Large time behaviour for the motion of a solid in a viscous incompressible fluid

In this article, we study the long-time behaviour of a system describing the coupled motion of a rigid body and of a viscous incompressible fluid in which the rigid body is contained. We assume that the system formed by the rigid body and the fluid fills the entire space R3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathbb {R}}^3$$\end{document}. In the case in which the rigid body is a ball, we prove the local existence of mild solutions and, when the initial data are small, the global existence of solutions for this system with a precise description of their large time behavior. Our main result asserts, in particular, that if the initial datum is small enough in suitable norms then the position of the center of the rigid ball converges to some h∞∈R3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h_\infty \in {\mathbb {R}}^3$$\end{document} as time goes to infinity. This result contrasts with those known for the analogues of our system in 2 or 1 space dimensions, where it has been proved that the body quits any bounded set, provided that we wait long enough. To achieve this result, we use a “monolithic” type approach, which means that we consider a linearized problem in which the equations of the solid and of the fluid are still coupled. An essential role is played by the properties of the semigroup, called fluid-structure semigroup, associated to this coupled linearized problem. The generator of this semigroup is called the fluid-structure operator. Our main tools are new Lp-Lq\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L^p - L^q$$\end{document} estimates for the fluid-structure semigroup. Note that these estimates are proved for bodies of arbitrary shape. The main ingredients used to study the fluid-structure semigroup and its generator are resolvent estimates which provide both the analyticity of the fluid-structure semigroup (in the spirit of a classical work of Borchers and Sohr) and Lp-Lq\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L^p- L^q$$\end{document} decay estimates (by adapting a strategy due to Iwashita).

estimates which provide both the analyticity of the fluid-structure semigroup (in the spirit of a classical work of Borchers and Sohr) and L p − L q decay estimates (by adapting a strategy due to Iwashita).

Introduction
We consider a homogeneous rigid body which occupies at instant t = 0 a ball B of radius R > 0 and centered at the origin and we study the motion of this body in a viscous incompressible fluid which fills the remaining part of R 3 . We denote by h(t), S(t), F(t) the position of the centre of the ball, the domain occupied by the solid, which coincides with the ball of radius R centered at h, and the domain filled by the fluid, respectively, at instant t > 0. Moreover, the velocity and pressure fields in the fluid are denoted by u and p, respectively. With the above notation, the system describing the motion of the rigid ball in the fluid is , (y − h(t)) × σ (u, p)ν ds (t > 0), u(0, y) = u 0 (y) ( y ∈ F(0)), h(0) = 0,ḣ(0) = 0 , ω(0) = ω 0 . (1.1) In the above equations, ω(t) represents the angular velocity of the ball (with respect to its centre) and the fluid is supposed to be homogeneous with density equal to 1 and of constant viscosity μ > 0. Moreover, the unit vector field normal to ∂ S(t) and directed towards the interior of S(t) is denoted by ν(t, ·). The constant m > 0 and the matrix J stand for the mass and the inertia tensor of the rigid body. Since in the above equations the rigid body is a homogeneous ball of radius R, the inertia tensor is independent of time and Finally, the Cauchy stress tensor field in the fluid is given by the constitutive law where δ k stands for the Kronecker symbol.
The system (1.1) can be easily transformed into a system in which the fluid equation is written in a fixed spatial domain. Indeed, using the change of frame x → y(t, x) := x + h(t) and setting v 0 (x) = u 0 (x), v(t, x) = u(t, x + h(t)), π(t, x) = p(t, x + h(t)), and E := F(0) = R 3 \B, Eq. (1.1) can be written in the form of the following system of unknowns v, π , and ω: As far as we know, the initial and boundary value problem (1.2) has been first studied in Serre [24], where it is proved, in particular, that (1.2) admits global in time weak solutions (of Leray type). The existence and uniqueness of strong solutions, with initial velocity supposed to be small (in the Sobolev space W 1,2 ) has been first established in Cumsille and Takahashi [4]. For the L p theory for the local in time existence and uniqueness of strong solutions of (1.2), we refer to Geissert et al. [9]. Let us also mention that the analogue of (1.2) when the fluid-rigid body system fills a bounded cavity (instead of the whole R 3 ) has also been studied in a quite important number of papers (see, for instance, Maity and Tucsnak [20] and references therein).
A natural question when considering (1.2) is the large time behaviour of the position of the mass centre of the ball, i.e., of the function h defined by It is, in particular, important to establish whether the centre of the rigid ball stabilizes around some position in R 3 or its distance to the origin tends to infinity when t → ∞. As far as we know, this question is open in the three dimensional context of (1.2). However, if one replaces the rigid ball by an infinite cylinder (so that the fluid can be modeled by the Navier-Stokes equations in two space dimensions) the question is studied in Ervedoza et al. [6], where it is established that the norm of (t) behaves like like Vázquez and Zuazua [29], or one dimensional viscous compressible fluids, like Koike [17].
The main novelty brought in by our work is twofold. Firstly, we prove that (1.2) is well-posed (globally in time) for initial data which are small in appropriate L q type spaces. Secondly, by appropriately choosing q, we prove that there exists h ∞ ∈ R 3 such that lim t→∞ h(t) = h ∞ , i.e., that the rigid body "stops" as t → ∞.
To state our main result we first recall that if G ⊂ R 3 is an open set, q > 1 and s ∈ R, the notation L q (G) and W s,q (G) stands for the standard Lebesgue and Sobolev-Slobodeckij spaces, respectively. Our main result can be stated as follows: Theorem 1.1 With the above notation for the set E. There exists ε 0 > 0 such that for every v 0 ∈ L 3 (E) 3 and 0 , ω 0 ∈ R 3 with there exists a unique solution (v, , ω) of (1. (1.5) Moreover, for q ∈ (1,3], there exists ε 0 (q) ∈ (0, ε 0 ] such that if v 0 ∈ L q (E) 3 ∩ then, for every p ∈ max 3 2 , q , ∞ the solution (v, , ω) of (1.2) satisfies In particular, if q < 3/2, taking p = ∞ in (1. 6) we have that ∈ L 1 ([0, ∞); R 3 ), hence that the position of the centre of the moving rigid ball converges to some point at finite distance h ∞ ∈ R 3 as t → ∞.

Remark 1.2
In fact, one can prove that, for every v 0 ∈ L 3 (E) 3  As precisely stated in Theorem 8.2 below, the results in Theorem 1.1 can be completed to include a local in time existence result without any smallness assumption on the initial data, see Sect. 8 for more precise statements.
The proof of Theorem 1.1 is based on decay estimates for the solutions of the linearized version of (1.2). Therefore, an important part of this work is devoted to the study of the semigroup associated to the linearized problem. As shown in the forthcoming sections, this semigroup called the fluid-structure semigroup, and its generator (called the fluid-structure operator) share several important properties of the Stokes semigroup and Stokes operator in an exterior domain. To establish this fact, an essential step consists in proving that the resolvent estimates derived in Iwashita [15] and Giga-Sohr [10] for the Stokes operator also hold for the fluid-structure operator (see also the corresponding estimates for the non-autonomous system describing the Navier-Stokes flow around a rotating obstacle, which have been obtained in Hishida [13,14]). Our results on the linearized problem will be derived for a solid of arbitrary shape, opening the way to a generalization of Theorem 1.1 for solids of arbitrary shape. However, the fixed point methodology used in the present paper to pass from the linearized equations to the full nonlinear problem is strongly using the fact that the rigid body is a ball (see the comments in Sect. 9 below concerning some tracks towards the modification of this procedure for tackling rigid bodies of arbitrary shape).
Note that Theorem 1.1 refers to mild solutions of (1.2), i.e., satisfying the integral equation is the fluid-structure semigroup and P is a Leray type projector on the space of free divergence vector fields on R 3 which coincide with a rigid velocity field on B. A precise definition of these objects requires some preparation and notation, so it is postponed to Sect. 3. However, we mention here that the roles of the projector P and of the fluid-structure semigroup in this paper are very close to those played by the Leray projector and the Stokes semigroup in the analysis of the Navier-Stokes equations. Consequently, the construction and study of the fluid-structure semigroup and of its generator are essential steps of our analysis, which are detailed in Sects. 4-7.
The outline of the paper is as follows. In Sect. 2, we introduce the notation (in particular several function spaces) that will be used throughout the article and we recall several results on the Stokes system in exterior domains. In Sect. 3 we introduce the fluid-structure operator and we give some of its basic properties. Section 4 is devoted to resolvent estimates for the fluid-structure operator. We use existing results on the Stokes system in exterior domains to derive our results. In Sects. 5 and 6 we show that the fluid-structure operator generates a bounded analytic semigroup on a suitable Banach space. We prove, in particular, L p − L q decay estimates for the fluidstructure semigroup in Sect. 7. Section 8 is devoted to the proof of Theorem 1.1. In Sect. 9, we formulate some open problems. Some technical results are collected in Appendix A and Appendix B.

Notation and preliminaries
Throughout this paper, the notation N, Z, R, C stands for the sets of natural numbers (starting with 1), integers, real numbers and complex numbers, respectively. For n ∈ N, the euclidian norm on C n will be simply denoted by | · |. For θ ∈ (0, π) we define the sector θ in the complex plane by θ = {λ ∈ C\{0} | |argλ| < θ}.
(2.1) Moreover, Z + stands for N ∪ {0}. For n, m ∈ N, u : R n → R m and α ∈ Z n + we set |α| = n k=1 α k and we use the notation ∂ α u for the partial derivative ∂ |α| u If G ⊂ R 3 is an open set, q > 1 and k ∈ N, we denote the standard Lebesgue and Sobolev spaces by L q (G) and by W k,q (G), respectively. For s ∈ R, W s,q (G) denotes the Sobolev-Slobodeckij spaces. The norms on [L q (G)] n and W k,q (G) n with n ∈ N, will be denoted by · q,G and · k,q,G , respectively. When G = R 3 , these norms will be simply denoted by · q and · k,q , respectively. Moreover, the space W k,q 0 (G) is the completion of C ∞ 0 (G) with respect to the W k,q (G) norm. We use repeatedly below the following well known result due to Bogovskiȋ [1]: Lemma 2.1 Let G be a smooth bounded domain in R 3 , q ∈ (1, ∞) and k ∈ Z + and let

Then there exists a linear bounded operator
We also introduce the homogeneous Sobolev spaces with the norm where we identify elements differing by a constant. Moreover, the function space will often appear in the remaining part of this work. For k ∈ N, and s, q ∈ R with 1 < q < ∞, we define the weighted Sobolev spaces W k,q,s (G) by and we set L q,s (G) = W 0,q,s (G). For ϕ ∈ [W 1,q (G)] 3 we denote by D(ϕ) the associated strain field defined by To end this section, we recall several results due to Borchers and Sohr [2] and Iwashita [15], on the Stokes system in the exterior domain is an open bounded set with ∂O of class C 2 . More precisely, we consider the stationary Stokes problem: (2.5) By combining Theorem 1.2 in [2] and Corollary 3.2 in [15] we have: Theorem 2.2 Let θ ∈ π 2 , π and let θ be the set defined in (2.1). Then

1.
Then there exist two families of operators (R(λ)) λ∈ θ and (P(λ)) λ∈ θ such that for every λ ∈ θ we have and the functions v = R(λ) f and p = P(λ) f satisfy (2.5). Moreover, there exists a positive constant M, depending only on O, q and θ such that for every λ ∈ θ we have Moreover, the functions λ → R(λ) and λ → P(λ) are holomorphic from θ to L W m,q,s (E) 3 , W m+2,q,s (E) 3 and L W m,q,s (E) 3 , W m+1,q,s (E) , respectively. Finally, there exist 3 Some background on the fluid-structure operator

Definition and first properties
In this section, we introduce the fluid-structure operator and the fluid-structure semigroup and we remind some of their properties, as established in the existing literature.
For the remaining part of this section the notation designs either an open, connected and bounded subset of R 3 , with ∂ of class C 2 , or we have = R 3 . Let O be an open bounded set with smooth boundary such that O ⊂ and such that 0 is its center of mass. We denote E = \O and we set E R 3 := E. Moreover, we denote by ν the unit normal vector on ∂O oriented towards the interior of O.
Reminding notation (2.4) for the tensor field D, we introduce the function space associated to the sets and O, which plays an important role in this work. Note that, for every q ∈ (1, ∞) the dual (X q ( )) * of X q ( ) can be identified with X q ( ), where ρ is the constant density of the rigid body. Our notation is making explicit only the dependence of X q on since these spaces will be used later on for various and with fixed O. For = R 3 , we simply set Since every in X q ( ) satisfies D( ) = 0 in O, there exist a unique couple ω ∈ where 1 U stands for the characteristic function of the set U (see for instance [27,Lemma 1.1]). We can thus use the identification: The two results below will allow us to precisely introduce the projection operator P q, from L q ( ) 3 onto X q ( ) which will be used in the following, and which will be denoted by P q when = R 3 .
Then for every u ∈ L q (R 3 ) 3 there exists a unique triple The map u → v, denoted P q , is a projection operator form L q ( ) 3 onto X q ( ).
Moreover, the dual of the operator P q is P q , where 1 q For the proof of Proposition 3.1 we refer to Wang Then for every u ∈ L q ( ) 3 there exists a unique triple (v, w 1 , The map u → v, denoted P q, , is a projection operator form L q ( ) 3 onto X q ( ).
Furthermore, the dual of the operator P q, is P q , , where 1 q The proof of Proposition 3.2 is similar to the proof of [30,Theorem 2.2]. However, for the sake of completeness we provide a short proof in Appendix A.
We also need some density results. Let us define As before, for = R 3 , we simply set Using Propositions 3.1 and 3.2, we have the following result The proof of this proposition is similar to [7, Theorem 2] and [23, Theorem 1.6]. We provide a short proof in Appendix A.
The fluid-structure operator on is the operator A q, : D(A q, ) → X q ( ) defined, for every q > 1, by where P q, is the projector introduced in Proposition 3.1, and the operator A q, : where m and J are given in terms of the constant density ρ of the body by (3.11) Note that the tensor of inertia J is positive. Also note that in the following, the density ρ of the homogeneous body will not intervene anymore directly: it will only appear through the constants m and J defined above.
In the case = R 3 , the operators P q, , A q, and A q, are denoted by P q , A q and A q , respectively and A q : D(A q ) → X q is defined, for every q > 1, by In the case q = 2 and when O is a ball, the fluid-structure operator A q has been introduced in Takahashi and Tucsnak [26], where it has been proven that this operator generates an analytic semigroup on X 2 . Later, Wang and Xin [30] proved that the operator A q generates an analytic semigroup on X 6/5 ∩ X q if q ≥ 2 and that if the solid is a ball in R 3 the operator A q generates an analytic semigroup (not necessarily bounded) on X 2 ∩ X q if q ≥ 6. One of our main result improves the result of Wang and Xin [30]. Actually, in Theorem 6.1 we will prove that A q generates a bounded analytic semigroup on X q for any q > 1. Moreover, this result is true for bodies of arbitrary shape. It is important for future use to rephrase the resolvent equation for A q, in a form involving only PDEs and algebraic constraints. To this aim, for λ ∈ C, we consider the system (3.14) In the above system the unknowns are v, π, and ω, whereas By slightly adapting the methodology used in [25,26] for the case q = 2, it can be checked that we have the following equivalence: Proposition 3.4 Let ⊂ R 3 be an open, connected and bounded set with ∂ of class C 2 or = R 3 . Let 1 < q < ∞ and λ ∈ C. Assume that f ∈ L q (E ) 3 and then Conversely, assume that F ∈ X q ( ) and V ∈ D(A q, ) satisfy (3.15). Then there

The fluid-structure semigroup on bounded domains
In this subsection we assume that is an open bounded set in R 3 with boundary of class C 2 . In this case the operator A q, has been extensively studied in Maity and Tucsnak [20]. In particular, by combining the density result from Proposition 3. , on The above result has the following consequence, which follows by standard analytic semigroups theory: Corollary 3.6 With the notation and under the assumptions in Theorem 3.5, for every θ ∈ π 2 , π the exists a constant M, possibly depending on q, θ, O and , such that By combining Corollary 3.6 and Proposition 3.4 we obtain the following result: Proposition 3.7 Let θ ∈ (π/2, π), q ∈ (1, ∞) and assume that ⊂ R 3 is bounded, with ∂ of class C 2 . Then there exists a constant C > 0, possibly depending on θ, q, and O, such that for all λ ∈ θ , f ∈ L q (E ) 3 and f , f ω ∈ C 3 , there exists a unique solution (v, π, , ω) ∈ W 2,q (E ) 3 × W 1,q (E ) × C 3 × C 3 of (3.14) satisfying We need below the following slight generalization of Proposition 3.7: Corollary 3.8 With the notation and assumptions in Proposition Then for every λ 0 > 0 there exists a constant C = C( , p, λ 0 , θ) such that for every λ ∈ θ with |λ| ≤ λ 0 .
Proof According to Lemma 2.1 there existsṽ ∈ W 2,q 0 (E ) 3 such that divṽ = div v on E and ṽ 2,q,E ≤ C div v 1,q,E , (3.18) where C is a constant depending only on and on q. Setting u = v −ṽ we see that u ∈ W 2,q (E ) 3 and By applying Proposition 3.7 and elementary inequalities, it follows that The above estimate and (3.18) imply the conclusion (3.17).

From the Stokes operator in exterior domains to the fluid-structure operator in the whole space
In this section, we study the fluid structure operator A q, , defined in (3.12) and (3.13), in the case = R 3 . As mentioned in Sects. 2 and 3, in this case the space X q ( ), defined in (3.1), and the operators P q, , A q, are simply denoted by X q , P q and A q , respectively. The main idea developed in this section is that the resolvent of the fluidstructure operator can be expressed in terms of the resolvent of the Stokes operator with homogeneous Dirichlet conditions on the boundary of an obstacle of arbitrary shape O. The connection between these two families of resolvents is then used to study the behaviour of the of (λI − A q ) −1 for λ close to zero, in the spirit of the similar results for the Stokes operator in exterior domains obtained by Iwashita [15].
Let O be an open, bounded subset of R 3 with ∂O of class C 2 and let E = R 3 \O. We consider the system 3 , f , f ω ∈ C 3 and λ ∈ C. In the above system the unknowns are u, π, and ω, whereas σ (u, π) := −π I + 2μD(u).
To study the solvability of (4.1) we introduce several auxiliary operators. Firstly, given λ ∈ C and , ω ∈ C 3 , we consider the boundary value problem: and we remind the notation (2.3) (and more generally the notation in Sect. 2) for the possibly weighted Sobolev spaces in unbounded domains.

Moreover, let (D λ ) λ∈ θ be the family of operators defined by
is the solution of (4.2). Then for every λ ∈ θ and m ∈ N, we have Finally, there exists for every m ∈ N, q > 1 and s < − 3 q .
Proof We choose two balls B 1 and for every x ∈ R 3 and We set We can apply classical regularity results for Stokes (e.g. [15, Proposition 2.7(i)]) to get (4.4) and Theorem 2.2 to obtain (4.5) and (4.6).
The above result allows us to introduce the family of operators (T λ ) λ∈ θ ⊂ L(C 6 ) defined by where (w, η) is the solution of (4.2), given by D λ according to (4.3).

Proposition 4.2
Let θ ∈ (0, π). For every λ ∈ θ let (T λ ) λ∈ θ be the operators defined in (4.7) and let (K λ ) λ∈ θ be the family of operators defined by Moreover, K λ is invertible for every λ ∈ θ and lim sup λ∈ θ ,λ→0 where D 0 is the operator introduced in Proposition 4.1. Applying Proposition 4.1 and a standard trace theorem it follows that (4.9) holds. The fact that K 0 (which is called the resistance matrix of O) is invertible is a classical result (see, for instance, Happel and Brenner [11,Section 5.4], where it is shown that this matrix is strictly positive).
On the other hand, taking the inner product in L 2 (E) 3 of the first equation in (4.2) by w, integrating by parts and using the second equation in (4.2) it follows that (4.11) Assume now that , ω ∈ C 3 and λ ∈ θ are such that Taking the inner product in C 6 of the two sides of the above formula by ω and using (4.11) it follows that If λ ∈ θ with Im λ = 0 it follows that = 0 and ω = 0. On the other hand, if λ ∈ θ and Imλ = 0 we have Reλ > 0. In this case, we obtain w = 0 and consequently = ω = 0. We have thus shown that the operator in (4.8) is invertible for every λ ∈ θ . This fact, (4.9) and the fact that K 0 is invertible finally imply (4.10).
We are now in a position to state the main result in this section.

Further properties of the fluid-structure semigroup in R 3
In this section we study the fluid structure operator A q, , defined in (3.12) and (3.13), in the case = R 3 . More precisely, we give several results opening the way to the proofs of the facts that A q generates a bounded analytic semigroup and of the decay estimates for the fluid-structure operator by collecting several results which follow quite easily from the existing literature. The first one is: Then there exist γ > 0 and m q,θ > 0 such that Consequently, A q generates an analytic semigroup on X q .
The proof of the above result can be obtained by a perturbation argument. Since this argument is a slight variation of the proof of Theorem 3.1 in [20], where the similar estimate is detailed for the case of fluid-structure system confined in a bounded domain, we omit the proof. We also note that by combining Proposition 3.4 and the first statement of Theorem 4.3, we have Proposition 5.2 For every λ ∈ θ and F ∈ X q , setting where the family (R(λ)) has been introduced in (4.12) and The result below provides some simple but important properties of the fluidstructure operator A q .
We consider the equation which according to Proposition 3.4 is equivalent to the system Assume that u ∈ W 2,q (E) 3 , π ∈ W 1,q (E), ∈ C 3 and ω ∈ C 3 satisfy the system (4.1). Taking the inner product in C 3 , of (5.6 1 ) by u and of (4.1) by ϕ, integrating by parts and summing up the two formulas we obtain Using the boundary conditions, the above relation can be written as In terms of the operator A q and A q , the above equality reads as . In order to prove the reverse inclusion, we first note that, for Let U ∈ D(A q ). Then using the last two formulas, we obtain In particular, we have Therefore W = U and this completes the proof.
The last result in this section provides some information on the resolvent equation associated to A q .
We first consider the case q = 2. Multiplying, (4.1 1 ) by u, (4.1 4 ) by and (4.1 5 ) by ω, we obtain after integration by parts: Note that, to justify properly these computations, we should multiply (4.1 1 ) by ϕ R u, where ϕ R = ϕ(x/R), ϕ being a smooth cut-off function taking value one close to the unit ball and vanishing outside the ball of radius 2, and R being a large positive parameter. One should then prove the following convergences, the first limit coming from Lebesgue dominated convergence theorem and the second from the fact that u ∈ L 2 (E) and ∇u ∈ L 2 (E). The last limit is more delicate and is based on the fact that, since ∇π ∈ L 2 (E), there exists a constant c π such that π + c π ∈ L 6 (E). Then we can write To conclude (5.13), it then remains to check that ∇ϕ R 3,R 3 is bounded uniformly in R, while u 2,R 3 \B(R) goes to 0 as R → ∞. If Imλ = 0, we take the imaginary part of identity (5.10) and obtain that u = π = = ω = 0. If Imλ = 0, then Reλ ≥ 0, hence using the above identity and the boundary conditions we also obtain u = π = = ω = 0.
Let us then consider the case q > 2 and λ = 0. Let B 1 and B 2 be two open balls in R 3 such that Then ϕ 1 u satisfies the following system . Therefore, by using Similarly, (ϕ 2 u, ϕ 2 π) satisfies the following system By standard results on Stokes operator in the whole space, we also get (ϕ 2 u, Combining the above results we obtain u ∈ W 2,2 (E) 3 and π ∈ W 1,2 (E).
Let us consider the case 1 < q < 2 and λ = 0. We use a bootstrap argument here.
we are reduced to the previous case. Otherwise, we continue the process until we get u ∈ W 2,2 (E) 3 .
We next consider the case λ = 0, which only consists in justifying identity (5.10) in that case, since u = π = = ω = 0 would then follow immediately.

Analyticity of the fluid-structure semigroup
We begin by stating the main result in this section, which, besides being of independent interest, is an important ingredient in the proof of our main results. In fact, as mentioned earlier, this result improves the existing result of [30, Theorem 2.5, Theorem 2.9].
The guiding idea in proving the above result is borrowed from Borchers and Sohr [2] and it consists in using a contradiction argument and appropriate cut-off functions, combined with Proposition 3.7 and classical results for the Stokes operator in the whole space.
Then there exists a constant M q,θ > 0 such that, for every
We are now in position to prove the main result in this section.
Proof of Theorem 6. 1 We first note that from Proposition 3.4, Theorem 4.3 and Proposition 6.2, we obtain (6.1) for 1 < q < 3 2 . In the case 3 2 ≤ q ≤ 2 we take q 0 ∈ (1, 3 2 ). We define 0 ≤ s ≤ 1 by Since (6.1) holds for q 0 , there exists a constant M θ,q 0 > 0 such that On the other hand, A 2 is a self-adjoint operator on X 2 (see [26]). Therefore, we also have for some M θ,2 depending only on θ. Then by Riesz-Thorin interpolation theorem (see for instance [28, Theorem 1, Section 1.18.7]), we obtain This ends the proof of (6.1) for In the case 2 < q < ∞, we take 1 < q ≤ 2 such that 1 q We have already seen that (6.1) holds for 1 < q ≤ 2. Thus from the above identity we infer that, (6.1) holds for any 2 < q < ∞, which ends the proof.
We end this section with the result below, whose proof can be easily obtained by combining Theorem 6.1 and the results from Lunardi [19, Chapter 3]: Corollary 6.3 With the assumptions and notations of Theorem 6.1, for any ε > 0 and k ∈ N, there exists C ε > 0 such that ). (6.23) 7 Decay estimates for the fluid-structure semigroup Based on Theorem 6.1, we consider the fluid-structure semigroup which is, for each q ∈ (1, ∞), the bounded analytic semigroup T q introduced in Theorem 6.1. Our main result in this section is: Then for any R > R 0 , there exists a constant C > 0, depending on q and R, such that Then there exists a constant C > 0, depending on q and r , such that Then there exists a constant C > 0, depending on q and r , such that (iv) Estimate (7.2) also holds for 1 < q < ∞ and r = ∞.
Let us emphasize that Theorem 7.1 holds for the linearized fluid-structure equations for bodies O of arbitrary shapes. It seems thus likely that these properties can be used to derive the well-posedness for solids of arbitrary shape, see the discussion in Sect. 9 below.
Let us also mention that, Maremonti and Solonnikov in [21] proved that, while considering Stokes equation in the exterior domain, the same decay estimates hold, and the estimate (7.2) are sharp for 3/2 ≤ q ≤ r ≤ ∞. It is then expected that same holds for the fluid-structure operator also. This is indeed the case, at least in the case of the ball, see Theorem B.1 in the appendix for more details.
Our methodology to prove the above result is inspired by [15] and it consists in using the resolvent estimates developed in Sects. 4-6. However, applying the strategy proposed in [15] requires several adaptations which are described below.
To start with, we state the following regularity result of the projection operator P q .
such that div u = 0 in D (R 3 ) and ∂ α u ∈ L r (E) 3 for every multi-index α ∈ Z 3 + with |α| = k. Then ∂ α (P q u) ∈ L r (E) 3 for every multi-index α ∈ Z 3 + with |α| = k. Moreover, there exists a constant C independent of the choice of u with the above properties, such that Since div u = 0, we have that w 1 from the decomposition (3.4) of u vanishes and, according to [30, Proof of Theorem 2.2, Eq. (3.15)], w 2 from the same decomposition satisfies w 2 = ∇π 2 , with π 2 satisfying results characterising the graph norm of A m q in terms of Sobolev spaces. 3 and there exists a constant C m > 0 such that , then U | E ∈ W 2m,q (E) and there exists a constant C m > 0 such that ). (7.10) Proof Let us set A q U = −F, so that F| E ∈ W m,q (E) 3 . Moreover, we denote Then according to Proposition 5.2 there exists π ∈ W 1,q (E) such that u, π, and ω satisfy Let where D 1 is the Dirichlet map introduced in Proposition 4.1.
According to Proposition 4.1, for every k ∈ N there exists positive constants C 1,k , C 2,k such that We denote u = u − w 1 and π = π − η 1 . Then u and π satisfy According to [15, Proposition 2.7(i)], for every m ∈ N there exists a positive constant C 3,m such that u m+2,q,E ≤ C 3,m F m,q,E + w 1 m,q,E + u q,E .
The above estimate together with (7.11) implies the estimate (7.9).
To prove (7.10), we use an induction argument. We first note that (7.10) is true for m = 1, since it is nothing else but the estimate (7.9) for m = 0. Let us assume that (7.10) is true for some m ∈ N and U ∈ D(A m+1 q ). Then by (7.9) and induction hypothesis, there exists a positive constant C m > 0 such that Then the assertion (7.10) holds for m replaced by m + 1 by applying Corollary 6.3 repeatedly and (7.12). This completes the proof of the proposition.

Proposition 7.4 Let q ∈ (1, ∞). Then:
(i) For any m ∈ N, there exists a positive constant C m > 0 such that , π and m ∈ N. Then there exists a positive constant C m > 0 such that Proof We use an induction argument to prove (7.13). Using Proposition 3.1, (3.9) and (3.10) 111 note that the estimate (7.13) is true for m = 1. Assume that (7.13) holds for some m ∈ N and U ∈ D(A m+1 q ). By the induction hypothesis, we have By applying Proposition 7.2 and Corollary 6.3, the above estimate implies that ) . (7.16) Thus (7.13) also holds when m is replaced by m + 1. Finally (7.14) follows from the facts that together with the estimates (7.10) and (7.13).

Remark 7.5
Putting together (7.10) and (7.13), it follows that, for every m ∈ N, the graph norm of A m q is equivalent to · 2m,q,E + · X q . We also note that this equivalence also holds for the bounded domain version of the fluid-structure operator, i.e., ⊂ R 3 open and bounded, and the operator A q, defined in (3.9). Moreover, elements ϕ of To state the next results, which yield decay estimates for the fluid-structure semigroup in weighted L p spaces, we remind from Sect. 2 the notation L q,s for the weighted Lebesgue spaces introduced in (2.3). Theorem 7.6 Let 1 < q < ∞. Let s and s be real numbers such that s > 3(1 − 1/q) and s < −3/q. Then there exists a positive constant C, depending only on q, s and s , such that 3 ).

(7.18)
Proof We first note that Theorem 4.3 is a complete analogue of Corollary 3.2 in [15], and Theorem 6.1 is the analogue of the main result in [2]. We can thus complete the proof following line by line the proof of Theorem 1.1 in [15].

Remark 7.7
For U 0 ∈ D(A q ), we denote by Moreover for every t ≥ 0, we set U (t) = T q t U 0 and Then there exists π ∈ C([0, ∞); W 1,q (E)) such that (u, π, , ω) satisfies the following system Our next result in this section provides L q − L r smoothing estimates for the fluidstructure semigroup T q for small time: Then for each τ ∈ (0, ∞), there exists a constant C > 0, depending on τ, q and r , such that denotes the integer part function. Let us assume that N is even. Then by (7.10), there exists a constant C > 0 depending on τ, q and r , such that 0, τ ]).

(7.22)
In a similar manner, we also obtain Thus by Sobolev embedding, interpolation and using (7.22)-(7.23), we obtain If N is odd then we replace N by N − 1. This completes the proof of (7.20). The proof of (7.21) is completely similar, thus omitted here.
The next step towards the proof of Theorem 7.1 is the following result:  (7.24) for every U ∈ D(A m q ) with U = 0 for |x| > d. (ii) There exists a constant C > 0 depending on d and m such that for all t > 0, Proof The proof can be obtained following line by line the proof Lemma 5.2 from Iwashita [15]. More precisely, it suffices to use instead of Proposition 2.7 and Lemma 2.8 in [15] our results in Proposition 7.3 and Proposition 7.4 above, respectively, and to replace expansion (3.2) in [15] by (4.15) above. Proposition 7.10 With the notation and assumptions of Remark 7.7 and Theorem 7.1, let d > R 0 + 5 and m ∈ N. Moreover, assume that U 0 ∈ Ran(T q 1 ). Then there exists a positive constant C, depending only on E, d, m and q, such that, for every t ≥ 0 we have where [s] denotes the integer part of s ∈ R.
Proof We follow with minor modifications the steps of the proof of Lemma 5.3 in [15].
where B O (div u 0 ) is seen as a function in W 2m,q (R 3 ) 3

after its extension by 0 in E.
Then ψ ∈ W 2m,q (R 3 ) 3 has the following properties Step 2. We consider the following Stokes system in R 3 Let q and r be such that 1 < q ≤ r ≤ ∞ and define σ = 3 2 ( 1 q − 1 r ). According to classical estimates (see, for instance, [15,Lemma 5.1]) for the heat kernel, for every m ∈ Z + , there exists a constant C m > 0, depending on q and r , with By applying (7.32) and (7.34), it follows that there exists a constant C m > 0, depending on q, such that Step 3. We now set Then v 2 , π, and ω satisfy for every m ∈ N and there exists a constant C > 0, depending on m and q, such that (7.44) Using (7.31), (7.32), (7.36) and (7.37), we infer that there exists a constant C > 0, depending only on m and q, such that On the other hand, applying the variation of the constants formula to (7.39), we have The last estimate, combined with Lemma 7.9, can be used, following line by line the end of the third step of the proof of Lemma 5.3 in [15], to obtain the existence of a constant C (depending only on d, m and q), such that for every t > 0 we have Final step. Estimates (7.26) easily follow by combining (7.38) with the estimates (7.31)-(7.33), (7.36), (7.37), (7.47) and (7.48). The estimate (7.27) can be obtained similarly. Putting together (7.26) and (7.27), from (7.19 1 ) we obtain Then the estimate (7.28) follows from the above estimate after redefining π as π − E d π dx and applying Poincaré type inequalities.
The results in Lemma 7.9 and Proposition 7.10 provide decay estimates for the restrictions to bounded sets of the solution u of the linearized problem. The result below provides decay estimates for the restriction of u(t, ·) to the exterior of the bounded set E d introduced in Lemma 7.9. Proposition 7.11 With the notation and assumptions of Remark 7.7 and Theorem 7.1, let d > R 0 + 5. Moreover, assume that U 0 ∈ Ran(T q 1 ). Then, for q ∈ (1, ∞) and r ∈ [q, ∞), there exists a positive constant C, depending only on E, d, q and r such that, where [s] denotes the integer part of s ∈ R.
for some constant C > 0 depending on m and q. To derive the last two estimates we have used Bogovskii Lemma, (7.26) and (7.27). We now define Note that, since all the functions appearing in (7.54) are supported away from O, the function v 4 shares all the properties derived in Lemma 5.5 and proofs of Theorems 1.2 and 1.3 of [15]. In particular, (7.58) By combining (7.57), (7.51) (with m = [2σ ] + 1) and Sobolev's embedding theorem we conclude that This completes the proof of (7.49). Finally, the proof of (7.50) is obtained similarly from (7.58), together with (7.51) (with m = [2σ ] + 2).
We are now in a position to prove the main result in this section.

Proof of Theorem 7.1, items (i)-(iii)
For small times, Theorem 7.1 items (i)-(iii) simply is Theorem 7.8, and we thus only focus on the estimates of Theorem 7.1 items (i)-(iii) for times larger than 1.
To complete the proof of Theorem 7.1, it remains to show that (7.2) holds for r = ∞ and 1 < q < ∞. To this aim, we first prove the following result Proposition 7.12 With the notation and assumptions of Remark 7.7 and Theorem 7.1, let d > R 0 + 5. Moreover, assume that 1 < q < ∞ and U 0 ∈ Ran(T q 1 ). Then, for q ∈ (1, ∞), there exists a positive constant C, depending only on E, d and q, such that for every t ≥ 0, we have where [s] denotes the integer part of s ∈ R.
Proof Combining (7.38) together with the estimates (7.32)(with m = 0, r = ∞), (7.36)(with m = 3) and (7.47) (with m = 2), we deduce that where the set E d has been defined in Lemma 7.9. Moreover, using (7.51) with m = 3 we also have where v 3 is defined as in Proposition 7.11. Therefore, by virtue of the decomposition (7.53), it remains to show the L ∞ estimate of v 4 , where v 4 is defined by (7.54). Recall the definition of h from (7.55). Using (7.26), (7.28), (7.51) and (7.52), we obtain for Let us take q 0 = min{5/4, q}. Then using (7.32) and the above estimate, we evaluate Note that, in the above estimate, we have used the fact that h has compact support, which comes from its definition in (7.55) and the fact that v 3 is compactly supported. The above estimate together with (7.61) and (7.62) implies (7.60).
Note that, for every U ∈ X q we have T q 1 U ∈ D(A k q ) for any k ∈ N. Thus applying (7.60) we have This proves (7.2) for q ∈ (1, ∞), r = ∞, and t ≥ 1.
For t ∈ (0, 1), 1 < q < ∞, and q 0 > max{3, q}, we apply Theorem 7.8 to obtain where the constant C is independent of t. This completes the proof of item (iv) of Theorem 7.1.

Proof of the main results
In this section, we focus on the analysis of the non-linear fluid-structure model, assuming that the rigid body is a ball. Note that this assumption has been already used when fixing the frame via a simple translation, which drastically simplifies the structure of the non-linear terms. Given q > 1 we continue to use the notation X q for the space defined in Eq. (3.1) and T q for the fluid structure semigroup introduced in the previous sections. However, to simplify the notation and when there is no risk of confusion, the fluid-structure semigroup will be simply denoted by T. Similarly, if the appropriate value of q is clear from the context, the projector P q , introduced in Proposition 3.2, is simply denoted by P. The arguments we are using are close to those in Kato [16], with several adaptations necessary to tackle the extra term coming from the motion of the rigid body, in a spirit close to [6], and with the extensive use of the results obtained in the previous sections on the fluid-structure semigroup, and in particular Theorem 7.1.
We rely, in particular, on the following lemma, which is a rather straightforward consequence of Theorem 7.1: Lemma 8.1 Let p 0 and q 0 be such that q 0 ∈ [3/2, ∞) and p 0 ∈ [q 0 , ∞]. Then there exists C > 0 such that for every F ∈ L q 0 (R 3 ; R 3×3 ) satisfying F = 0 in B and div F ∈ L r (R 3 ) 3 for some r ∈ (1, p 0 ]\{∞} we have Proof The proof follows the same steps as those appearing in [6, Proof of Corollary 3.10]. More precisely, for F ∈ L q 0 (R 3 ; R 3×3 ) satisfying F = 0 in B and div F ∈ L r (R 3 ) 3 for some r ∈ (1, p 0 ]\{∞}, we necessarily have F · ν = 0 on ∂ B and for t > 0, T r t P r div F is a well-defined element of X r ∩ X ∞ (see Theorem 7.1). Setting r 0 = max{r , q 0 }, we thus have that Finally, using (7.3), we obtain where the last estimate comes from (7.2).
We state below a result which, taking in consideration the notation recalled at the beginning of this section, clearly includes our main result in Theorem 1.1. More precisely, this result provides local in time existence of solutions for initial data in X 3 and global existence of solutions of (1.2) for small data in X 3 , with a description of the long time behavior of the solutions when further assuming that the initial datum belongs to X q for some q ∈ (1, 3]. 9 ) of (1.2), such that Furthermore, this solution is such that for all p ∈ [3, ∞], t 3/2(1/3−1/ p) V ∈ C 0 ([0, T 0 ]; X p ).
For q ∈ (1, 3], there exists ε 0 (q) ∈ (0, ε 0 ] such that if V 0 belongs to X q ∩X 3 and satisfies V 0 3 ≤ ε 0 (q), then the solution V also satisfies, for all p ∈ [max{q, 3/2}, ∞], Proof Existence theory for V 0 ∈ X 3 . We first focus on the existence of solutions V of (1.2). As mentioned in the introduction, we are looking for mild solutions V of the nonlinear problem (1.2), i.e. solutions of the Eq. (1.8).
We first remark that, since In particular the triple V = ⎡ ⎣ v ω ⎤ ⎦ is a mild solution of (1.2) iff it satisfies where F is defined in (8.3). The above formulation will be intensively used in the remaining part of the section, in conjunction with Lemma 8.1.
For T > 0, we introduce the class which we endow with the norm Note in particular that, for V = ⎡ ⎣ v ω ⎤ ⎦ ∈ C (T ), we have the estimate We start by remarking that we are looking for a solution V of (1.8) or equivalently (8.4). We then define the map T : (8.5) or, equivalently, Next we claim the following lemma: Proof Since estimate (8.6) can be easily deduced from (8.7) by taking V a = 0 and V b = V , we prove only (8.7). Let V a and V b be in C (T ). Then easy computations show that where G is given by Writing we easily deduce that for all s ∈ (0, T ), Besides, for all s ∈ (0, T ], div G(s) belongs to [L 3 (R 3 )] 3 since V a and V b belongs to C (T ). Therefore, using Lemma 8.1, with p 0 = q 0 = 6, for some C independent of T , where we used that, by scaling, for all t > 0, Similarly, using again Lemma 8.1, with p 0 = ∞, q 0 = 6, we get, for some C independent of T , where g is given that we decompose as We then bound g in [L 3 (R 3 )] 9 : for s ∈ (0, T ], We also bound g v in [L 2 (R 3 )] 9 : for s ∈ (0, T ], Accordingly, estimate (7.3) yields We then show that the supremum in t ∈ [0, T ] can in fact be bounded by a constant independent of T . For t ∈ (0, 1), where c α,β is defined for (α, β) ∈ (0, 1) 2 by For t ≥ 1, we write  Consequently, there exists some C independent of T > 0 such that Putting together estimates (8.8), (8.9) and (8.14), we conclude the estimate (8.7) and Lemma 8.3.
According to Lemma 8.3, for K > 0, the set C (T , where C 0 is the constant in Lemma 8.3. Therefore, for K is a continuous increasing function of T , which is bounded by C V 0 X 3 according to estimates (7.2)-(7.3), and which goes to 0 as T to 0 by density of D(A 3 ) in X 3 and the decay estimates (7.2)-(7.3).
Therefore, for V 0 ∈ X 3 , we can guarantee that there exists a time T K > 0 such that so that by (8.6), for K ≤ K 0 = 1/(4C 0 ), the set C (T K , K ) is stable by T K , and T K is strictly contractive in it. Therefore, by Banach-Picard fixed point theorem, there exists a fixed point V ∈ C (T K ), which is by construction a mild solution of (1.2). Besides, this solution is such that for all T < T K , V | (0,T ) is the fixed point of T in C (T , K ). Therefore, for T such that T t V 0 C (T ) ≤ 1/(8C 0 ), it is easy to check from (8.6) that V | (0,T ) C (T ) ≤ 2 T t V 0 C (T ) , and thus goes to 0 as T → 0.
Furthermore, V can be constructed as the limit of the sequence V n+1 = T K V n for n ∈ N, V 1 = 0, for which we have, for all n ∈ N, V n ∈ C (T K , K ). Elements of this sequence satisfies In particular, using Lemma 8.1, we get Therefore, taking K smaller if necessary to guarantee that C K < 1, we get that the sequence V n is also uniformly bounded in L ∞ (0, T K ; X 3 ), so that its limit V is also bounded in L ∞ (0, T K ; X 3 ). We then easily deduce by interpolation that, for all Besides, since T t V 0 C (∞) ≤ C 1 V 0 X 3 according to estimates (7.2)-(7.3) for some constant C 1 , if V 0 X 3 is small enough (namely ≤ 1/(8C 0 C 1 )), we can take K = 2C V 0 X 3 and T K = ∞. In this case, we have from the above computations that the above sequence V n stays in C (∞, K ) and stays bounded in Consequently, when V 0 X 3 is small enough, we can deduce by interpolation that there exists C > 0 such that for all n ∈ N, sup p∈ [3,∞] 17) and this also holds for the limit V of the sequence V n : sup p∈ [3,∞] Now, we prove that we also have that for all θ ∈ [0, 1/2), min{t 1/2 , t θ }∇v ∈ L ∞ (0, ∞; [L 3 (E)] 9 ), taking V 0 X 3 smaller if needed. Indeed, starting from (8.16), that we rewrite 18) using (7.3) with r = 3 and q = 3 and q = 3/2, we have, for all t > 0 and n ∈ N, where we used that the sequence (V n ) satisfies (8.17), and Now, arguing as in (8.10)-(8.13), one can easily check that there exists a constant C independent of t such that for all t > 0, Consequently, combining the above estimates, we have, for all n ∈ N, 9 ) . (8.19) It follows that if V 0 X 3 is small enough, then the sequence (min{t 1/2 , t θ }∇v n (t)) is bounded in L ∞ (0, ∞; [L 3 (E)] 9 ), and since v n converges to v in C (∞), we obtain Uniqueness. Let V a and V b be two mild solutions of (1.2) with the same initial datum V 0 in the class C (T 0 ) such that for V a C (T ) and V b C (T ) go to 0 as T → 0. Then, setting according to (8.7), we have Since V a C (T ) and V b C (T ) go to 0 as T → 0, there exists t 0 ∈ (0, T 0 ], such that and thus e(t 0 ) = 0, and V a and V b coincides on [0, t 0 ]. If t 0 < T 0 , it is easily seen that this argument can be repeated on time intervals of the form [t * , T ] with t * ≥ t 0 : Using that for t ∈ (0, 1) such that t + t * ≤ T 0 , . we immediately have that there exists t 1 > 0 such that if V a and V b coincide in [0, t * ] with t * ≥ t 0 , then they coincide on [0, min{t * + t 1 , T 0 }]. This argument proves that V a and V b in fact coincide on the whole time interval [0, T 0 ]. The case of an initial datum in X q for q ∈ (1, 3). Let q ∈ (1, 3) and V 0 ∈ X q ∩ X 3 with V 0 X 3 ≤ ε 0 . Then we know that the solution V of (1.2) is global in time and belong to C (∞), and we know that the sequence given by V 1 = 0 and converges to V in C (∞), and we have the estimates (8.17).
To prove that V is bounded in some class, it is enough to check that the sequence (V n ) n∈N is uniformly bounded in this class.

Concluding remarks and open questions
The main result in this paper, namely Theorem 1.1, concerns the wellposedness of the system modelling the motion of a rigid ball in a viscous incompressible fluid filling the remaining part of R 3 and asserts that the position of the centre of the ball tends, when t → ∞, to some position h ∞ ∈ R 3 . This result differs from those previously obtained in two space dimensions in [6], or for a simplified 1D model in [29], where it has been shown that the distance of the centre of the ball to the origin tends to +∞ when t → ∞. Several open questions seem natural in view of our results.
One of the most challenging ones, for which we have no track at this stage, is determining h ∞ from the initial data.
Another natural question is the generalization of Theorem 1.1 for a body of arbitrary shape. When the rigid body is not a ball, writing the equations in a fixed domain requires the use of more delicate changes of variables, since it has to include the rotation of the body. There are basically two ways of doing that: one consists in setting v(t, x) = Q * (t)u(t, h(t) + Q(t)x) , where Q(t) is the rotation matrix of the body, that is the solution ofQ(t)Q * (t)x = ω(t) × x starting from Q(0) = I d. The problem is that such change of frame would induce in the fixed frame a term of the form (ω × x) · ∇v which our estimate does not allow to handle since the identity mapping does not belong to L ∞ (E). The alternative approach proposed in [4], which consists in constructing a change of variable which follows the structure in a neighbourhood of it and equals the identity far from the body, seems more suitable to deal with the non-linear terms. However, this change of variable introduces a lot of delicate terms which we do not know how to handle in the above setting so far. In fact, even in two space dimensions, the existing results (see [6]) provide an analysis of the motion of a rigid body in a viscous incompressible fluid in R 2 only in the case when the rigid body is a disk.
Finally, let us mention that the counterparts in two space dimensions of some of our results in Sects. 4-7 have been used in Takahashi and Lacave [18] to study the behaviour of solutions of (1.2) when the radius of the rigid ball tends to zero (see also He and Iftimie [12] and references therein). We believe that the approach in [18] can be adapted to the three dimensional case by using our results on the fluid structure-semigroup and its generator, but this deserves further work.

A.1 Proof of Proposition 3.2
We first show that every u ∈ L q ( ) 3 can be written in the form u = v + w 1 + w 2 , with v ∈ X q ( ), w 1 ∈ G q 1 ( ) and w 2 ∈ G q 2 ( ). To this aim, let q 11 be the solution of the problem q 11 = div u in , q 11 = 0 on ∂ .
Setting then q 1 = q 11 + q 12 , and w 1 = ∇q 1 , w 1 ∈ G q 1 ( ), and div (u − w 1 ) = 0 in and (u − w 1 ) · ν = 0 on ∂ . Since we are looking for v ∈ X q ( ), we know that for some v ∈ R 3 and ω v ∈ R 3 . We set and where m and J are defined in (3.11).
Now we define q 2 as the solution of the Neumann problem ⎧ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎩ q 2 = 0 i n E , ∂q 2 ∂ν = 0 o n ∂ , where v and ω v are defined in (A.4) and (A.5), respectively. Note that, q 2 solves a Laplace equation with non-local boundary condition. As shown below, we have that q 2 ∈ W 1,q (E ) and there exists a constant C depending on q, and O such that In this case we can determine v and ω v from (A.4) and (A.5) respectively. Consequently, we obtain ϕ and w 2 from (A.2) and (A.3) respectively. In particular, we have that w 2 ∈ G q 2 ( ) and by setting v = u − w 1 − w 2 we can verify that v ∈ X q ( ). We still have to prove that q 2 ∈ W 1,q ( ) and (A.7) holds. If q = 2, this is a consequence of Lax-Milgram Theorem (see for instance [5,Lemma 1]). If q = 2, we employ a density argument. Assume that u ∈ C ∞ 0 ( ) 3  where C is a positive constant depending only on q, and O. The above two estimates yield that there exists a positive constant C, depending only on q, and O, such that estimate (A.7) holds. Thus the conclusion follows by a density argument. This completes the proof of the existence of a decomposition with the required properties. The proof of uniqueness of the decomposition is similar to that of [30, Theorem 2.2].

A.2 Proof of Proposition 3.3
Let Y q ( ) ⊥ and (Y q ) ⊥ be the annihilators of Y q ( ) and Y q respectively, i.e., Let us set G q ( ) = G q 1 ( ) ⊕ G q 2 ( ) and G q = G q 1 ⊕ G q 2 .
Step 1. We claim that Y q ( ) ⊥ = G q ( ) and (Y q ) ⊥ = G q . Let us assume that u ∈ G q ( ) = G q 1 ( ) ⊕ G q 2 ( ), i.e., u = u 1 + u 2 with u 1 ∈ G q 1 ( ) and u 2 ∈ G q 2 ( ). Let us take v ∈ C ∞ c ( ), div v = 0 in , and Dv = 0 in O. Then we have Thus by density u · v = 0 for all v ∈ Y q ( ). This shows that u ∈ Y q ( ) ⊥ and G q ( ) ⊂ Y q ( ) ⊥ . In a similar manner, we can show G q ⊂ (Y q ) ⊥ .
Since u ∈ Y q ( ) ⊥ , by similar calculation as above, we obtain for all v ∈ C ∞ c ( ), with div v = 0 in and Dv = 0 in O, We claim that w ∈ G q 2 ( ). By choosing, v ∈ C ∞ c (E ) with div v = 0 in E , we have E w · v dx = 0 for all v ∈ C ∞ c (E ), with div v = 0 in E .
Then by de Rham's theorem there is a distribution q 2 such that w = ∇q 2 in E . Further, from the fact that w ∈ X q ( ) we have q 2 ∈ L 1 loc (E ), and if is a bounded domain we have q 2 ∈ W 1,q (E ). Furthermore, we have w = w + ω w × y in O, and In particular, Therefore w ∈ G q ( ). Since by construction, w = P q , u belongs to X q ( ), and X q ( ) ∩ G q ( ) = {0}, w = 0 and u = w 0 belongs to G q ( ). The proof is similar when = R 3 .

Appendix B Optimality of the decay estimates in (7.2) in the case of a ball
In this section, following [21, p.441], we prove the optimality of the decay estimates (7.2) when O is the unit ball (the extension to any ball can be done analogously and is left to the reader).

Theorem B.1
Let O be the unit ball, and T q t be the fluid-structure semigroup introduced in Theorem 6.1 for q ∈ (1, ∞).
There is no ε > 0 and r ≥ q ≥ 3/2 with r ≤ ∞ such that the semigroup T t = T q t satisfies T t U X r ≤ Ct −σ −ε U X q (t > 0, U ∈ X q ), (B.1) where σ = 3 2 1 q − 1 r . In other words, item (ii) and (iv) of Theorem 7.1 are sharp decay estimates.