p u B ˙ p , 1 n p v B ˙ p , 1 n p .

If s = min { n p , n p } and r = , then

(2.4) u v B ˙ p , s u B ˙ p , s v B ˙ p , 1 n p .

Suppose that s R , p , r , β [ 1 , ] , and 0 < T , suppose also f S ( ( 0 , T ) × R n ) such that f ( t ) S h ( R n ) for all 0 < t < T . We say f L ˜ β ( ( 0 , T ) ; B ˙ p , r s ) (or L ˜ T β ( B ˙ p , r s ) if T < ), if

f L ˜ β ( ( 0 , T ) ; B ˙ p , r s ) ( 2 q s Δ ˙ q f L β ( ( 0 , T ) ; L p ) ) l r < .

Direct calculation shows that

L ˜ β ( ( 0 , T ) ; B ˙ p , r s ) L β ( ( 0 , T ) ; B ˙ p , r s ) , f L β ( ( 0 , T ) ; B ˙ p , r s ) f L ˜ β ( ( 0 , T ) ; B ˙ p , r s )

for r β , and

L β ( ( 0 , T ) ; B ˙ p , r s ) L ˜ β ( ( 0 , T ) ; B ˙ p , r s ) , f L ˜ β ( ( 0 , T ) ; B ˙ p , r s ) f L β ( ( 0 , T ) ; B ˙ p , r s )

for β r .

Lemma 2.2

[1] Let T > 0 , s R , and p , r [ 1 , ] . If we assume g 0 B ˙ p , r s and f L ˜ T 1 ( B ˙ p , r s ) , then equation

g ( t ) = e ν t Δ g 0 + 0 t e ν ( t τ ) Δ f ( τ ) d τ , t > 0

has a unique solution g L ˜ T ( B ˙ p , r s ) L ˜ T 1 ( B ˙ p , r s + 2 ) verifying the following estimate:

(2.5) g L ˜ t ( B ˙ p , r s ) + ν g L ˜ t 1 ( B ˙ p , r s + 2 ) C ( g 0 B ˙ p , r s + f L ˜ t 1 ( B ˙ p , r s ) )

for all 0 < t T .

Lemma 2.3

[11] Let T > 0 , p ( 1 , ) , 1 min { n p , n p } < α < n p , and r [ 1 , ] or α = 1 min { n p , n p } and r = . Assume that v is a vector field such that div v = 0 and v L T 1 ( B ˙ p , n p L ) or respectively v L T 1 ( B ˙ p , 1 n p ) . Assume also g 0 B ˙ p , r α and f L 1 ( B ˙ p , r α ) . If g C ( [ 0 , T ] ; B ˙ p , r α ) L T 1 ( B ˙ p , r α ) is a solution of the initial problem of the transport-diffusion equation

t g ν Δ g + v g = f ( t ) , g ( 0 ) = g 0 ,

then it holds that

(2.6) g L ˜ t ( B ˙ p , r α ) + ν g L ˜ t 1 ( B ˙ p , r α + 2 ) C e C V ( t ) ( g 0 B ˙ p , r α + f L t 1 ( B ˙ p , r α ) )

for all 0 < t T , where V ( t ) = v L t 1 ( B ˙ p , n p L ) or V ( t ) = v L t 1 ( B ˙ p , 1 n p ) .

Hereinafter, C > 0 denotes a universal constant, it may even change from line to line, but does not depend on the time t and involved functions.

Now we can state main results of the article.

Theorem 2.4

Suppose 1 < p < , s R such that min { n p , 2 n p } < s n p , and ( θ 0 , u 0 ) B ˙ p , 1 s 1 × B ˙ p , 1 n p 1 . Then there exists T > 0 such that on the interval [ 0 , T ] , Boussinesq system (1.1) has a uniqueness strong solution

( θ , u , π ) ( C ( [ 0 , T ] ; B ˙ p , 1 s 1 ) L ˜ T ( B ˙ p , 1 s 1 ) L T 1 ( B ˙ p , 1 s + 1 ) ) × ( C ( [ 0 , T ] ; B ˙ p , 1 n p 1 ) L ˜ T ( B ˙ p , 1 n p 1 ) L T 1 ( B ˙ p , 1 n p + 1 ) ) × L ˜ T ( B ˙ p , 1 n p 1 ) .

Theorem 2.5

Suppose that n p < and n p < s n p . Then there exists a small number ε > 0 , such that under the initial condition ( θ 0 , u 0 ) ( B ˙ p , 1 s 1 L n 3 ) × ( B ˙ p , 1 n p 1 L n ) with the restriction

(2.7) θ 0 L n 3 ε min { μ , ν } 2 , u 0 L n ε min { μ , ν } ,

Problem (1.1) permits a unique global solution ( θ , u , π ) belonging to

( L ( ( 0 , ) ; B ˙ p , σ 1 ) L ˜ 1 ( ( 0 , ) ; B ˙ p , σ + 1 ) C ( [ 0 , ) ; B ˙ p , 1 s 1 ) L ˜ loc ( ( 0 , ) ; B ˙ p , 1 s 1 ) L loc 1 ( ( 0 , ) ; B ˙ p , 1 s + 1 ) ) × ( C ( [ 0 , ) ; B ˙ p , 1 n p 1 ) L loc 1 ( ( 0 , ) ; B ˙ p , 1 n p + 1 ) ) × L loc ( ( 0 , ) ; B ˙ p , 1 n p 1 ) ,

and obeying the following estimates:

(2.8) θ L t ( B ˙ p , σ 1 ) + ν θ L ˜ t 1 ( B ˙ p , σ + 1 ) C ( θ 0 B ˙ p , 1 s 1 + θ 0 L n 3 ) ,

(2.9) θ L ˜ t ( B ˙ p , 1 s 1 ) + ν θ L t 1 ( B ˙ p , 1 s + 1 ) C θ 0 B ˙ p , 1 s 1 exp { C μ 1 [ u 0 B ˙ p , 1 n p 1 + ( t + ν 1 ) ( θ 0 B ˙ p , 1 s 1 + θ 0 L n 3 ) ] } ,

and

(2.10) u L t ( B ˙ p , 1 n p 1 ) + μ u L t 1 ( B ˙ p , 1 n p + 1 ) + π L t 1 ( B ˙ p , 1 n p 1 ) d τ C [ u 0 B ˙ p , 1 n p 1 + ( t + ν 1 ) ( θ 0 B ˙ p , 1 s 1 + θ 0 L n 3 ) ]

for all t > 0 , where 1 < σ < min { n p 1 , s } .

If we assume further n p < 2 n and n p < s < n p 1 , then the strong solution ( θ , u , π ) also lies in

( C ( [ 0 , ) ; B ˙ p , 1 s 1 ) L ( ( 0 , ) ; B ˙ p , 1 s 1 ) L 1 ( ( 0 , ) ; B ˙ p , 1 s + 1 ) ) × ( C ( [ 0 , ) ; B ˙ p , 1 n p 1 ) L loc 1 ( ( 0 , ) ; B ˙ p , 1 n p + 1 ) ) × L loc ( ( 0 , ) ; B ˙ p , 1 n p 1 )

and verifies the following estimates:

(2.11) θ L ( ( 0 , ) ; B ˙ p , 1 s 1 ) + ν θ L 1 ( ( 0 , ) ; B ˙ p , 1 s + 1 ) C θ 0 B ˙ p , 1 s 1

and

(2.12) u L t ( B ˙ p , 1 n p 1 ) + μ u L t 1 ( B ˙ p , 1 n p + 1 ) + π L t 1 ( B ˙ p , 1 n p 1 ) C ( u 0 B ˙ p , 1 n p 1 + t 1 ( n p s ) 2 ν ( n p s ) 2 θ 0 B ˙ p , 1 s 1 )

for all t > 0 .

Remark 2.6

Evidently, θ L ( ( 0 , ) ; B ˙ p , 1 n p 1 ) L 1 ( ( 0 , ) ; B ˙ p , 1 n p + 1 ) is better than θ L loc ( ( 0 , ) ; B ˙ p , 1 n p 1 ) L loc 1 ( ( 0 , ) ; B ˙ p , 1 n p + 1 ) . Moreover, estimates (2.11) and (2.12) are superior to (2.9) and (2.10) respectively. Hence, the strong solution exhibits better asymptotic behaviour on [ 0 , ) in the special case n p < 2 n and n p < s < n p 1 .

Remark 2.7

Here assumptions on the value of p , s , and σ are natural. In fact, assumption min { n p , 2 n p } < s n p comes from Lemma 2.2 and the requirement B ˙ p , 1 s 1 B ˙ p , 1 s + 1 B ˙ p , 1 n p 1 , which guarantee the consistency of the different equations in system (1.1). Moreover, assumption σ < min { n p 1 , s } fits the requirement B ˙ p , 1 s 1 L n 3 B ˙ p , σ 1 , meanwhile p n and σ > 1 fit the treatment of θ u L t 1 ( B ˙ p , σ ) (or θ u L t 1 ( B ˙ p , 1 s ) ). In fact, it goes well through Bony’s decomposition (see (3.15) and (3.16) (or (3.24) and (3.26))).

3 Proofs and remarks

Proof of Theorem 2.4

Let 0 < T < and introduce two function spaces

X T = L ˜ T ( B ˙ p , 1 n p 1 ) L T 1 ( B ˙ p , 1 n p + 1 ) , Y T = L ˜ T ( B ˙ p , 1 s 1 ) L T 1 ( B ˙ p , 1 s + 1 ) ,

where the norms are defined by

u X T = u L ˜ T ( B ˙ p , 1 n p 1 ) + μ u L T 1 ( B ˙ p , 1 n p + 1 ) , θ Y T = θ L ˜ T ( B ˙ p , 1 s 1 ) + ν θ L T 1 ( B ˙ p , 1 s + 1 ) .

Step 1. Construction of the approximate solutions.

Define θ L ( t ) = e t ν Δ θ 0 and u L ( t ) = e t μ Δ u 0 . In light of Lemma 2.2, we have θ L Y T , u L X T , and

θ L Y T C θ 0 B ˙ p , 1 s 1 , u L X T C u 0 B ˙ p , 1 n p 1 .

Consider the function series { ( θ ¯ k , u ¯ k ) } , where ( θ ¯ 0 , u ¯ 0 ) = ( 0 , 0 ) , and ( θ ¯ k + 1 , u ¯ k + 1 ) is the integral solution of the following parabolic equation system:

(3.1) t θ ¯ k + 1 ν Δ θ ¯ k + 1 = div ( θ k u k ) , t > 0 , t u ¯ k + 1 μ Δ u ¯ k + 1 = P div ( u k u k ) + P ( θ k e n ) , t > 0 , θ ¯ k + 1 ( 0 ) = 0 , u ¯ k + 1 ( 0 ) = 0 ,

where θ k = θ L + θ ¯ k , u k = u L + u ¯ k , and P = I + ( Δ ) 1 div is the Leray projection.

According to Lemmas 2.1 and 2.2, with the method of interpolation, we can deduce that

(3.2) θ ¯ k + 1 Y T C 0 T div ( θ k ( τ ) u k ( τ ) ) B ˙ p , 1 s 1 d τ C 0 T θ k ( τ ) u k ( τ ) B ˙ p , 1 s d τ C θ L + θ ¯ k L T 2 ( B ˙ p , 1 s ) u L + u ¯ k L T 2 ( B ˙ p , 1 n p ) C ( θ L L T 2 ( B ˙ p , 1 s ) + θ ¯ k L ˜ T ( B ˙ p , 1 s 1 ) 1 2 θ ¯ k L T 1 ( B ˙ p , 1 s + 1 ) 1 2 ) ( u L L T 2 ( B ˙ p , 1 n p ) + u ¯ k L ˜ T ( B ˙ p , 1 n p 1 ) 1 2 u ¯ k L T 1 ( B ˙ p , 1 n p + 1 ) 1 2 ) C ( θ L L T 2 ( B ˙ p , 1 s ) + ν 1 2 θ ¯ k Y T ) ( u L L T 2 ( B ˙ p , 1 n p ) + μ 1 2 u ¯ k X T ) .

Similarly, by taking

ρ = 2 ( n p s ) 1 , if s < n p , , if s = n p ,

one can deduce that

(3.3) u ¯ k + 1 X T C 0 T div ( u k ( τ ) u k ( τ ) ) B ˙ p , 1 n p 1 d τ + θ k L T 1 ( B ˙ p , 1 n p 1 ) C 0 T u k ( τ ) u k ( τ ) B ˙ p , 1 n p d τ + T 1 1 ρ θ L + θ ¯ k L T ρ ( B ˙ p , 1 n p 1 ) C u L + u ¯ k L T 2 ( B ˙ p , 1 n p ) 2 + C T 1 1 ρ ( θ L L ˜ T ( B ˙ p , 1 s 1 ) 1 1 ρ θ L L T 1 ( B ˙ p , 1 s + 1 ) 1 ρ + θ ¯ k L ˜ T ( B ˙ p , 1 s 1 ) 1 1 ρ θ ¯ k L T 1 ( B ˙ p , 1 s + 1 ) 1 ρ ) C ( u L L T 2 ( B ˙ p , 1 n p ) + μ 1 2 u ¯ k X T ) 2 + C T 1 1 ρ ν 1 ρ ( θ 0 B ˙ p , 1 s 1 + θ ¯ k Y T ) .

Let

c 1 = ( 4 C ) 1 μ 1 2 ν 1 2 , c 2 = ( 8 C ) 1 μ 1 2 min { μ 1 2 , ν 1 2 } ,

and take T > 0 so small that

T 1 1 ρ < ( 2 C ) 1 min { c 2 ( θ 0 B ˙ p , 1 s 1 + c 1 ) 1 , 1 } ν 1 ρ ,

and u L L T 2 ( B ˙ p , 1 s ) μ 1 2 c 2 , θ L L T 2 ( B ˙ p , 1 n p ) ν 1 2 c 1 . Under this setting and the assumptions: θ ¯ k Y T c 1 , u ¯ k X T c 2 , it follows from (3.2) and (3.3) that

θ ¯ k + 1 Y T 4 C μ 1 2 ν 1 2 c 1 c 2 c 1

and

u ¯ k + 1 X T 4 C μ 1 c 2 2 + c 2 2 c 2 ,

Considering that θ ¯ 0 = 0 , u ¯ 0 = 0 , by means of iteration, we obtain the uniform boundedness of { ( θ ¯ k , u ¯ k ) } in X T × Y T .

Step 2. Convergence of the approximate solutions.

Let δ θ k + 1 = θ k + 1 θ k and δ u k + 1 = u k + 1 u k . Evidently, ( δ θ k + 1 , δ u k + 1 ) is the integral solution of the following problem:

(3.4) t δ θ k + 1 ν Δ δ θ k + 1 = div ( δ θ k u k ) div ( θ k 1 δ u k ) , t δ u k + 1 μ Δ δ u k + 1 = P div ( δ u k u k ) P div ( u k 1 δ u k ) + P ( δ θ k e n ) , δ θ k + 1 ( 0 ) = 0 , δ u k + 1 ( 0 ) = 0 .

Similar to (3.2) and (3.3), we can derive that

δ θ k + 1 Y T C 0 T div ( δ θ k ( τ ) u k ( τ ) ) B ˙ p , 1 s 1 d τ + C 0 T div ( θ k 1 ( τ ) δ u k ( τ ) ) B ˙ p , 1 s 1 d τ C ( δ θ k L T 2 ( B ˙ p , 1 s ) u k L T 2 ( B ˙ p , 1 n p ) + θ k 1 L T 2 ( B ˙ p , 1 s ) δ u k L T 2 ( B ˙ p , 1 n p ) ) C ν 1 2 δ θ k Y T ( u L L T 2 ( B ˙ p , 1 n p ) + μ 1 2 u ¯ k X T ) + C μ 1 2 δ u k X T ( θ L L T 2 ( B ˙ p , 1 s ) + ν 1 2 θ ¯ k 1 Y T ) 2 1 ( δ θ k Y T + δ u k X T )

and

δ u k + 1 X T C 0 T div ( δ u k ( τ ) u k ( τ ) ) B ˙ p , 1 n p 1 d τ + C 0 T div ( u k 1 ( τ ) δ u k ( τ ) ) B ˙ p , 1 n p 1 d τ + C 0 T δ θ k ( τ ) B ˙ p , 1 n p 1 d τ C δ u k L T 2 ( B ˙ p , 1 n p ) ( u k L T 2 ( B ˙ p , 1 n p ) + u k 1 L T 2 ( B ˙ p , 1 n p ) ) + C T 1 1 ρ δ θ k L ˜ T ( B ˙ p , 1 s 1 ) 1 1 ρ δ θ k L T 1 ( B ˙ p , 1 s + 1 ) 1 ρ C μ 1 2 δ u k X T [ 2 u L L T 2 ( B ˙ p , 1 n p ) + μ 1 2 ( u ¯ k X T + u ¯ k 1 X T ) ] + C T 1 1 ρ ν 1 ρ δ θ k Y T 2 1 ( δ θ k Y T + δ u k X T ) .

Summing up, we have

( δ u k + 1 , δ θ k + 1 ) X T × Y T 2 1 ( δ u k , δ θ k ) X T × Y T ,

which means that { ( θ ¯ k , u ¯ k ) } is a Cauchy sequence in X T × Y T . Thus, there exists ( θ ¯ , u ¯ ) X T × Y T such that θ ¯ k θ ¯ Y T 0 and u ¯ k u ¯ X T 0 as k . Let θ = θ L + θ ¯ and u = u L + u ¯ , then we have ( θ , u ) X T × Y T , and

div ( θ k u k ) div ( θ u ) L T 1 ( B ˙ p , 1 s 1 ) 0 , P div ( u k u k ) P div ( u u ) L T 1 ( B ˙ p , 1 n p 1 ) 0 and P θ k P θ L T 1 ( B ˙ p , 1 n p 1 ) 0 .

Consequently, ( θ , u ) is an integral solution of the system

(3.5) t θ ν Δ θ + div ( θ u ) = 0 , t u μ Δ u + P div ( u u ) = P ( θ e n ) , θ ( 0 ) = θ 0 , u ( 0 ) = u 0

on [ 0 , T ] .

Step 3. Regularity and uniqueness of the integral solution.

Define

F ( t ) = 0 t e ν ( t τ ) Δ div ( θ u ) d τ and G ( t ) = 0 t e μ ( t τ ) Δ P [ div ( u u ) θ e n ] d τ .

Since Δ θ L T 1 ( B ˙ p , 1 s 1 ) and Δ u L T 1 ( B ˙ p , 1 n p 1 ) , we have

Δ F ( t ) = Δ θ L ( t ) Δ θ ( t ) B ˙ p , 1 s 1 and Δ G ( t ) = Δ u L ( t ) Δ u ( t ) B ˙ p , 1 n p 1

a.e. on [ 0 , T ] . Noting that div ( θ u ) L T 1 ( B ˙ p , 1 s 1 ) and P [ div ( u u ) θ e n ] L T 1 ( B ˙ p , 1 n p 1 ) , by the closedness of Δ and analyticity of e ν t Δ and P e μ t Δ in B ˙ p , 1 s 1 and in Besov spaces (see for example [12,13]), we can assert that F ( t ) and G ( t ) are both differentiable in B ˙ p , 1 s 1 and B ˙ p , 1 n p 1 for a.e. t [ 0 , T ] , respectively, and

t F ( t ) = ν 0 t Δ e ν ( t τ ) Δ div ( θ u ) d τ div ( θ ( t ) u ( t ) ) = ν Δ F ( t ) div ( θ ( t ) u ( t ) ) in B ˙ p , 1 s 1 , t G ( t ) = μ 0 t Δ e μ ( t τ ) Δ P [ div ( u u ) θ e n ] d τ P [ div ( u ( t ) u ( t ) ) θ ( t ) e n ] = μ Δ G ( t ) P [ div ( u ( t ) u ( t ) ) θ ( t ) e n ] in B ˙ p , 1 n p 1 ,

which combined with the a.e. differentiability of θ L ( t ) and u L ( t ) leads to the a.e. differentiability of θ ( t ) and u ( t ) and the validity of (3.5). As a straight consequence, we have t θ L T 1 ( B ˙ p , 1 s 1 ) , t u L T 1 ( B ˙ p , 1 n p 1 ) , and there exists π L T 1 ( B ˙ p , 1 n p 1 ) such that

t u μ Δ u + div ( u u ) + π = θ e n in B ˙ p , 1 n p 1 for a.e. t [ 0 , T ] .

In a word, ( θ , u , π ) is the strong solution of the Boussinesq system (1.1).

Proof of uniqueness of the local strong solution is much similar to Step 2, and we omit it here.□

Remark 3.1

In light of the classical theory of abstract evolution equations, we know that every local solution ( θ , u ) of (1.1) can be extended rightward step by step onto a maximal interval [ 0 , T * ) for some 0 < T * to become a saturated solution such that ( θ , u ) ( C ( [ 0 , T * ) ; B ˙ p , 1 s 1 ) L loc 1 ( ( 0 , T * ) ; B ˙ p , 1 s + 1 ) ) × ( C ( [ 0 , T * ) ; B ˙ p , 1 n p 1 ) L loc 1 ( ( 0 , T * ) ; B ˙ p , 1 n p + 1 ) ) . We should mention that under the additional condition:

(3.6) ( θ , u ) ( L ( ( 0 , T * ) ; B ˙ p , 1 s 1 ) L 1 ( ( 0 , T * ) ; B ˙ p , 1 s + 1 ) ) × ( L ( ( 0 , T * ) ; B ˙ p , 1 n p 1 ) L 1 ( ( 0 , T * ) ; B ˙ p , 1 n p + 1 ) ) ,

it becomes T * = + , and ( θ , u ) exists globally. This can be proved by contraction as follows.

Assume that 0 < T * < + , then similar to step 3 of the previous proof, we find that Δ θ , div ( θ u ) L T * 1 ( B ˙ p , 1 s 1 ) and Δ u , P div ( u u ) , P ( θ e n ) L T * 1 ( B ˙ p , 1 n p 1 ) , consequently t θ L T * 1 ( B ˙ p , 1 s 1 ) and t u L T * 1 ( B ˙ p , 1 n p 1 ) . Thus, continuity of θ and u keeps in B ˙ p , 1 s 1 and B ˙ p , 1 n p 1 , respectively, up to the left side of T * . By defining θ ( T * ) = lim t ( T * ) θ ( t ) and u ( T * ) = lim t ( T * ) u ( t ) , we have θ C ( [ 0 , T * ] ; B ˙ p , 1 s 1 ) and u C ( [ 0 , T * ] ; B ˙ p , 1 n p 1 ) .

We next show that for each ( θ 0 , u 0 ) B ˙ p , 1 s 1 × B ˙ p , 1 n p 1 , existing time T of the local solution is uniform for the initial points in every small neighbourhood of ( θ 0 , u 0 ) . By reviewing the arguments in step 1 of the previous proof, it reduces to prove e ν t Δ θ 1 L T 1 ( B ˙ p , 1 s + 1 ) 0 and e μ t Δ u 1 L T 1 ( B ˙ p , 1 n p + 1 ) 0 as T 0 uniformly for all ( θ 1 , u 1 ) in ( θ 0 , u 0 ) ’s neighbourhood with small radius. We only deal with the former limit, and the latter one can be treated in the same way. Assume that θ 1 θ 0 B ˙ p , 1 s 1 < δ for some 0 < δ < 1 specified later. For each q Z , from the estimate for the heat semigroup (refer to [10, Ch.2, §2.1.2])

e ν t Δ Δ ˙ q θ 1 L p C e c ν t 2 2 q Δ ˙ q θ 1 L p ,

where the constants C , c > 0 are independent of t , q , and θ 1 , it follows that

e ν t Δ θ 1 L T 1 ( B ˙ p , 1 s + 1 ) C q Z 1 e c ν T 2 2 q c ν 2 q ( s 1 ) Δ ˙ q θ 1 L p C q k + q > k 1 e c ν T 2 2 q c ν 2 q s Δ ˙ q θ 1 L p C ( 1 e c ν T 2 2 k ) c ν θ 1 B ˙ p , 1 s 1 + C c ν θ 1 θ 0 B ˙ p , 1 s 1 + q > k 2 q ( s 1 ) Δ ˙ q θ 0 L p C ( 1 e c ν T 2 2 k ) c ν ( θ 0 B ˙ p , 1 s 1 + 1 ) + C c ν δ + q > k 2 q ( s 1 ) Δ ˙ q θ 0 L p .

For arbitrary ε > 0 , take k N so large that

q > k 2 q ( s 1 ) Δ ˙ q θ 0 L p < c ν ε 4 C ,

and take 0 < δ < ( 4 C ) 1 c ν ε . Thus, there exists T 0 > 0 such that for all 0 < T < T 0 and all θ 1 B ˙ p , 1 s 1 fulfilling θ 1 θ 0 B ˙ p , 1 s 1 < δ , it holds

1 e c ν T 2 2 k < c ν ε 2 C ( θ 0 B ˙ p , 1 s 1 + 1 ) ,

and consequently e ν t Δ θ 1 L T 1 ( B ˙ p , 1 s + 1 ) < ε , which leads to the desired conclusion.

Now for each t [ 0 , T * ] , by the continuity of ( θ ( t ) , u ( t ) ) in B ˙ p , 1 s 1 × B ˙ p , 1 n p 1 and uniqueness of the local strong solution, there are two lengths r , h > 0 such that for each t ( t r , t + r ) [ 0 , T * ] , the strong solution ( θ , u ) viewed as arising from the initial point ( θ ( t ) , u ( t ) ) can be extended rightwards onto the interval [ t , t + h ] . Thus, by the compactness of [ 0 , T * ] , we can find two lengths h r > 0 independent of t [ 0 , T * ] such that for all t [ T * 2 , T * ] , ( θ , u ) can be extended rightward uniformly from t r 2 to t + h , which implies that ( θ , u ) can be eventually extended beyond T * . This however contradicts the definition of the maximal interval of existence.

Proof of Theorem 2.5

Assume that [ 0 , T * ) is the maximal interval of existence for the strong solution ( θ , u ) to system (1.1). By Remark 3.1, to show T * = , it suffices to verify Condition (3.6). For this purpose, we make estimates for the approximate solutions at first.

Given a Banach space X , for each α > 0 and 0 < T < T * , define

C α ( ( 0 , T ] ; X ) = { f C ( ( 0 , T ] ; X ) : sup 0 < t T t α f ( t ) X < } .

Evidently, endowed with the norm

f C α ( ( 0 , T ] ; X ) = sup 0 < t T t α f ( t ) X ,

C α ( ( 0 , T ] ; X ) is also a Banach space. Using the estimates for heat semigroups, one can easily check that, under the initial assumption ( θ 0 , u 0 ) L n 3 × L σ n , it holds

θ L C ( [ 0 , T ] ; L n 3 ) C 1 2 ( ( 0 , T ] ; L n 2 ) , u L C ( [ 0 , T ] ; L n ) C 1 4 ( ( 0 , T ] ; L 2 n ) ,

and

θ L C ( [ 0 , T ] ; L n 3 ) + ν 1 2 θ L C 1 2 ( ( 0 , T ] ; L n 2 ) c 0 θ 0 L n 3 , u L C ( [ 0 , T ] ; L n ) + μ 1 4 u L C 1 4 ( ( 0 , T ] ; L 2 n ) d 0 u 0 L n ,

where the constants c 0 , d 0 > 0 depend only on n .

Assume that θ k C ( [ 0 , T ] ; L n 3 ) C 1 2 ( ( 0 , T ] ; L n 2 ) and u k C ( [ 0 , T ] ; L n ) C 1 4 ( ( 0 , T ] ; L 2 n ) , by using the following estimates (refer to [3]):

e t ν Δ div f L q C ( ν t ) n ( 1 p 1 q ) 2 1 2 f L p , e t μ Δ P div F L q C ( μ t ) n ( 1 p 1 q ) 2 1 2 F L p ,

where 1 p q < , f and F are vector and tensor fields, respectively, we can deduce that

(3.7) θ k + 1 ( t ) L n 3 θ L ( t ) L n 3 + 0 t e ( t τ ) ν Δ div ( θ k ( τ ) u k ( τ ) ) L n 3 d τ θ L ( t ) L n 3 + C ν 1 2 0 t ( t τ ) 1 2 θ k ( τ ) L n 2 u k ( τ ) L n d τ θ L C ( [ 0 , t ] ; L n 3 ) + C ν 1 2 θ k C 1 2 ( ( 0 , t ] ; L n 2 ) u k C ( [ 0 , t ] ; L n ) ,

(3.8) t 1 2 θ k + 1 ( t ) L n 2 t 1 2 θ L ( t ) L n 2 + C ν 3 4 t 1 2 0 t ( t τ ) 3 4 θ k ( τ ) L n 2 u k ( τ ) L 2 n d τ θ L C 1 2 ( ( 0 , t ] ; L n 2 ) + C ν 3 4 θ k C 1 2 ( ( 0 , t ] ; L n 2 ) u k C 1 4 ( ( 0 , t ] ; L 2 n ) ,

and

(3.9) u k + 1 ( t ) L n u L ( t ) L n + 0 t e ( t τ ) μ Δ div ( u k ( τ ) u k ( τ ) ) L n d τ + 0 t e ( t τ ) μ Δ P θ k ( τ ) e n L n d τ u L ( t ) L n + C μ 3 4 0 t ( t τ ) 3 4 u k ( τ ) L n u k ( τ ) L 2 n d τ + C μ 1 2 0 t ( t τ ) 1 2 θ k ( τ ) L n 2 d τ u L C ( [ 0 , t ] ; L n ) + C μ 3 4 u k C ( [ 0 , t ] ; L n ) u k C 1 4 ( ( 0 , t ] ; L 2 n ) + C 1 μ 1 2 θ k C 1 2 ( ( 0 , t ] ; L n 2 ) ,

(3.10) t 1 4 u k + 1 ( t ) L 2 n t 1 4 u L ( t ) L 2 n + C μ 3 4 t 1 4 0 t ( t τ ) 3 4 u k ( τ ) L 2 n 2 d τ + C μ 3 4 t 1 4 0 t ( t τ ) 3 4 θ k ( τ ) L n 2 d τ u L C 1 4 ( ( 0 , t ] ; L 2 n ) + C μ 3 4 u k C 1 4 ( ( 0 , t ] ; L 2 n ) 2 + C 1 μ 3 4 θ k C 1 2 ( ( 0 , t ] ; L n 2 ) ,

where the constant C 1 > 0 depends only on n . Meanwhile, continuity of θ k + 1 , u k + 1 in L n 3 , L n 2 and L n , L 2 n , respectively, can be checked directly.

Let W T = ( C ( [ 0 , T ] ; L n 3 ) C 1 2 ( ( 0 , T ] ; L n 2 ) ) × ( C ( [ 0 , T ] ; L n ) C 1 4 ( ( 0 , T ] ; L 2 n ) ) with the norm

( θ , u ) W T 4 C 1 min { μ , ν } 1 ( θ C ( [ 0 , T ] ; L n 3 ) + ν 1 2 θ C 1 2 ( ( 0 , T ] ; L n 2 ) ) + ( u C ( [ 0 , T ] ; L n ) + μ 1 4 u C 1 4 ( ( 0 , T ] ; L 2 n ) ) .

Then putting (3.7), (3.8), (3.9), and (3.10) together, we have

(3.11) ( θ k + 1 , u k + 1 ) W T ( θ L , u L ) W T + C min { μ , ν } 1 ( θ k , u k ) W T 2 + 2 1 ( θ k , u k ) W T k 0 + k 1 ( θ k , u k ) W T 2 + 2 1 ( θ k , u k ) W T ,

where k 0 = 4 C 1 c 0 min { μ , ν } 1 θ 0 L n 3 + d 0 u 0 L n and k 1 = C min { μ , ν } 1 .

Take 0 < ε ( 64 C max { 4 C 1 c 0 , d 0 } ) 1 , and let

θ 0 L n 3 min { μ , ν } 2 ε , u 0 L n min { μ , ν } ε .

Under this setting, we have k 0 ( 32 k 1 ) 1 . So if we take λ 1 = ( 1 1 16 k 0 k 1 ) 4 k 1 and assume ( θ k , u k ) W T λ 1 , then from (3.11), we can derive ( θ k + 1 , u k + 1 ) W T λ 1 . Noting that 2 k 0 λ 1 4 k 0 , we have ( θ L , u L ) W T k 0 λ 1 2 . Hence by means of induction and letting ( θ 0 , u 0 ) = ( θ L , u L ) , we conclude that

(3.12) ( θ k , u k ) W T λ 1 4 k 0

for all k N .

Making the same arguments on (3.4), we can derive the following estimates:

( δ θ k + 1 , δ u k + 1 ) W T k 1 [ ( θ k , u k ) W T + ( θ k 1 , u k 1 ) W T ] ( δ θ k , δ u k ) W T + 2 1 ( δ θ k , δ u k ) W T ( 8 k 0 k 1 + 2 1 ) ( δ θ k , δ u k ) W T 3 4 ( δ θ k , δ u k ) W T , k = 1 , 2 , ,

which means that { ( θ k , u k ) } is a Cauchy sequence in W T . Consequently, it converges to ( θ , u ) in W T , and ( θ , u ) verifies (3.12) for each 0 < T < T * . By the arbitrariness of T ( 0 , T * ) , we assert that

(3.13) sup 0 t < T * u ( t ) L n 4 k 0 C 2 ε min { μ , ν } ,

where the constant C > 0 depends only on n . Since L n B ˙ , 1 , from (3.13), we obtain immediately

sup 0 t < T * u ( t ) B ˙ , 1 C 3 ε min { μ , ν }

for some C 3 > 0 depending only on n .

Now we can investigate global existence of the strong solution in Besov spaces. We begin with the general case n p < and n p < s n p .

Take 1 < σ < min { n p 1 , s } . To deal with the term div ( θ u ) in L ˜ t 1 ( B ˙ p , σ 1 ) , we first consider Bony’s decomposition θ u = T ˙ θ u + T ˙ u θ + R ˙ ( θ , u ) , where T ˙ θ u = q Z S ˙ q 1 θ Δ ˙ q u , T ˙ u θ = q Z S ˙ q 1 u Δ ˙ q θ , and R ˙ ( θ , u ) = q Z Δ ˙ q θ Δ ˜ ˙ q u , where Δ ˜ ˙ q = Δ ˙ q 1 + Δ ˙ q + Δ ˙ q + 1 (cf. [10, §2.6]). Since sup q Z 2 q S ˙ q u L is equivalent to the norm u B ˙ , 1 , for each 0 < t < T * and q Z , it holds that

(3.14) 2 q σ Δ ˙ q T ˙ u θ L t 1 ( L p ) C 2 q σ q q 4 S ˙ q 1 u L t ( L ) Δ ˙ q θ L t 1 ( L p ) C u L t ( B ˙ , 1 ) θ L ˜ t 1 ( B ˙ p , σ + 1 ) ,

(3.15) 2 q σ Δ ˙ q T ˙ θ u L t 1 ( L p ) C 2 q ( σ + 1 n p ) q q 4 q q 2 Δ ˙ q θ L t 1 ( L ) Δ ˙ q u L t ( L n ) C 2 q ( σ + 1 n p ) u L t ( L n ) q q 4 q q 2 2 q n p Δ ˙ q θ L t 1 ( L p ) C θ L ˜ t 1 ( B ˙ p , σ + 1 ) u L t ( L n ) q q 4 q q 2 2 ( q q ) ( σ + 1 n p ) ,

and

(3.16) 2 q σ Δ ˙ q R ˙ ( θ , u ) L t 1 ( L p ) C 2 q ( σ + 1 ) q q 3 Δ ˙ q θ L t 1 ( L p ) Δ ˜ ˙ q u L t ( L n ) C θ L ˜ t 1 ( B ˙ p , σ + 1 ) u L t ( L n ) q q 3 2 ( q q ) ( σ + 1 ) .

Putting them together, we have

div ( θ u ) L ˜ t 1 ( B ˙ p , σ 1 ) C θ u L ˜ t 1 ( B ˙ p , σ ) C θ L ˜ t 1 ( B ˙ p , σ + 1 ) ( u L t ( L n ) + u L t ( B ˙ , 1 ) ) C ( C 2 + C 3 ) ν ε θ L ˜ t 1 ( B ˙ p , σ + 1 ) .

Using this estimate, with application of Lemma 2.2 to the first equation of (1.1), we can deduce from the assumption 0 < ε ( 2 C ( C 2 + C 3 ) ) 1 that θ L t ( B ˙ p , σ 1 ) L ˜ t 1 ( B ˙ p , σ + 1 ) , and

(3.17) θ L t ( B ˙ p , σ 1 ) + ν 2 θ L ˜ t 1 ( B ˙ p , σ + 1 ) C θ 0 B ˙ p , σ 1 C θ 0 B ˙ p , s 1 1 δ θ 0 B ˙ p , n p 3 δ C ( θ 0 B ˙ p , 1 s 1 + θ 0 L n 3 ) ,

where δ = ( σ + 2 n p ) ( s + 2 n p ) 1 .

Moreover, by performing the operators P and Q = I P , respectively, on both sides of the second equation of (1.1), we obtain

(3.18) t u μ Δ u = P ( u u ) + P ( θ e n )

and

π = Q ( u u ) + Q ( θ e n ) .

Applying Lemma 2.2 to (3.18), we have

u ( t ) B ˙ p , 1 n p 1 + μ 0 t u ( τ ) B ˙ p , 1 n p + 1 d τ C u 0 B ˙ p , 1 n p 1 + 0 t P ( u u ) B ˙ p , 1 n p 1 d τ + 0 t P ( θ e n ) B ˙ p , 1 n p 1 d τ .

By invoking [1, Lemma 4.3], we obtain

0 t P ( u u ) B ˙ p , 1 n p 1 d τ + 0 t Q ( u u ) B ˙ p , 1 n p 1 d τ C 0 t u ( τ ) B ˙ , 1 u ( τ ) B ˙ p , 1 n p + 1 d τ C C 3 ε μ 0 t u ( τ ) B ˙ p , 1 n p + 1 d τ .

So if we take 0 < ε ( 2 C C 3 ) 1 and use the boundedness of P , Q on B ˙ p , 1 n p 1 , with application of logarithmic interpolation inequality (cf. [14]), then we can deduce that

(3.19) u ( t ) B ˙ p , 1 n p 1 + μ 2 0 t u ( τ ) B ˙ p , 1 n p + 1 d τ + 0 t π ( τ ) B ˙ p , 1 n p 1 d τ C u 0 B ˙ p , 1 n p 1 + 0 t θ ( τ ) B ˙ p , 1 n p 1 d τ C u 0 B ˙ p , 1 n p 1 + θ L ˜ t 1 ( B ˙ p , n p 1 ) log e + θ L ˜ t 1 ( B ˙ p , 2 n p σ 3 ) + θ L ˜ t 1 ( B ˙ p , σ + 1 ) θ L ˜ t 1 ( B ˙ p , n p 1 ) .

Here two terms remain untreated. One is θ L ˜ t 1 ( B ˙ p , n p 1 ) , the other is θ L ˜ t 1 ( B ˙ p , 2 n p σ 3 ) . From (3.17), it follows that

(3.20) θ L ˜ t 1 ( B ˙ p , n p 1 ) t 1 1 β θ L t β ( B ˙ p , n p 1 ) t 1 1 β θ L t ( B ˙ p , σ 1 ) 1 1 β θ L ˜ t 1 ( B ˙ p , σ + 1 ) 1 β ( 1 1 β ) t θ L t ( B ˙ p , σ 1 ) + ( 1 β ) θ L ˜ t 1 ( B ˙ p , σ + 1 ) C ( t + ν 1 ) ( θ 0 B ˙ p , s 1 + θ 0 L n 3 )

and

(3.21) θ L ˜ t 1 ( B ˙ p , 2 n p σ 3 ) + θ L ˜ t 1 ( B ˙ p , σ + 1 ) t 1 1 γ θ L t γ ( B ˙ p , 2 n p σ 3 ) + θ L ˜ t 1 ( B ˙ p , σ + 1 ) t 1 1 γ θ L t ( B ˙ p , σ 1 ) 1 1 γ θ L ˜ t 1 ( B ˙ p , σ + 1 ) 1 γ + θ L ˜ t 1 ( B ˙ p , σ + 1 ) ( 1 1 γ ) t θ L t ( B ˙ p , σ 1 ) + ( 1 + 1 γ ) θ L ˜ t 1 ( B ˙ p , σ + 1 ) C ( t + ν 1 ) ( θ 0 B ˙ p , 1 s 1 + θ 0 L n 3 ) ,

where β = 2 ( n p σ ) 1 and γ = ( n p σ 1 ) 1 . Since for any M > 0 , r r log ( e + M r ) is an increasing function on ( 0 , ) , plugging (3.20) and (3.21) into (3.19), we can derive that

(3.22) u ( t ) B ˙ p , 1 n p 1 + μ 2 0 t u ( τ ) B ˙ p , 1 n p + 1 d τ + 0 t π ( τ ) B ˙ p , 1 n p 1 d τ C [ u 0 B ˙ p , 1 n p 1 + ( t + ν 1 ) ( θ 0 B ˙ p , 1 s 1 + θ 0 L n 3 ) log ( e + C ) ] C [ u 0 B ˙ p , 1 n p 1 + ( t + ν 1 ) ( θ 0 B ˙ p , 1 s 1 + θ 0 L n 3 ) ] .

Let us return to the first equation of (1.1). By invoking Lemma 2.3, we can deduce the following estimate:

(3.23) θ L ˜ t ( B ˙ p , 1 s 1 ) + ν θ L t 1 ( B ˙ p , 1 s + 1 ) C θ 0 B ˙ p , 1 s 1 exp { C u L t 1 ( B ˙ p , 1 n p + 1 ) } C θ 0 B ˙ p , 1 s 1 exp { C μ 1 [ u 0 B ˙ p , 1 n p 1 + ( t + ν 1 ) ( θ 0 B ˙ p , 1 s 1 + θ 0 L n 3 ) ] } .

Noting that estimates (3.22) and (3.23) both hold on ( 0 , T * ) . Hence, if we assume T * < , then from (3.22) and (3.23), we obtain (3.6) easily, which just leads to the contradiction as shown in Remark 3.1. This proves global existence of the strong solution. Besides, from (3.17), (3.23), and (3.22), we reach the desired estimates (2.8), (2.9), and (2.10), respectively. Finally, uniqueness of the global strong solution comes from uniqueness of the local strong solution and connectedness of the interval [ 0 , ) .

In the remaining part of the proof, we will derive (2.11) and (2.12) in the special case n p < 2 n and n p < s < n p 1 . We also begin with the inequality div ( θ u ) B ˙ p , 1 s 1 C θ u B ˙ p , 1 s and Bony’s decomposition. Noting that in the case s + 1 n p < 0 , it follows that

(3.24) 2 q s Δ ˙ q T ˙ θ u L p C 2 q s q q 4 q q 2 Δ ˙ q θ L Δ ˙ q u L p C 2 q s q q 4 q q 2 2 q n p Δ ˙ q θ L p 2 q ( 1 n p ) Δ ˙ q u L n C θ B ˙ p , 1 s + 1 u L n q q 4 2 ( q q ) s q q 2 2 ( q q ) ( s + 1 n p ) a q ,

where ( a q ) l 1 1 . Additionally, similar to (3.14), we have

(3.25) 2 q s Δ ˙ q T ˙ u θ L p C 2 q s q q 4 S ˙ q 1 u L Δ ˙ q θ L p C θ B ˙ p , 1 s + 1 u B ˙ , 1 q q 4 2 ( q q ) s a q .

Furthermore, since 1 n + 1 p < 1 and s + 1 > 0 , it holds that

(3.26) 2 q s Δ ˙ q R ˙ ( θ , u ) L p C 2 q ( s + 1 ) q q 3 Δ ˙ q θ L p Δ ˜ ˙ q u L n C θ B ˙ p , 1 s + 1 u L n q q 3 2 ( q q ) ( s + 1 ) a q .

Putting these estimates together, we obtain

0 t div ( θ ( τ ) u ( τ ) ) B ˙ p , 1 s 1 d τ C 0 t θ ( τ ) B ˙ p , 1 s + 1 ( u ( τ ) B ˙ , 1 + u ( τ ) L n ) d τ C ( C 2 + C 3 ) ε ν 0 t θ ( τ ) B ˙ p , 1 s + 1 d τ .

Now taking 0 < ε ( 2 C ( C 2 + C 3 ) ) 1 , and applying Lemma 2.2 to the first equation of (1.1), we have

(3.27) θ ( t ) B ˙ p , 1 s 1 + ν 2 0 t θ ( τ ) B ˙ p , 1 s + 1 C θ 0 B ˙ p , 1 s 1 .

Plugging (3.27) into (3.19), we obtain

(3.28) u ( t ) B ˙ p , 1 n p 1 + μ 2 0 t u ( τ ) B ˙ p , 1 n p + 1 d τ + 0 t π ( τ ) B ˙ p , 1 n p 1 d τ C u 0 B ˙ p , 1 n p 1 + 0 t θ ( τ ) B ˙ p , 1 n p 1 d τ C ( u 0 B ˙ p , 1 n p 1 + C t 1 1 ρ θ L t ρ ( B ˙ p , 1 n p 1 ) ) C ( u 0 B ˙ p , 1 n p 1 + C t 1 1 ρ θ L t ( B ˙ p , 1 s 1 ) 1 1 ρ θ L t 1 ( B ˙ p , 1 s + 1 ) 1 ρ ) C ( u 0 B ˙ p , 1 n p 1 + C t 1 1 ρ ν 1 ρ θ 0 B ˙ p , 1 s 1 ) ,

where ρ = 2 ( n p s ) 1 as before.

Finally, estimates (2.11) and (2.12) for ( θ , u , π ) come directly from (3.27) and (3.28), respectively. Thus, the proof has been completed.□

Remark 3.2

Because of the existence of the diffusion term, the first equation of (1.1) exhibits different mechanical characteristics to the transport equation (please compare it with [1,5]). So Lemmas 2.12.3 could not be applied directly to make a time-independent estimate in Besov space for the strong solution. To overcome the obstacle, we first derive a time-independent L n 3 L n estimate for the pair ( θ , u ) . By putting u L t ( L n ) in Bony’s decomposition, we then make an auxiliary estimate for θ in L ( ( 0 , ) ; B ˙ p , σ 1 ) L ˜ 1 ( ( 0 , ) ; B ˙ p , σ + 1 ) for the lower index σ , based on which, with the application of Lemma 2.2, 2.3, and logarithmic interpolation inequality, desired estimates for ( θ , u ) are eventually derived.

Acknowledgement

The authors would like to thank the referees for their valuable comments and suggestions.

  1. Funding information: This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

  2. Author contributions: The first part of the manuscript, including introduction and preliminaries, was written by Lu Wang and Qinghua Zhang. The second part, including main results and proofs, was written by Shuokai Yan and Qinghua Zhang. All authors reviewed and approved the manuscript.

  3. Conflict of interest: The authors declare that there are no conflicts of interest regarding the publication of this article.

References

[1] R. Danchin and M. Paicu, Les théorèmes de Leray et de Fujita-Kato pour le système de Boussinesq partiellement visqueux, Bull. Soc. Math. France 136 (2008), no. 2, 261–309, DOI: https://doi.org/10.24033/bsmf.2557. 10.24033/bsmf.2557Search in Google Scholar

[2] J. R. Cannon and E. DiBenedetto, The initial problem for the Boussinesq equations with data in Lp, Lectures Notes in Mathematics, Vol. 771, Springer, Berlin Heidelberg, 1980, pp. 129–144. 10.1007/BFb0086903Search in Google Scholar

[3] L. Brandolese and M. E. Schonbek, Large time decay and growth for solutions of a viscous Boussinesq system, Trans. Amer. Math. Soc. 364 (2012), 5057–5090, DOI: https://doi.org/10.1090/S0002-9947-2012-05432-8. 10.1090/S0002-9947-2012-05432-8Search in Google Scholar

[4] P. Han, Algebraic L2 decay for weak solutions of a viscous Boussinesq system in exterior domains, J. Differential Equations 252 (2012), 6306–6323, DOI: https://doi.org/10.1016/j.jde.2012.03.002. 10.1016/j.jde.2012.03.002Search in Google Scholar

[5] R. Danchin and M. Paicu, Existence and uniqueness results for the Boussinesq system with data in Lorentz spaces, Phys. D 237 (2008), no. 10–12, 1444–1460, DOI: https://doi.org/10.1016/j.physd.2008.03.034. 10.1016/j.physd.2008.03.034Search in Google Scholar

[6] R. A. Adams and J. J. F. Fournier, Sobolev Spaces, 2nd ed., Pure and Applied Mathematics, Vol. 140, Academic Press, Amsterdam, 2003. Search in Google Scholar

[7] H. Fujita and T. Kato, On the Navier-Stokes initial problem I, Arch. Ration. Mech. Anal. 16 (1964), 269–315, DOI: https://doi.org/10.1007/bf00276188. 10.1007/BF00276188Search in Google Scholar

[8] T. Kato, Strong Lp-solutions of the Navier-Stokes equation in Rm, with applications to weak solutions, Math. Z. 187 (1984), 471–480, DOI: https://doi.org/10.1007/BF01174182. 10.1007/BF01174182Search in Google Scholar

[9] H. Kozono, Global Ln-solution and its decay property for the Navier-Stokes equations in half-space R+n, J. Differential Equations 79 (1989), 79–88, DOI: https://doi.org/10.1016/0022-0396(89)90114-9. 10.1016/0022-0396(89)90114-9Search in Google Scholar

[10] H. Bahouri, J.-Y. Chemin, and R. Danchin, Fourier Analysis and Nonlinear Partial Differential Equations, Grundlehren der mathematischen Wissenschaften, Vol. 343, Springer-Verlag, Heidelberg, 2011. 10.1007/978-3-642-16830-7Search in Google Scholar

[11] H. Abidi and M. Paicu, Existence globale pour un fluide inhomogène, Ann. Inst. Fourier (Grenoble) 57 (2007), 883–917, DOI: https://doi.org/10.5802/aif.2280. 10.5802/aif.2280Search in Google Scholar

[12] H.-O. Bae, Analyticity for the Stokes operator in Besov spaces, J. Korean Math. Soc. 40 (2003), no. 6, 1061–1074, DOI: https://doi.org/10.4134/JKMS.2003.40.6.1061. 10.4134/JKMS.2003.40.6.1061Search in Google Scholar

[13] Z.-M. Chen, Analytic semigroup approach to generalized Navier-Stokes flows in Besov spaces, J. Math. Fluid Mech. 19 (2017), 709–724, DOI: https://doi.org/10.1007/s00021-016-0302-5. 10.1007/s00021-016-0302-5Search in Google Scholar

[14] R. Danchin, Density-dependent imcompressible viscous fluids in critical spaces, Proc. Roy. Soc. Edinburgh Sect. A 113 (2003), no. 6, 1311–1334, DOI: https://doi.org/10.1017/s030821050000295x. 10.1017/S030821050000295XSearch in Google Scholar

Received: 2023-10-27
Revised: 2024-01-20
Accepted: 2024-02-01
Published Online: 2024-03-14

© 2024 the author(s), published by De Gruyter

This work is licensed under the Creative Commons Attribution 4.0 International License.

Downloaded on 6.5.2024 from https://www.degruyter.com/document/doi/10.1515/math-2023-0182/html
Scroll to top button