Research article

On the analysis of a mechanically consistent model of fluid-structure-contact interaction

  • Received: 08 November 2023 Revised: 25 March 2024 Accepted: 04 May 2024 Published: 06 June 2024
  • This paper is devoted to the mathematical analysis of the contact capabilities of the fluid-structure interaction (FSI) model with seepage reported in [Comput. Methods Appl. Mech., 392:114637, 2022]. In the case of a rigid disk moving over a fixed horizontal plane, we show that this model encompasses contact and hence removes the non collision paradox of traditional FSI models which rely on Dirichlet or Dirichlet/Navier boundary conditions. Numerical evidence on the theoretical results is also provided.

    Citation: Marguerite Champion, Miguel A. Fernández, Céline Grandmont, Fabien Vergnet, Marina Vidrascu. On the analysis of a mechanically consistent model of fluid-structure-contact interaction[J]. Mathematics in Engineering, 2024, 6(3): 425-467. doi: 10.3934/mine.2024018

    Related Papers:

    [1] Ivan Fumagalli . Discontinuous Galerkin method for a three-dimensional coupled fluid-poroelastic model with applications to brain fluid mechanics. Mathematics in Engineering, 2025, 7(2): 130-161. doi: 10.3934/mine.2025006
    [2] M. M. Bhatti, Efstathios E. Michaelides . Oldroyd 6-constant Electro-magneto-hydrodynamic fluid flow through parallel micro-plates with heat transfer using Darcy-Brinkman-Forchheimer model: A parametric investigation. Mathematics in Engineering, 2023, 5(3): 1-19. doi: 10.3934/mine.2023051
    [3] Andrea Manzoni, Alfio Quarteroni, Sandro Salsa . A saddle point approach to an optimal boundary control problem for steady Navier-Stokes equations. Mathematics in Engineering, 2019, 1(2): 252-280. doi: 10.3934/mine.2019.2.252
    [4] Francesca Tedeschi, Giulio G. Giusteri, Leonid Yelash, Mária Lukáčová-Medvid'ová . A multi-scale method for complex flows of non-Newtonian fluids. Mathematics in Engineering, 2022, 4(6): 1-22. doi: 10.3934/mine.2022050
    [5] Yangyang Qiao, Qing Li, Steinar Evje . On the numerical discretization of a tumor progression model driven by competing migration mechanisms. Mathematics in Engineering, 2022, 4(6): 1-24. doi: 10.3934/mine.2022046
    [6] Camilla Nobili . The role of boundary conditions in scaling laws for turbulent heat transport. Mathematics in Engineering, 2023, 5(1): 1-41. doi: 10.3934/mine.2023013
    [7] Giuseppe Procopio, Massimiliano Giona . Bitensorial formulation of the singularity method for Stokes flows. Mathematics in Engineering, 2023, 5(2): 1-34. doi: 10.3934/mine.2023046
    [8] Jason Howell, Katelynn Huneycutt, Justin T. Webster, Spencer Wilder . A thorough look at the (in)stability of piston-theoretic beams. Mathematics in Engineering, 2019, 1(3): 614-647. doi: 10.3934/mine.2019.3.614
    [9] Paola F. Antonietti, Chiara Facciolà, Marco Verani . Unified analysis of discontinuous Galerkin approximations of flows in fractured porous media on polygonal and polyhedral grids. Mathematics in Engineering, 2020, 2(2): 340-385. doi: 10.3934/mine.2020017
    [10] Boubacar Fall, Filippo Santambrogio, Diaraf Seck . Shape derivative for obstacles in crowd motion. Mathematics in Engineering, 2022, 4(2): 1-16. doi: 10.3934/mine.2022012
  • This paper is devoted to the mathematical analysis of the contact capabilities of the fluid-structure interaction (FSI) model with seepage reported in [Comput. Methods Appl. Mech., 392:114637, 2022]. In the case of a rigid disk moving over a fixed horizontal plane, we show that this model encompasses contact and hence removes the non collision paradox of traditional FSI models which rely on Dirichlet or Dirichlet/Navier boundary conditions. Numerical evidence on the theoretical results is also provided.



    The numerical simulation of systems involving fluid-structure-contact interaction is of fundamental importance in many engineering and biomedical applications. For instance, contact modeling is a crucial ingredient in the simulation of the dynamics of native or artificial cardiac valves (see, e.g., [22,23]).

    Modeling contact between solids within a fluid-structure interaction (FSI) framework raises many modeling, mathematical and numerical issues. First, in the case of a ball immersed in a viscous incompressible fluid with no-slip boundary conditions and falling over a fixed horizontal plane, the resulting FSI models are unable to predict contact, both in 2D (see, e.g., [17]) and in 3D (see, e.g., [19]), which is known as the no collision paradox. One of the most widespread explanation of this paradox is that one can no longer consider ideal smooth surfaces when solids come into contact: roughness-induced effects play a fundamental role into enabling collision. Indeed, contact between smooth solids with no-slip boundary conditions seems possible only in very specific configurations like, for example, grazing collision in 3D (see [21]). On the other hand, many studies show that the no collision paradox is circumvented by taking into account roughness in FSI models, either by enabling the fluid to slip through Navier-type boundary conditions (see, e.g., [20,29] in 2D and [12,13,15] in 3D) or by considering rough solids (see, e.g., [11] in 2D and [12] in 3D). A second major difficulty is related to the mechanical consistency of the model. Indeed, for FSI models which allow for contact, the simple addition of a non-penetration constraint to the solid can lead to unphysical void creation (at release from contact) or unbalanced stresses at contact. Recently, these mechanical inconsistencies have been avoided by considering a poroelastic modeling of the fluid seepage induced by the roughness of the contacting wall (see, e.g., [1,7]). Yet, very little is known on the mathematical foundations of these modeling approaches.

    In this work, we investigate the capability to encompass contact of the FSI model with seepage reported in [7]. For this purpose, we consider a simplified 2D setting of a rigid disk immersed in a Stokesian flow and falling over an horizontal plane (the contacting wall), modeled as a porous layer. We provide a well-posedness analysis for the fluid problem and describe the asymptotics with respect to the porous layer parameters. By building on the arguments reported in [12], we also derive an estimate of the fluid drag force, acting on the disk boundary, with respect to the gap between the disk and the porous layer. A salient feature of this analysis is that it shows that the considered FSI model with seepage allows for contact between the disk and the wall. In other words, since the porous layer allows for seepage, the incompressibility constraint does not create any singularity which prevents contact. To the best of our knowledge, this is the first time in which contact is allowed in a FSI model with Dirichlet interface conditions on the falling disk, for the considered geometrical setting. From the analysis reported in [12,17], one can indeed show that the combination of Dirichlet and Navier boundary conditions prevents contact. The mathematical analysis of the paper is complemented by a comprehensive numerical study which illustrates the theoretical results obtained.

    The rest of the paper is organized as follows: Section 2 presents the considered 2D simplified setting and the fluid-structure-contact interaction model of [7]. An appropriate velocity scaling of the problem (see, e.g., [12]) reduces the coupled problem to the evaluation of the drag force operator in terms of the distance to the contacting wall. Section 3 represents the theoretical core of the paper. It provides a mathematical and asymptotic analysis of the resulting scaled fluid system and analytical estimations of the drag force, which allow to describe the contact dynamics of the disk. Numerical evidence on these theoretical results is provided in Section 4. Finally, the main results of the paper are summarized in Section 5 together with some perspectives of future work.

    In order to investigate the effect of the surface porous modeling of the contact wall, we consider a simplified fluid-structure interaction system involving a quasi-steady Stokes flow with an immersed rigid disk. This simplified setting has already been investigated in previous studies (see, e.g., [12,17,20]), but with a different treatment of the contact wall (notably in terms of boundary conditions).

    The Stokesian fluid is assumed to be contained in a rectangular domain Ωdef=(l,l)×(0,˜l) and the current configuration of the immersed rigid disk is denoted by S(t)Ω, for all t>0. We can hence introduce the fluid domain Ω(t)def=ΩS(t) and its associated non-cylindrical trajectory

    Tdef=tR+Ω(t)×{t}.

    The boundary of Ω is partitioned as Ω=ΓΣ, where Σ denotes the bottom contacting wall (see Figure 1). In what follows, the rigid disk is assumed to be of radius one and to move only vertically (without rotation), so that S(t) can be defined as follows:

    S(t)def=B((d(t)+1)e2,1),

    where (e1,e2) denotes the canonical basis of R2, B(x,R) the disk of radius R centered in xR2, and d:R+R stands for the so called gap function, viz., the quantity d(t) represents the (signed) distance function between S(t) and the contacting wall Σ. Owing to the geometrical symmetry of the problem, it is worth noting that this purely vertical motion can be (physically) expected under appropriate initial conditions.

    Figure 1.  Geometrical description.

    The fluid. The state of the fluid can be described in terms of its velocity u:TR2 and pressure p:TR fields, which are governed by the following Stokes system

    {divσ(u,p)=0inΩ(t),divu=0inΩ(t),u=0onΓ, (2.1)

    where the fluid Cauchy stress tensor σ reads

    σ(u,p)def=2D(u)pI,D(u)def=12(u+uT).

    The fluid viscosity is assumed to be one, and no-slip boundary conditions are considered on the exterior boundary Γ.

    The solid. Owing to the vertical motion of the rigid disk, the dynamics of the solid is simply given in terms of the momentum conservation law. Assuming that the solid has unit mass, the following relation holds

    ¨d(t)+F(t)=0, (2.2)

    for tR+ and where F(t) stands for the external force acting on the solid.

    Fluid-solid coupling. The fluid and solid equations (2.1) and (2.2) have to be coupled with standard geometric, kinematic and dynamic interface conditions on S(t), namely:

    {S(t)=B((d(t)+1)e2,1),Ω(t)=ΩS(t),u=˙de2onS(t),F(t)=S(t)σ(u,p)ne2, (2.3)

    where n denotes the exterior unit normal to Ω(t). Note that the last expression simply relates F(t) to the fluid drag force exerted on the rigid disk at time t.

    Surface roughness model on the contact wall. Traditionally, fluid-structure interaction models based on (2.1)–(2.3) involve a no-slip boundary condition on the contact wall Σ, which is known to prevent contact (see, e.g., [17]) in contrast to what is physically observed. In order to circumvent this issue, we consider the alternative modeling approach reported in [7], which consists in taking into account the roughness of the contact wall through a thin-walled porous model of fluid seepage.

    The matrix of the porous layer is assumed to be rigid and thin-walled, with thickness εp>0 around its mid-surface Σ. We denote by τ the tangential vector on the surface Σ; here τ is simply equal to e1. The normal and tangential conductivities of the porous medium are denoted by κ and η, respectively. The averaged porous fluid pressure ˆp:Σ×R+R can be described by the following surface Darcy model:

    {divτ(εpητˆp)=uΣonΣ,εpητˆpτ=0onΣ, (2.4)

    where uΣ denotes the seepage velocity. This model is derived in [7,25] by averaging the original Darcy system across the thickness and by applying a trapezoidal rule to approximate the integrals involving the interstitial pressure and the normal velocity across the thickness. Note that (2.4)1 enforces mass conservation over the porous layer Σ. In particular, from (2.4), it follows that the net seepage velocity vanishes, viz., ΣuΣ=0.

    Remark 1. As Σ is a 1D line, the symbols divτ and τ, denoting respectively the divergence and gradient along the direction τ, are here both equivalent to x.

    The above system has to be coupled with the bulk fluid equation (2.1) on Σ, via appropriate kinematic and dynamic relations, namely,

    {uΣ=unonΣ,σ(u,p)nn=(ˆp+εp4κun)onΣ,σ(u,p)nτ=0onΣ. (2.5)

    The first relation simply enforces that the seepage velocity in the porous layer has to be the normal component of the bulk fluid velocity. The second corresponds to the balance at the interface of normal stress between the bulk and porous fluids. In particular, the quantity ˆp+εp4κun represents the reconstruction of the interfacial porous pressure from the averaged porous pressure ˆp. Finally, free slip tangential stress is enforced with the last relation.

    Remark 2. It should be noted that, for the sake of simplicity, a pure slip condition has been considered in (2.5)3. In practice, it is however more appropriate to consider the so-called Beavers-Joseph-Saffman (BJS) law:

    σ(u,p)nτ+αεpηuτ=0onΣ, (2.6)

    where the coefficient α0 depends on the structure of the porous layer (see, e.g., [3,26,27]). As indicated in Remark 11, the analysis below can be adapted straightforwardly to the case in which (2.5)3 is replaced by the BJS condition (2.6) and, in particular, it does not modify the collision result of Section 3.3.1.

    Fully coupled problem. The considered coupled problem can be summarized as follows: Find the gap d:R+R, the fluid velocity u:TR2, the fluid pressure p:TR and the porous pressure ˆp:Σ×R+R such that:

    {divσ(u,p)=0 in Ω(t),(2.7a)divu=0 in Ω(t),(2.7b)u=0 on Γ,(2.7c)u=˙de2 on S(t),(2.7d)σ(u,p)nn=(ˆp+εp4κun) on Σ,(2.7e)σ(u,p)nτ=0 on Σ,(2.7f)divτ(εpητˆp)=un on Σ,(2.7g)εpητˆpτ=0 on Σ,(2.7h)¨d(t)+F(t)=0,F(t)=S(t)σ(u,p)ne2,(2.7i)S(t)=B((d(t)+1)e2,1),Ω(t)=ΩS(t),(2.7j)

    with d(0)=d0 and ˙d(0)=˙d0 as initial conditions.

    Velocity scaling. System (2.7) is a highly non-linear coupled problem, notably due to the fact that the fluid domain Ω(t) is unknown. In order to mitigate this difficulty, we adopt here a scaling argument, considered for instance in [12], which leverages the linearity of problem (2.7) with respect to ˙d and basically consists in performing the following change of variables for t>0:

    {u(t)=˙d(t)ud(t)inΩd(t),p(t)=˙d(t)pd(t)inΩd(t),ˆp(t)=˙d(t)ˆpd(t)inΩd(t). (2.8)

    By inserting the above relations into (2.7) and since the equations are linear for a given location of the rigid disk, we infer that, for a given gap d(0,˜l2) and associated solid and fluid domains

    Sddef=B((d+1)e2,1),Ωddef=ΩSd,

    the triplet ud:ΩdR2, pd:ΩdR, ˆpd:ΣR is solution of the following pure (steady) fluid coupled problem:

    {divσ(ud,pd)=0 in Ωd,(2.9a)divud=0 in Ωd,(2.9b)ud=0 on Γ,(2.9c)ud=e2 on Sd,(2.9d)σ(ud,pd)nn=(ˆpd+εp4κudn) on Σ,(2.9e)σ(ud,pd)nτ=0 on Σ,(2.9f)divτ(εpητˆpd)=udn on Σ,(2.9g)τˆpdτ=0 on Σ.(2.9h)

    Note that the velocity scaling (2.8) yields a constant unit velocity as kinematic constraint in (2.9d), and hence independent of ˙d. The analysis reported in Section 3.1 below shows that the operator d(0,˜l2)(ud,pd,ˆpd) is well defined, so that the change of variables (2.8) is feasible.

    By inserting the scaling (2.8) into (2.7i), it follows by linearity that

    F(t)=˙d(t)Fd(t),

    with the notation

    Fddef=Sdσ(ud,pd)ne2 (2.10)

    for any given d(0,˜l2). As a result, the rigid disk dynamics can be rewritten as:

    ¨d(t)+˙d(t)Fd(t)=0. (2.11)

    We will show in Section 3.3 that Fd0, so that from (2.11) we can infer that the fluid-solid coupling (2.7) acts as a friction force into the dynamics of the rigid disk. The amount of this friction is nothing but the scaled drag force (2.10), which only depends on the instantaneous gap d, trough the solution of the pure fluid problem (2.9). The analysis of the system (2.9) and, more specifically, of the behavior of its associated drag force Fd with respect to d provides a practical way to estimate the disk dynamics near the wall and, in particular, to conclude on the possibility of contact between the rigid disk and the bottom boundary Σ.

    Previous studies (see, e.g., [12,13,17,19,29]) have shown that Dirichlet or Navier boundary conditions on the contact wall Σ lead to a singularity of Fd when d0. In 2D with Dirichlet boundary conditions on both the wall and the disk or a Navier boundary conditions on one side and a Dirichlet boundary conditions on the other side, the singularity of Fd is

    Fdd0d32

    and it prevents collision, as proven in Theorem 1 of the seminal paper [17]. If Navier boundary conditions are considered on both the wall and the disk, the singularity is weaker (see [29, Proposition 6.1]) and allows collision:

    Fdd0d12.

    In this paper, we investigate the case in which Dirichlet boundary conditions are applied on the disk and a porous layer is considered on the bottom wall. To the best of our knowledge, this new setting has not yet been investigated from a mathematical analysis perspective. In the next section, we prove existence and uniqueness of a weak solution to the coupled system (2.9). Then we study the asymptotic behavior of the system with respect to the porous layer parameters κ and η. Finally, we provide an estimate of the drag force when d0 and show that

    0FdC(κ,η),

    from which we conclude that the porous layer enables contact in the 2D case.

    This section gathers the main theoretical results of the paper. We provide a thorough mathematical analysis of system (2.9), whose main distinctive feature lies in the coupling with the porous layer on the bottom wall Σ. As, to our knowledge, this system has not yet been studied, we address its well-posedness in Section 3.1, using standard arguments from saddle point problems theory. Then, in order to highlight the role of the normal and tangential conductivity parameters κ and η, we provide an asymptotic analysis of (2.9) with respect to these parameters in Section 3.2. Finally, in Section 3.3 we investigate the influence of the porous layer on the disk contact dynamics, by estimating analytically the drag force acting on the disk.

    In this section, the existence and uniqueness of a weak solution to system (2.9) is investigated. We also derive a priori estimates satisfied by this weak solution.

    Let ω be a given bounded domain of R2 and let Υ denote a subset of its boundary. We denote by L20(ω) the space of L2(ω) functions with zero mean value in ω and by H10,Υ(ω) the space of functions in H1(ω) with zero trace on Υ. The scalar product in L2(ω) is denoted by (,)0,ω, the L2(ω) norm by 0,ω and the H1(ω) norm by 1,ω. We recall that the H1 semi-norm ||1,ω is a norm in H10,Υ(ω), since the Poincaré inequality holds true for functions in H10,Υ(ω). On Υ, we also consider the Hs norms, denoted by s,Υ, for every 0<s1, and the H1 semi-norm, denoted by ||1,Υ.

    Since only the gradient of pd and of ˆpd is involved in (2.9a) and (2.9g), respectively, both pd and ˆpd are in principle defined up to a constant. Nonetheless, owing to (2.9e), once one of the constants is fixed, the other is uniquely defined, therefore, only one of them needs to be fixed to guarantee uniqueness. In the analysis below, we choose to fix the constant of the Darcy pressure ˆpd by considering the space

    Ddef=H1(Σ)L20(Σ), (3.1)

    so that the fluid pressure belongs to L2(Ωd). As regards the fluid velocity, we denote by Ud and U0d the following spaces,

    Uddef={uH10,Γ(Ωd)2: u=e2 on Sd},U0ddef=H10,ΓSd(Ωd)2.

    In order to derive a weak formulation for problem (2.9), we test the fluid momentum equation (2.9a) with vU0d, the incompressibility condition (2.9b) with qL2(Ωd) and the surface Darcy equation (2.9g) with ˆqD. The weak formulation then follows by integrating by parts and using the different boundary and coupling conditions: Find (ud,ˆpd,pd)Ud×D×L2(Ωd) such that

    2(D(ud),D(v))0,Ωd+εp4κ(udn,vn)0,Σ+εpη(τˆpd,τˆq)0,Σ          +(ˆpd,vn)0,Σ(udn,ˆq)0,Σ(pd,divv)0,Ωd+(divud,q)0,Ωd=0 (3.2)

    for all (v,ˆq,q)U0d×D×L2(Ωd).

    The main results of this section are stated in the following theorem.

    Theorem 1. For any d(0,˜l2), problem (3.2) admits a unique solution (ud,pd,ˆpd)Ud×L2(Ωd)×D. Moreover, there exist three positive constants C1,C2,C3>0, which only depend on Ωd, such that (ud,pd,ˆpd) satisfies the following a priori estimates

    D(ud)20,Ωd+εp4κudn20,Σ+εpητˆpd20,ΣC1e2212,Sd, (3.3)
    ˆpd1,ΣC2ε32pκηe212,Sd, (3.4)
    pd0,ΩdC3(1+εp4κ+1εpη)e212,Sd. (3.5)

    Proof. The sketch of the proof is as follows. We first introduce a lifting of the non-homogeneous Dirichlet boundary condition (2.9d) and we derive the associated homogeneous weak formulation. Then, we establish an inf-sup condition in order to prove existence and uniqueness of a solution to problem (3.2). The end of the proof deals with the derivation of estimates (3.3)–(3.5). For the sake of clarity, we drop the d subscript of all variables in the remaining of the proof, since d>0 is fixed. The steps of the proof are detailed hereafter.

    Homogeneous problem. We first consider a lifting of the non-homogeneous boundary condition (2.9d). Since Sde2n=0, it follows from Theorem 5.1 and Remark 5.3 of [14] that there exists a unique (w,q)H1(Ωd)×L20(Ωd), solution of the Stokes systems (2.9a) and (2.9b) with the following Dirichlet boundary conditions:

    {w=e2onSd,w=0onΣΓ. (3.6)

    Moreover, there exists a constant C1>0 which only depends on Ωd such that

    w1,ΩdC1e21/2,Sd. (3.7)

    Therefore, we define for every uUd the new velocity ˉuU0d given by

    ˉu=uw. (3.8)

    Rewriting (3.2) with u=ˉu+w, we obtain that the triplet ˉu:ΩdR2, p:ΩdR, ˆp:ΣR is solution of the following homogeneous problem: Find (ˉu,ˆp,p)U0d×D×L2(Ωd) such that

    2(D(ˉu),D(v))0,Ωd+εp4κ(ˉun,vn)0,Σ+εpη(τˆp,τˆq)0,Σ+(ˆp,vn)0,Σ(ˉun,ˆq)0,Σ(p,divv)0,Ωd+(divˉu,q)0,Ωd=2(D(w),D(v))0,Ωd (3.9)

    for all (v,ˆq,q)U0d×D×L2(Ωd). Problems (3.9) and (3.2) are obviously equivalent. We will then prove existence and uniqueness of a solution (ˉu,ˆp,p) to problem (3.9) and deduce that problem (3.2) is also well-posed.

    For the sake of simplicity, we rewrite the weak formulation (3.9) with usual notations for saddle point problem so the reader can easily find his way through the standard arguments to prove existence and uniqueness. We consider the Hilbert spaces

    Xdef=U0d×D,Mdef=L2(Ωd).

    Let a:X×XR, b:X×MR and lX be the bi-linear and linear forms defined by

    a((u,ˆp),(v,ˆq))def=2(D(u),D(v))0,Ωd+εp4κ(un,vn)0,Σ+εpη(τˆp,τˆq)0,Σ+(ˆp,vn)0,Σ(un,ˆq)0,Σ,b((v,ˆq),q)def=(q,divv)0,Ωd,l,(v,ˆq)X,Xdef=(D(w),D(v))0,Ωd. (3.10)

    Problem (3.9) can therefore be reformulated in an abstract fashion as:

    Find((ˉu,ˆp),p)X×M such that{a((ˉu,ˆp),(v,ˆq))+b((v,ˆq),pd)=l,(v,ˆq)X,X(v,ˆq)X,b((ˉu,ˆp),q)=0qM. (3.11)

    The form l is continuous on X, the form b is continuous on X×M, by continuity of the divergence operator and, using the continuity of the trace operator, we show that a is continuous on X×X. Moreover, using Körn's inequality (see, e.g., [10, Theorem 3.77]), it follows that the form a is coercive on X×X.

    Inf-sup condition. To prove that problem (3.11) is well-posed, it remains to show that the continuous bilinear form b satisfies the usual inf-sup condition. We first observe that

    sup(v,ˆq)U0d×D{(0,0)}(divv,q)(v,ˆq)U0d×Dq0,ΩdsupvU0d{0}(divv,q)v1,Ωdq0,Ωd,

    for all qL2(Ωd){0}. From Theorem 3 of [9] and by adapting Bogovskii's lemma stated in [5] to the case of homogeneous boundary condition on part of the boundary, there exists a constant C>0 such that for any qL2(Ωd) there exists wqU0d such that divwq=q and wq1,ΩdCq0,Ωd. As a result

    supvU0d{0}(divv,q)v1,Ωd(divwq,q)wq1,Ωd=q20,Ωdwq1,Ωd1Cq0,Ωd,

    for all qL2(Ωd){0}, which yields the inf-sup condition

    infqL2(Ωd){0}sup(v,ˆq)U0d×D{(0,0)}(divv,q)(v,ˆq)U0d×Dq0,Ωd1C. (3.12)

    Therefore, since the bilinear form a is continuous and coercive on X×X, that l is continuous on X and that b is continuous on X×M and satisfies the inf-sup condition (3.12), we conclude that problem (3.11) admits a unique solution in X×M (see, e.g., [14, Corollary 4.1]). Moreover, it also proves that problem (3.2) admits a unique solution in Ud×D×L2(Ωd).

    A priori estimates. To obtain a priori estimates on u and ˆp, we first consider (ˉu,ˆp)X as a test function in the weak formulation (3.9), which yields

    2D(ˉu)20,Ωd+εp4κˉun20,Σ+εpητˆp20,Σ=2(D(w),D(ˉu))0,Ωd.

    Thus, from (3.6) and (3.8), we obtain

    2D(u)20,Ωd+εp4κun20,Σ+εpητˆp20,Σ=2(D(w),D(u))0,Ωd.

    Finally, by using the Cauchy-Schwarz and Young inequalities, we get

    2D(u)20,Ωd+εp4κun20,Σ+εpητˆp20,ΣD(w)20,Ωd+D(u)20,Ωd. (3.13)

    The a priori estimate (3.3) then simply follows from (3.13) and (3.7).

    Thanks to (3.3), we can derive a sharper estimate on ˆpd. Indeed using the variational formulation (3.2) with v=0 and ˆq=ˆpd, we obtain

    εpητˆpd20,Σ=(udn,ˆpd)0,Σ,

    so that, by using Cauchy-Schwarz and Poincaré-Wirtinger inequalities we get

    εpηˆpd1,ΣCudn0,Σ,

    with C>0. From (3.3), we get

    udn0,Σ4κC1εpe212,Sd,

    from which we finally deduce estimate (3.4) with C2=2CC1.

    Then, to estimate the fluid pressure p, we test the weak formulation (3.9) with ˆq=0 and q=0, which yields

    (p,divv)0,Ωd=2(D(ˉu),D(v))0,Ωd+2(D(w),D(v))0,Ωd+εp4κ(ˉun,vn)0,Σ+(ˆp,vn)0,Σ,

    for all vU0d. Therefore, owing to (3.6) and (3.8), we have

    (p,divv)0,Ωd=2(D(u),D(v))0,Ωd+εp4κ(un,vn)0,Σ+(ˆp,vn)0,Σ, (3.14)

    for all vU0d. By combining this expression with the inf-sup condition (3.12), we get the following pressure estimate

    p0,ΩdCsup(v,ˆq)U0d×D{(0,0)}2(D(u),D(v))0,Ωd+εp4κ(un,vn)0,Σ+(ˆp,vn)0,Σ(v,ˆq)U0d×D,

    so that, using Cauchy-Schwarz inequality, the continuity of the trace operator and that (v,ˆq)U0d×DvU0d, we finally have

    p20,Ωd˜C(|u|1,Ωd+εp4κun0,Σ+ˆp0,Σ),

    with ˜C>0 a positive constant. Estimate (3.5) then follows from (3.3) and (3.4), which concludes the proof of Theorem 1.

    Estimates (3.3)–(3.5) will be essential for the asymptotic analysis (with respect to κ and η) conducted in Section 3.2. On one hand, we can infer from (3.3) that the fluid velocity ud is bounded in H1(Ωd), independently of κ and η. On the other hand, we loose control on the fluid pressure in (3.5) when either κ or η vanish. The purpose of the next proposition is to show that the zero-mean part of the fluid pressure can be bounded independently of κ and η. We hence decompose the fluid pressure pd=pd+cd, using the direct sum L2=L20R, so that

    cd=1|Ωd|Ωdpd,pd=pdcdL20(Ωd).

    Proposition 1. For any d(0,˜L2), there exists a positive constant C4>0, independent of the conductivity parameters, such that the zero-mean part pd of the fluid pressure pd, solution to system (2.9), satisfies the a priori estimate

    pd0,ΩdC4. (3.15)

    Proof. The proof is very similar to the proof of estimate (3.5). We use the notations introduced in the proof of Theorem 1 and, as before, we drop the subscript d of all variables for readability.

    We consider a test function vU0d which also satisfies vn=0 on the bottom wall Σ. Thus, the constant c does not play any role in the weak formulation (3.9) since

    (c,divv)0,Ωd=cΩddivv=cΩdvn=0.

    The weak formulation (3.2) with ˆq=0 and q=0 therefore gives

    (p,divv)0,Ωd=2(D(ˉu),D(v))0,Ωd (3.16)

    for all

    vU0d{vH1(Ωd)2, vn=0 on Σ}.

    According to Bogovskii's Lemma [5], there exists C4>0 such that for all pL20(Ωd) there exists ˜vH10(Ωd) such that div˜v=p and

    |˜v|1,ΩdC4p0,Ωd.

    Taking v=˜v in (3.16), applying the Cauchy-Schwarz inequality and using estimate (3.3) yields

    p20,Ωd=2(D(ˉu),D(˜v))0,Ωd|ˉu|1,Ωd|˜v|1,ΩdC4p0,Ωd.

    This completes the proof.

    Remark 3. Now, to bring insight on the fluid pressure constant cd, let us note that the weak formulation (3.2) gives the following link between ˆpd, pd and cd

    σ(ud,pd+cd)nn,vn(H1200(Σ)),H1200(Σ)=(ˆpd+εp4κudn,vn)0,Σ

    for all vU0d, where H1200(Σ) denotes the Lions-Magenes space and (H1200(Σ)) its dual (see Remark 4 below). Thanks to the H1 regularity of ˆpd and the H12 regularity of the trace of udn and using the Hahn-Banach theorem and density arguments, we infer that

    σ(ud,pd+cd)nn=(ˆpd+εp4κudn)inL2(Σ). (3.17)

    We can get rid of the Darcy pressure term by integrating this relation over Σ and by using that ˆpdL20(Ωd) and Σudn=0, which yields Σσ(ud,pd)nn2Lcd=Σˆpdεp4κΣudn=0, so that the pressure constant is given by

    cd=12LΣσ(ud,pd)nn. (3.18)

    Note that it is not obvious to obtain an a priori estimate on cd with respect to the data of the problem.

    Remark 4. The space H1200(Σ) can be characterized as the trace on Σ of functions belonging to H10,Γ(Ω) (see [24,28]). It is hence a sub-space of H12(Σ) but with further regularity. In particular, for ξH1200(Σ) we have that x(l,l)ξ(x)x+l belongs to L2(Σ).

    In this section, we use the a priori estimates of Theorem 1 to study the asymptotic behavior of system (2.9) with respect to κ or η. These results are stated in Theorem 2 and summarized in Table 1 below.

    Table 1.  Asymptotic behavior of the solution (ud,pd,ˆpd) of system (2.9), with respect to the normal and tangential conductivity parameters κ and η. We denote by pd def =pd1|Ωd|Ωdpd the zero mean value of pd.
    0 cst.
    0 Navier Navier Navier
    udH1uNad udH1uNad udH1uNad
    pdL2pNad pdL2pNad pdL2pNa
    ˆpdH10 ˆpdH10
    cst. Navier Robin
    udH1uNad Darcy udH1uRd
    pdL2pNad pdL2pRd
    ˆpd1ΩdΩdpd(H1/200)σNann ˆpdH10
    Navier Darcy no cor. Neumann
    udH1uNad udH1uDwcd udH1uNed
    pdL2pNad pdL2pDwcd pdL2pNed
    ˆpd1ΩdΩdpd(H1/200)σNadnn ˆpdH1ˆpDwcd ˆpdH10

     | Show Table
    DownLoad: CSV

    We first introduce the different boundary conditions on Σ that can be obtained as a limit of problem (2.9). We look for the fluid velocity ud:ΩdR2 and the fluid pressure pd:ΩdR solution of (2.9a)–(2.9d), that we recall here,

    {divσ(ud,pd)=0inΩd,divud=0inΩd,ud=0onΓ,ud=e2onS,

    supplemented with either one of the following boundary conditions on the bottom wall Σ:

    ● Navier boundary conditions:

    {udn=0onΣ,σ(ud,pd)nτ=0onΣ. (3.19)

    ● Neumann boundary conditions:

    σ(ud,pd)n=0onΣ. (3.20)

    ● Robin boundary conditions:

    {σ(ud,pd)nn=εp4κudnonΣ,σ(ud,pd)nτ=0onΣ. (3.21)

    ● Darcy boundary conditions without correction term

    {σ(ud,pd)nn=ˆpdonΣ,σ(ud,pd)nτ=0onΣ,divτ(εpητˆpd)=udnonΣ,τˆpdτ=0onΣ. (3.22)

    In what follows, we shall denote the weak solutions of systems (2.9a)–(2.9d) with boundary conditions of type Navier (3.19), Neumann (3.20), Robin (3.21) or Darcy without correction (3.22) by

    (uNad,pNad)U0d×L20(Ωd),(uNed,pNed)U0d×L2(Ωd),(uRd,pRd)U0d×L2(Ωd),(uDwcd,ˆpDwcd,pDwcd)U0d×D×L2(Ωd),

    respectively. The Navier, Neumann and Robin boundary conditions are classical in different fluid-structure interaction settings. Existence and uniqueness of the associated weak solutions can be established using standard arguments. For the coupling with the Darcy without correction (3.22), the well-posedness of the solution can be proved by using the same arguments as for system (2.9) in Section 3.1.

    In what follows, we shall make use of the following functional spaces involving Navier boundary conditions

    UNaddef={uH1(Ωd)2: u=e2  on Sd, u=0  on Γ, un=0  on Σ},U0,Naddef={uH1(Ωd)2: u=0  on SdΓ, un=0  on Σ}

    and the divergence-free subspace Vddef={uH1(Ωd)2: divu=0}.

    The following result provides the asymptotic behavior of system (2.9) with respect to the parameters of the Darcy layer.

    Theorem 2 (Asymptotic analysis). Let d>0 be given and let (ud,ˆpd,pd)Ud×D×L2(Ωd) be the solution to (2.9) which depends on the conductivity parameters κ and η. The following propositions hold true:

    (i) When either κ or η goes to zero, (ud,pd1|Ωd|Ωdpd) weakly converges towards (uNad,pNad) in H1(Ωd)×L2(Ωg) solution to (2.9a)–(2.9d) with Navier boundary conditions (3.19) on Σ;

    (ii) When both κ and η go to infinity, (ud,pd) weakly converges towards (uNed,pNed) in H1(Ωd)2×L2(Ωg) solution to (2.9a)–(2.9d) with Neumann boundary conditions (3.20) on Σ;

    (iii) When κ fixed and η, (ud,pd) weakly converges towards (uRd,pRd) in H1(Ωd)2×L2(Ωg) solution to (2.9a)–(2.9d) with Robin boundary conditions (3.21) on Σ;

    (iv) When κ and η fixed, (ud,pd,ˆpd) weakly converges towards (uDwcd,pDwcd,ˆpDwcd) in H1(Ωd)2×L2(Ωg)×H1(Σ) solution to (2.9a)–(2.9d) coupled with Darcy without correction system (3.22) on Σ.

    Remark 5. Theorem 2 mainly focuses on the asymptotic behavior of the pure fluid part of (2.9). Some asymptotic results for ˆp in the limit cases mentioned above are discussed in Remark 6. In particular, it is worth mentioning that, in the different cases studied in Theorem 2 (except in case (iv)), the fluid unknowns (u,p) and the porous pressure ˆp are not coupled anymore.

    Proof. We first gather the cases which converge towards the Navier boundary conditions (κ or η goes to 0) and then derive the other cases. This distinction is motivated by estimates (3.3) and (3.5), in which some control on the solution is lost when either κ or η goes to 0, so that the passage to the limit is more involved than in the other cases.

    In this proof, we assume the gap distance d is fixed and we drop the d subscript in all variables. Instead, we introduce the notations uκ,η, pκ,η and ˆpκ,η that make the dependence on the conductivity parameters explicit. Thanks to a priori estimate (3.3) on uκ,η, we know there exists a subsequence that converges weakly in H1(Ωd)2. By abuse of notation, we also denote by uκ,η the subsequence and by ul its limit:

    uκ,ηH1κ or η  0ul.

    By continuity of the trace and of the divergence operators, we have ulVdUd. We now consider the different cases.

    Proof of (i). When κ or η goes to 0, to show that uln=0, we have to consider two cases. When κ0, a priori estimate (3.3) gives straightforwardly uln=0 in L2(Σ). When η0, taking the weak formulation (3.2) with v=0 yields

    (uκ,ηn,ˆq)0,Σ=εpη(εpητˆpκ,η,τˆq)0,Σ,ˆqD.

    Owing to (3.3), we have εpηˆpκ,η1,ΣC, so that we can pass to limit in the previous identity and obtain

    (uln,ˆq)0,Σ=0,ˆqD,

    and, by the density of H1(Σ) in L2(Σ), we have

    (uln,ˆq)0,Σ=0,ˆqL20(Σ),

    which implies that uln is constant on Σ. Morever, since ulnH1200(Σ) (see Remark 4), we recover uln=0 also in this case. We therefore have that ulVdUNad either when κ or η goes to zero.

    Testing (3.2) with vU0,NadVd and both ˆq=0 and q=0, yields

    (2D(uκ,η),D(v))=0,vU0,NadVd,

    so that, passing to the limit,

    (2D(ul),D(v))=0,vU0,NadVd.

    By uniqueness of the weak solution associated to the Navier problem (2.9a)–(2.9d) with boundary conditions (3.19), we obtain that ul=uNa. Furthermore, thanks to the sequential characterization of the limit, we conclude that

    uκ,ηH1κ or η  0uNa. (3.23)

    For the pressure, we consider its zero mean part

    pκ,ηdef=pκ,η1|Ωd|Ωdpκ,η,

    which, according to Proposition 1, is bounded independently of κ and η. We then have a subsequence (uκ,η,pκ,η) in Ud×L20(Ωg) which weakly converges towards (uNa,pl). Testing (3.2) with vU0,Nad and both ˆq=0 and q=0, yields

    2(D(uκ,η),D(v))0,Ωd(pκ,η,divv)0,Ωd=0,vU0,Nad,

    which, passing to the limit, gives

    2(D(uNa),D(v))0,Ωd(pl,divv)0,Ωd=0,vU0,Nad.

    We hence have pl=pNa and

    pκ,ηL2κ or η  0pNa. (3.24)

    This concludes the proof of (i).

    Proof of (ii) and (iii). When neither η nor κ vanish, we have control on uκ,η, pκ,η and ˆpκ,η thanks to the estimates (3.3) and (3.5). When η, from (3.3) we get that

    ˆpκ,ηH1η0.

    So, passing to the limit in the weak formulation (3.2) with ˆq=0, gives

    (2D(ul),D(v))0,Ωd+εp4κ(uln,vn)0,Σ(pl,divv)0,Ωd+(divul,q)0,Ωd=0,(v,q)U0d×L2(Ωd),

    which is exactly the weak formulation associated to Robin boundary conditions (3.21). For κ fixed, we therefore recover, when η, the solution to the problem with Robin boundary condition, which is unique, i.e.,

    uκ,ηH1κ fixed, ηuR,pκ,ηL2κ fixed, ηpR.

    When both η and κ, we have that the term εp4κ(uκ,ηn,vn)0,Σ disappears, so that we retrieve the Neumann boundary conditions at the limit and have

    uκ,ηH1κ, ηuNe,pκ,ηL2κ, ηpNe.

    Proof of (iv). For η fixed and κ, we can pass to the limit in the weak formulation (3.2) and note that only the correction term εp4κ(uκ,ηn,vn)0,Σ vanishes. One can easily prove that this limit problem admits a unique solution corresponding to the Darcy without correction, using the same arguments than for the proof of Theorem 1. This concludes the proof.

    Remark 6. The following asymptotic behaviors for the Darcy pressure can also be proved:

    ˆpdH1η 0,ˆpdH1κ0, η fixed0,ˆpdH1κη00,ˆpd1|Ωd|Ωdpd(H1/200)κ fixed or , η0σ(uNad,pNad)nn.

    Indeed, from (3.3), we have |ˆpd|21,ΣC1εpη, so that ˆpd strongly convergences to 0 in H1(Σ) when η. When κ0 and η is fixed, from estimate (3.4) we straightforwardly deduce that

    ˆpdH1κ0, η fixed0.

    When η0 and κ is fixed or goes to , we loose the control on ˆpd provided by (3.3). Nevertheless, from (3.2) with ˆq=0, we have

    (ˆpd,vn)0,Σ=2(D(ud),D(v))0,Ωdεp4κ(udn,vn)0,Ωd+(pd,divv)0,Ωd

    for all vU0d. By splitting the fluid pressure in terms of its zero mean part, pd=pd+cd, we have

    (ˆpdcd,vn)0,Σ=2(D(ud),D(v))0,Ωdεp4κ(udn,vn)0,Ωd+(pd,divv)0,Ωd (3.25)

    for all vU0d. As κ does not vanish, we can pass to the limit in this expression using the weak convergences (3.23) and (3.24), so that

    limη0(ˆpdcd,vn)0,Σ=2(D(uNad),D(v))0,Ωd+(pNad,div(v))0,Ωd,vU0d.

    By density arguments, we conclude that

    ˆpdcd(H1/200)η0, κ fixed or σ(uNad,pNad)nn.

    Finally, when κ and η both go to 0, we cannot pass to the limit in (3.25), but we can conclude in some particular cases. Indeed, from (3.4), we have

    ˆpd1,ΣCκη.

    When κ converges faster than η towards 0, the Darcy pressure ˆpd converges strongly in H1 towards 0. If η and κ converge at the same speed, ˆpd is bounded in H1 and therefore admits a subsequence that weakly converges in H1 towards an indeterminate limit. Nothing can be said when η converges faster than κ towards 0.

    We conclude this section by summarizing all the obtained convergence results in Table 1. It should be noted that, in principle, the sole cases which are physically relevant are those where the conductivity parameters are small. It is interesting to note also that we always recover the solution of the Stokes equations with Navier boundary conditions on the bottom wall when either κ or η goes to 0. This is in agreement with the following physical intuition: reducing κ or η makes the fluid seepage through the porous medium more difficult, as un converges weakly towards zero. In terms of the contact dynamics of the disk moving towards the bottom wall, this means that in the limit cases where either κ or η goes to 0, the drag force acting on it is expected to behave like the one with the Navier boundary conditions on the bottom wall. The next section is devoted to the estimation of this drag force.

    This section is devoted to the main result of the paper: modeling seepage with a porous layer on the bottom wall, as in (2.7), enables contact even with Dirichlet boundary conditions on the disk. The proof of this result is based on an estimate (Theorem 3 below) of the asymptotic behavior of the drag force (2.10) acting on the disk as it gets closer to the wall (i.e., d0). To this purpose, we build on the approach reported in [12,17]. For Navier or Dirichlet boundary conditions, similar arguments have previously shown that the drag force becomes singular as d0 (see, e.g., [17,29] in 2D and [13,19] in 3D). In some cases, such as Dirichlet boundary conditions on both the disk and the wall or the combination of Navier and Dirichlet boundary conditions, the singularity in the drag force scales as d32 in 2D. This prevents the disk to reach the wall in finite time, which is known as the no-collision paradox.

    The next result states that the drag force associated to the scaled fluid problem (2.9) is bounded as d0.

    Theorem 3 (Drag force estimate). The drag force Fd given by (2.10) is nonnegative and bounded from above independently of d. More precisely, we have the following estimate

    |Fd|=Od0(1+6i=11(εpη)i+εp4κ4i=21(εpη)i). (3.26)

    Before proceeding with the proof of Theorem 3, we can already combine estimate (3.26) with the disk dynamics equation (2.11) to conclude that system (2.7) enables contact between the disk and the bottom wall. This is the purpose of the next corollary.

    Corollary 1 (Collision result). Assume that Eq (2.11), with initial conditions ˙d(0)=˙d0<0 and d(0)=d0>0, admits a solution dC2(0,tc), where tcR+{+} denotes the time at which the disk hits the bottom wall. There exists a constant C>0, depending only on κ and η, such that for ˙d0<Cd0 we have tc<+ with d(tc)=0 and ˙d(tc)<0.

    Proof. We first recall that the disk dynamics associated to the FSI system (2.7) are given by the ordinary differential equation (2.11), namely,

    ¨d+Fd˙d=0,

    as long as d remains positive. Owing to Theorem 3, there exists a constant C(κ,η)>0 such that, for all d>0, the drag force satisfies

    FdC(κ,η). (3.27)

    The drag force is also positive Fd0. This is a direct consequence of the result, shown in the next section (see (3.34) and (3.35)), that the drag force is the minimum of an energy.

    Assume that, at time t=0, the disk is located at position x0=d0e2 and has the initial velocity ˙xG=˙d0e2 with ˙d0<0. In other words, it falls towards the bottom wall. By continuity, the velocity ˙d is still negative for a short time interval 0tt1, so that from (2.11) and (3.27) we get

    ¨d=Fd(˙d)0¨dC(κ,η)˙d (3.28)

    for all 0tt1, and thus

    ˙d(t)˙d0eCt. (3.29)

    This implies that, for all t>0 such that d(t)>0, the velocity ˙d(t) remains negative and (3.29) holds true. Integrating over time gives for any t>0 such that d(t)>0, we have

    d(t)d0+˙d0t0eCsdsd(t)d0+˙d0C(1eCt)def=f(t).

    The right-hand side function f is obviously decreasing and its limit when t is given by

    limtf(t)=d0+˙d0C.

    If ˙d0<Cd0, there exists a time tc>0 such that f(tc)=0, namely,

    tc=ln(1+C(κ,η)d0˙d0)C(κ,η). (3.30)

    By continuity, we show that the disk touches the wall at a time ttc, with ˙d(t)<0. This completes the proof.

    Remark 7. Note that in the current model nothing prevents the disk of penetrating the bottom wall. Indeed, if the initial velocity |˙d0| is large enough, the distance d can vanish with a negative velocity. To avoid this, one often adds a non-penetration constraint in the fluid-structure interaction model (2.7). Usually, this constraint is imposed by duality with a Lagrange multiplier λ that represents the upward force acting on the disk at contact. The resulting problem writes

    {¨d+Fd˙d=λ,d0,λ0,λd=0,

    where the last three conditions represent the so-called non-penetration, compressibility and complementary conditions.

    The next sections are devoted to the proof of Theorem 3. The first idea consists in expressing the drag force Fd as the minimum of the energy of the system. This is the purpose of Section 3.3.2. Then, in Section 3.3.3, we construct an admissible minimizer which catches the contact dynamics. Finally, in Section 3.3.4, we estimate the energy associated to the minimizer which yields the bound (3.26). Since the proof builds on [17], we put particular emphasis on the main difficulties arising in the present model with the porous layer.

    Let Ad be the admissible function space for the fluid velocity defined by

    Addef=UdVd={uH1Γ(Ωd)2:divu=0 in Ωd,u=e2 on Sd}. (3.31)

    In this section, we express the Darcy pressure ˆp in terms of the fluid velocity via the following operator.

    Definition 1 (Operator A) For any fL20(Σ) the problem

    Find ˆpD such thatεpηΣτˆpτˆq=Σfˆq,ˆqD, (3.32)

    is well-posed. Therefore, we denote by A:L20(Σ)D the operator that associates, to any f in L20(Σ), the unique solution ˆp of problem (3.32): Af=ˆp.

    In particular, for all vAd, vn|ΣL20(Σ) as v is divergence free. We have that A(vn) is the solution of the following strong form of problem (3.32) with f=vn:

    {divτ(εpητˆp)=vn on  Σ,(3.33a)τˆp(l)=0,τˆp(l)=0.(3.33b)

    The following proposition is the cornerstone of our approach to prove Theorem 3.

    Proposition 2 (Energy minimization problem). We introduce the energy functional Ed:AdR+ defined by

     Ed(u)def=2ΩdD(u)2+εpηΣ|τA(un)|2+εp4κΣ|un|2 (3.34)

    for all uAd. The drag force (2.10) satisfies

    Fd=minuAdEd(u). (3.35)

    Moreover, the energy minimization problem (3.35) admits a unique solution in Ad, ud, given by (3.2).

    Proof. We first prove that

    Fd=Ed(ud). (3.36)

    Owing to the weak formulation (3.2), we have

    div(σ(ud,pd))=0inD(Ωd),

    which by a standard density argument and thanks to Hahn-Banach Theorem leads to

    div(σ(ud,pd))=0inL2(Ωd).

    Since div(σ(ud,pd))L2(Ωd), the trace of σ(ud,pd)n can be defined by duality as follows:

    σ(ud,pd)n,ξH12(Ωd),H12(Ωd)=2(D(ud),D(Lξ))0,Ωd(pd,div(Lξ))0,Ωd

    for all ξH12(Ωd), where the symbol L denotes a standard lifting operator from H12(Σ)2 to H1(Ωd)2. Taking Lξ=ud and using that udUdVd and the free sliding condition in (2.5), we get

    σ(ud,pd)n,udH12(ΣΓ),H12(ΣΓ)+σ(ud,pd)n,e2H12(Sd),H12(Sd)=2D(ud)20,Ωd, (3.37)

    where the drag force appears in the term on Sd. It is standard to check that σ(ud,pd)nL2(Sd). Indeed, the C regularity of the disk and the Dirichlet boundary condition on Sd guarantee H2 regularity for the fluid velocity ud and H1 regularity for the pressure pd near the ball (see, e.g., [6]), so that we have

    Fd=σ(ud,pd)n,e2H12(Sd),H12(Sd)=Sdσ(ud,pd)ne2. (3.38)

    As regards the first term in (3.37), we use the relation (3.17), the free sliding condition (2.5)3 and the fact that ud|Γ=0 to get

    σ(ud,pd)n,udH12(ΣΓ),H12(ΣΓ)=σ(ud,pd)n,ud(H1200(Σ)),H1200(Σ)=(ˆpd,udn)0,Σεp4κ(udn,udn)0,Σ.

    Since by construction ˆpd=A(udn), taking ˆq=A(udn) in (3.32) yields

    (ˆpd,udn)0,Σ=(A(udn),udn)0,Σ=εpηΣ|τA(udn)|2,

    so that

    σ(ud,pd)n,udH12(ΓΣ),H12(ΓΣ)=εpηΣ|τA(udn)|2εp4κΣ|udn|2. (3.39)

    Finally, by inserting (3.38) and (3.39) into (3.37), we obtain

    Fd=Sdσ(ud,pd)ne2=Ωd2D(ud)2+ηεpΣ|τA(udn)|2+εp4κΣ|udn|2=Ed(ud),

    which yields (3.36).

    In order to show that ud is a minimizer of the functional Ed, we introduce the tangent Hilbert space to Ad, namely:

    A0ddef={uH1(Ωd)2,divu=0 in Ωd,u=0 on SdΓ}. (3.40)

    Straightforward computations yield

    Ed(ud+v)=Ed(ud)+2(D(ud),D(v))0,Ωd+εp4κ(udn,vn)0,Σ+εpη(τA(udn),τA(vn))0,Σ+Ed(v),vA0d. (3.41)

    On the other hand, from the definition of A (3.32), the following symmetry relation holds

    (A(udn),vn)0,Σ=(τA(udn),τA(vn))0,Σ=(udn,A(vn))0,Σ,vA0d.

    Therefore, since ˆpd=A(udn) and using the definition of the bi-linear form a in (3.10), we have that

    a((ud,A(udn)),(v,A(vn)))=2(D(ud),D(v))0,Ωd+εp4κ(udn,vn)0,Σ+εpη(τA(udn),τA(vn))0,Σ,vA0d,

    and, by inserting this identity into (3.41), we get

    Ed(ud+v)=Ed(ud)+a((ud,A(udn)),(v,A(vn)))+Ed(v),vA0d. (3.42)

    Yet, testing (3.2) with vA0dU0d, q=0 and ˆq=A(vn), gives

    a((ud,A(udn)),(v,A(vn)))=0,vA0d,

    so that, from (3.42), we finally have

    Ed(ud+v)=Ed(ud)+Ed(v),vA0d,

    which, since Ed(v)0, yields (3.35).

    Finally, to prove the uniqueness of the minimizer, we can use the equivalence

    minvAdEd(v)minvA0dJd(v),Jd(v)def=Ed(ud+v),

    and evaluate the second Fréchet differential of Jd to show that

    D2Jd(w)(v,v)=a((v,A(vn)),(v,A(vn)))>α|v|21,Ωd,

    for all non zero vA0d and wA0d, where the last inequality follows from the coercivity of the bilinear form a given in (3.10). This guarantees the strict convexity of Jd, which implies the uniqueness of the solution of the minimization problem (3.35) and, hence, completes the proof.

    In order to ease the computations, we rewrite the minimization problem, in an equivalent form, in terms of stream functions. This is the purpose of the next result.

    Lemma 1 (Stream function). The admissible space Ad, defined in (3.31), is also given by

    Ad={uH1(Ωd):!φH2(Ωd),u=2φe1+1φe2,1φ|S=1,2φ|S=0,1φ|Γ=0,2φ|Γ=0,φ(l,0)=0}. (3.43)

    Moreover, we have

    Ed(u)=Ed(u(φ))=Ωd(4|212φ|2+|21φ|2+|22φ|2222φ21φ)+1εpηΣ|φ|2+εp4κΣ|1φ|2 (3.44)

    for all uAd.

    Proof. We start by proving (3.43). Given uAd, u is a 2D divergence-free vector field so there exist a streamfunction φH2(Ωd), unique up to an additive constant, such that u=2φe1+1φe2 (see, e.g., [14, Theorem 3.1]). In (3.43), this constant is fixed by imposing φ(l,0)=0. The reasons of this choice will be made clear later in the proof. One can easily check that u|Sd=e2,u|Γ=0, enforce

    1φ|Sd=1,2φ|Sd=0,1φ|Γ=0,2φ|Γ=0. (3.45)

    Conversely, given φH2(Ωd), if φ satisfies (3.45) then obviously u=2φe1+1φe2 belongs to Ad.

    In order to get (3.44), we simply replace u by its associated stream-function φ in (3.34). Computations are straightforward for the bulk term D(u) and the last boundary term. For the term involving τA(un) we proceed as follows. The Darcy pressure ˆp=A(un) satisfies (3.33) which gives

    {divτ(εpητˆp(x))=u(x,0)e2=1φ(x,0)in(l,l),τˆp(l)=0. (3.46)

    Integrating the first relation yields

    εpη(τˆp(x)τˆp(l))=xl1φ(s,0)ds=φ(x,0)φ(l,0)in(l,l).

    On the other hand, using (3.46)2, we have

    εpητˆp=φφ(l,0)inΣ. (3.47)

    As a result, since from (3.43) we have φ(l,0)=0, we get

    εpηΣ|τA(un)|2=εpηΣ|φ|2(εpη)2=1εpηΣ|φ|2,

    which completes the proof.

    In the section, the energy functional is split into two contributions: one part leads to a relaxed problem, that we can solve analytically, while the remaining terms will be bounded in Section 3.3.4. In order to both catch the contact dynamics and to be able to simplify the computations, two main choices are made to relax problem (3.35).

    First, we limit the analysis to the following region under the disk

    Ωd,δdef={(x,y)Ωd:|x|<δ,0<y<d+γ(x)}, (3.48)

    for all d>0 and 0<δ<1, and where γ:x[δ,δ]11x2 is a parametrization of a subset of the boundary of the disk (see Figure 2). We also consider the notations

    Sd,δdef=Sd¯Ωd,δ={(x,d+γ(x)), x(δ,δ)},Σδdef=Σ¯Ωd,δ=(δ,δ).
    Figure 2.  Contact zone Ωd,δ.

    Second, we introduce the relaxed energy functional ˜Ed given by

    ˜Ed(u)def=Ωd,δ|2u1|2+εpηδδ|τ(Au2)|2=Ωd,δ|22φ|2+1εpηδδ|φ(x,0)|2. (3.49)

    Note that only the contribution 2u1=22φ has been kept in the volumetric energy term. This is motivated by physical intuition and previous studies (see, e.g., [12,17,29]) that show that, in the case of Dirichlet or Navier boundary conditions, the term 2u1=22φ causes the no-collision paradox. Indeed, high velocity gradients arise when the fluid has to escape tangentially in the narrow space between the disk and the bottom wall. The terms involving u2=1φ has been removed to avoid having to deal with terms on the side boundaries Ωd,δ(Sd,δΣδ). It should be noted that, all the left out terms will be estimated in Section 3.3.4.

    Similarly, we introduce the relaxed admissible function space ˜Ad as the restriction of functions of Ad to the contact zone Ωd,δ, viz.,

    ˜Addef={u|Ωd,δ:uAd}. (3.50)

    In order to give a precise characterization of ~Ad, we need the following two technical lemmas.

    Lemma 2 (Symmetry of the solution). Solution of (2.9) satisfies the following symmetries:

    u1(x,y)=u1(x,y),u2(x,y)=u2(x,y),p(x,y)=p(x,y),ˆp(x)=ˆp(x).

    Proof. The proof is straightforward thanks to the symmetries of the problem (see Figure 3) and the uniqueness of the solution of (2.9).

    Figure 3.  Symmetries of the problem.

    The next lemma gives the characterization of ~Ad.

    Lemma 3. Let ˜Ad be the relaxed space given by (3.50). There holds

    ˜Ad={uH1(Ωd,δ):!φH2(Ωd,δ),u=2φe1+1φe2,φ|Sd,δ=x,2φ|Sd,δ=0}. (3.51)

    Proof. Let u=2φe1+1φe2Ad. The boundary conditions u|S=e2 implies

    1φ|Sd,δ=1,2φ|Sd,δ=0,

    from which we deduce

    ddxφ(x,d+γ(x))=1φ(x,d+γ(x))+γ(x)2φ(x,d+γ(x))=1in(δ,δ),

    so that

    φ|Sd,δ=x+c,2φ|Sd,δ=0.

    As we already fixed the constant by imposing φ(l,0)=0 in (3.43), the constant c cannot be arbitrarily chosen. We have c=φ(0,d). Owing to Lemma 2, by a symmetry argument on u, we have u1(0,y)=0 in (0,d), so that

    0=d02φ(0,y)dyφ(0,d)=φ(0,0)=c.

    As ˆp is also symmetric, and thanks to (3.47), we have

    0=τˆp(0)=1εpη(φ(0,0)φ(l,0))=1εpηφ(0,0),

    which gives c=0. As a result, we finally recover the following boundary conditions on Sd,δ

    φ|Sd,δ=x,2φ|Sd,δ=0.

    For the boundary conditions on Ωd,δ(ΣδSd,δ), we remark that for all δ<1 the following restriction function on Ωd,δ

    {φH2(Ωd),1φ|Γ=2φ|Γ=0}H2(Ωd,δ)φφ|Ωd,δ

    is surjective (see for example the extension ˆφ in Section 3.3.4), that is why the boundary conditions on Ωd,δ(ΣδSd,δ) do not appear in the relaxed space definition (3.51). This completes the proof.

    The relaxed minimization problem reads therefore as follows: Find

    ˜ud=2˜φde1+1˜φde2˜Ad

    such that

    ˜Ed(˜ud)=min˜v˜Ad˜Ed(˜v). (3.52)

    We finally define the relaxed drag force as the minimum of this relaxed energy, namely,

    ˜Fddef=˜Ed(˜ud). (3.53)

    The next proposition solves explicitly the relaxed minimization problem (3.52).

    Proposition 3 (Explicit expression of ˜φd). The relaxed minimization problem (3.52) admits a unique solution ˜ud=2˜φde1+1˜φde2˜Ad, given by

    ˜φd(x,y)=xϕd(x,yd+γ(x)), (3.54)

    with

    αd(x)def=(d+γ(x))3εpη,ϕd(x,t)def=αd(x)t36+2αd(x)+3αd(x)t6+2αd(x)+66+2αd(x). (3.55)

    Proof. As in the proof of Proposition 2, we can rewrite (3.52) as a minimization problem in the tangent Hilbert space ˜A0d:

    ˜A0ddef={uH1(Ωd,δ):!φH2(Ωd),u=2φe1+1φe2,φ|Sd,δ=0,2φ|Sd,δ=0}.

    So, let u0˜Ad be given, instead of (3.52), we consider the auxiliary equivalent problem

    minv˜A0d˜Jd(v),˜Jd(v)def=˜Ed(v+u0). (3.56)

    Let us denote D2˜Jd(u) the second Fréchet differential of ˜Jd in u. For all u˜A0d, we have

    D2˜Jd(u)(v,v)=2(Ωd,δ|2v1|2+εpηδδ|τ(Av2)|2)0,v˜A0d.

    Let v˜A0d and ψ its associated streamfunction. The relation D2˜Jd(u)(v,v)=0 implies that

    {22ψ=0inΩd,δ,ψ=0onΣδ.

    Therefore, there exists a function g such that ψ(x,y)=g(x)y in Ωd,δ. Since ψ(x,d+γ(x))=0, we conclude that ψ=0, so that v=0. As a result, D2˜Jd(u)(v,v)>0 for all v˜A0d{0}. This guarantees the strict convexity of ˜Jd, so that, if a minimum exists for problem (3.56), it is unique.

    The minimum ˜ud˜Ad of (3.52) can be characterized by the Euler equations associated to the auxiliary minimization problem (3.56), which writes DJd(˜udu0)v=0 for all v˜A0d, or equivalently

    Ωd,δ22˜φd22ψ+δδ1εpη˜φdψ=0ψB,

    where B simply denotes the stream-function space associated to ˜A0d, namely,

    Bdef={φH2(Ωd,δ),φ|Sd,δ=0,2φ|Sd,δ=0}.

    After two integrations by parts, we obtain

    Ωd,δ42˜φdψ+δδ(32˜φd+1εpη˜φ)ψδδ22˜φd2ψ=0,ψB. (3.57)

    From (3.51) and (3.57) we obtain, using standard density and trace arguments, the following PDE system for ˜φd

    {42˜φd=0inΩd,δ,32˜φd+1εpη˜φd=0onΣδ,22˜φd=0onΣδ,2˜φd=0onSd,δ,˜φd=xonSd,δ. (3.58)

    whose solution is given by (3.54). Indeed, the streamfunction ˜φd defined by (3.54) is a polynomial of 3rd degree in y, so that it satisfies (3.58)1. For the second relation in (3.58), as

    32˜φd=x(d+γ(x))332ϕd,

    one can check that 32˜φd(x,0)+1εpη˜φd(x,0)=6x6+2αd(x)(αd(x)(d+γ(x))3+1εpη)=0. The rest of the computations are straightforward. The solution (3.54) therefore characterizes the unique minimizer ˜ud of (3.52). This completes the proof.

    The next theorem shows that the relaxed drag force ˜Fd is bounded irrespectively of d.

    Theorem 4. The relaxed drag force (3.53) is given by

    ˜Fd=3δδαd(x)3+αd(x)x2(d+γ(x))3dx. (3.59)

    In particular, we have that

    ˜Fd=Od0(1εpη). (3.60)

    Proof. In order to prove (3.59), we combine the definition (3.53) of the relaxed drag force ˜Fd with the explicit expression of the minimizer given in terms of the stream function ˜φd in (3.54). Straightforward computations then yield

    ˜Fd=Ωd,δ|22˜φd|2+1εpηδδ|˜φd(x,0)|2dx=δδ36αd(x)2(6+2αd(x))2x2(d+γ(x))6(d+γ(x)0y2dy)dx+δδ1εpη36(6+2αd(x))2x2dx=δδ12αd(x)2(6+2αd(x))2x2(d+γ(x))3dx+δδ36αd(x)(6+2αd(x))2x2(d+γ(x))3dx=3δδαd(x)2+3αd(x)(3+αd(x))2x2(d+γ(x))3dx=3δδαd(x)(3+αd(x))x2(d+γ(x))3dx.

    Now, by inserting into (3.59) the expression for αd given in (3.55), we get

    ˜Fd=3δδfd(x)x2dx,fd(x)def=13εpη+(d+γ(x))3.

    Since limd0fd(x)=1/(3εpη+γ(x)3), the dominated convergence theorem gives

    limd0˜Fd=3δδx23εpη+γ(x)3dx,

    which is clearly bounded when εpη is fixed. Indeed, as γ(x)0 for any x[δ,δ], we have

    limd0˜Fd3δδx23εpη=2δ33εpη,

    which yields (3.60) and completes the proof.

    We conclude this section with a series of remarks.

    Remark 8 (Darcy versus Navier or Dirichlet). A salient difference of the porous surface model considered in this paper with respect to Navier or Dirichlet boundary conditions lies in the fact that the fluid can escape vertically, i.e., un can be non-zero on Σ, so that 1φ can play an important role in the estimation of the drag force. On the contrary, the work reported in [12] only focuses on the horizontal velocity of the fluid. Note that we neglect the impact of κ in the relaxed energy (3.49) in order to facilitate the solution of (3.52) (Proposition 3). However, the dependance on κ is retrieved in the bound of the full drag force Fd, provided in the next section. It should be noted we do not have a natural lower bound on the drag force because the cross derivatives 22φ21φ cannot be simplified in the energy functional (3.44), so that we do not have ˜FdFd as in the Navier or Dirichlet case (see [12]).

    Remark 9 (Asymptotic of ˜Fd for εpη=O(dp), p3). If εpη=O(dp) with p3, we obtain an asymptotic behavior for the relaxed drag force (3.59) similar to the case with Dirichlet boundary conditions on both the disk and the wall

    ˜Fdd32. (3.61)

    Indeed, by using εpη=O(dp), p3, in (3.55) we infer that αd1. Since f:xxx+3 is continuous in [1,+) and limxf(x)=1, we have

    cf(αd)=αd3+αdC,

    from which we deduce

    3cδδx2(d+γ(x))3dx˜Fd=˜Ed(˜φ)3Cδδx2(d+γ(x))3dx.

    We therefore recover Dirichlet asymptotic behavior as δδx2(d+γ(x))3dxd32. We recall that the idea to retrieve this behavior consists in combining the expansion of γ(x)=x22+Ox0(x4) with the change of variables u=vd to exhibit d32 out of the integral (see [12] for the no-slip case).

    Note that the obtained asymptotic (3.61) of the relaxed drag force is consistent with the asymptotic analysis carried out at the Section 3.2. Indeed, as stated in Theorem 2, when η tends to zero the solution of problem (2.9) converges weakly towards solution of a problem with Navier boundary conditions on the wall and Dirichlet boundary conditions on the disk. So, for εpη=O(dp), p3 we expect to recover (3.61) as the asymptotic of the drag force associated to the Navier-Dirichlet boundary conditions, which is known to behave like the Dirichlet-Dirichlet case (see, e.g., [13]). Yet, the proved asymptotic behavior (3.61) only concerns the relaxed drag force ˜Fd. Nothing can be said on the asymptotic of the whole drag force due to the lack of lower bound on Fd (see Remark 8).

    Remark 10 (Asymptotic of ˜Fd for εpη=O(dp), 0<p<3). In this case, it can be shown that

    ˜Fdd0dp2.

    by combining the Taylor expansion γ(x)=x22+Ox0(x4) with the change of variable t=x/dp6. This asymptotic behavior could be useful in a more general setting, in which the compressibility of the porous matrix can induce a change in the conductivity parameters. For instance, when the ball gets closer to the wall, it compresses the porous matrix and thus might reduce its conductivity.

    The main idea to complete the proof consists in extending ˜φd to the whole domain Ωd in order to obtain an admissible velocity ˇudAd. To obtain a bound on Fd, we have to estimate the whole energy Ed(ˆud) and, in particular, all the terms left out above in the introduction of the relaxed energy functional ˜Ed. We consider the same extension as in [17] and use similar notations for the auxiliary functions.

    Proof. We consider the following extension of ˜φd (see [17]):

    ˇφd(x,y)def={χ2δ(1+|x|)˜φd(x,y)+[1χ2δ(1+|x|)]ˆφd(x,y),(x,y)Ωd,2δ,ˆφd(x,y), otherwise, (3.62)

    where the auxiliary function χ2δ is defined as

    χδ(x)def={1ifx1+δ2,1fh(x(1+δ2)(1+δ)(1+δ2))if1+δ2<x<1+δ,0ifx1+δ, (3.63)

    with

    h(x)def=1e111x,f(x)def=ex1x,

    and the remaining auxiliary function ˆφd is given by

    ˆφd(x,y)def=xχδ(xxd), (3.64)

    with xddef=(0,1+d) denoting the center of the disk. The auxiliary function χδ has a smooth C2 decreasing transition between 1+δ2 and 1+δ. Figure 4 illustrates how ˇφd is contructed by combining ˆφd and ˜φd. The different zones have been depicted in Figure 5 for illustration purposes.

    Figure 4.  Plots of ˆφd, ˜φd et ˇφd for d=0.5 with δ=0.4, η=0.1 et εp=0.01.
    Figure 5.  Definition of the extension ˇφd.

    Let us introduce the function

    ˇud=2ˇφde1+2ˇφde2.

    We first check that ˇud is admissible, i.e., that ˇudAd, with Ad the admissible space defined in (3.31). As χδ belongs to C((1,1+δ)) by construction, the function ˆφd also belongs to C(Ωd). The extension ˇφd has therefore the same H2-regularity than ˜φd, so that ˇud belongs to H1(Ωd). Then, one can easily check that ˇud=e2 on Sd.

    Finally, as the support of ˇφd is strictly included in Ωd, we do have ˇud=0 on the boundary Ωd(ΣSd). According to Proposition 2, since ˇudAd, we have the following bound on Fd

    FdEd(ˇud). (3.65)

    Thus in order to derive (3.26), we estimate the following energy

    Ed(ˇud(ˇφd))=Ωd2D(ˇud(ˇφd)):D(ˇud(ˇφd))+1εpηΣ|ˇφd|2+εp4κΣ|1ˇφd|2=˜Ed(˜φd)+Ωd,δ[2D(˜ud(˜φd)):D(~ud(˜φd))|22˜φd|2]+εp4κΣΩd,δ|1˜φd|2+Ωd,2δΩd,δ2D(ˇud(ˇφd)):D(ˇud(ˇφd))+1εpηΣΩd,2δΩd,δ|ˇφd|2+εp4κΣΩd,2δΩd,δ|1ˇφd|2+ΩdΩd,2δ2D(ˆud(ˆφd)):D(ˆud(ˆφd)).

    We notice that the first term corresponds to the relaxed drag force, already estimated in Theorem 4. We proceed by bounding all the remaining terms.

    Under the disk. We consider the following energy:

    Ed(˜φd)def=Ωd,δ(2D(˜φd):D(˜φd)|22˜φ2d|)+εp4κδδ|1˜φd|2=Ωd,δ(|21˜φd|2+4|12˜φd|2221˜φd22˜φd)+εp4κδδ|1˜φd|2.

    We then estimate each term separately. For the first term, using Proposition 3, we have

    21˜φd(x,y)=1εpηf(x)y3+3εpηg(x)y+f(x),

    with αd defined in (3.55) and the auxiliary functions

    f(x)def=x6+2αd(x),g(x)def=x(d+γ(x))26+2αd(x).

    Since d<˜l, the derivatives of αd can be bounded for all x[δ,δ] as

    |αd(x)|(˜l+1)3εpη,|αd(x)|3(˜l+1)2εpηγ(δ),|αd(x)|C(˜l)εpη(γ(δ)+γ(δ)).

    We can show that if δ<1 there exist strictly positive constants C1 and C2 depending only on ˜l and δ such that

    |f(x)|C1(˜l,δ)(1εpη+1(εpη)2),|g(x)|C2(˜l,δ)(1+1εpη+1(εpη)2).

    We finally obtain the estimate of the first term

    Ωd,δ|21˜φ(x,y)|2C3(˜l,δ)(εpη)2(1+1εpη+1(εpη)2+1(εpη)3+1(εpη)4).

    Similarly, we can obtain the following estimates for the other terms

    Ωd,δ|12˜φ(x,y)|2δC4(˜l,δ)(εpη)2(1+1εpη+1(εpη)2),Ωd,δ|12˜φ(x,y)|2δC5(˜l,δ)(εpη)2(1+1εpη+1(εpη)2),εp4κΣΩd,δ|1˜φ(x,0)|2εp4κδC6(˜l,δ)(εpη)2(1+1εpη+1(εpη)2).

    Finally, we obtain the following estimate

    |Ed(˜φ)|C7(˜l,δ)(εpη)2(1+1εpη+1(εpη)2+1(εpη)3+1(εpη)4)+εp4κC8(˜l,δ)(εpη)2(1+1εpη+1(εpη)2). (3.66)

    It is worth noting that if κ or η vanishes we loose the upper bound.

    Away from the disk ΩdΩd,2δ. Outside of Ωd,2δ, ˇφd=ˆφd and ˆφd depends on the gap distance d only through a translation (see [17]HY__HY, Proof of Lemma 8]). Thus, energy in ΩΩd,2δ of ˇud=ˇφd does not depend on d and is therefore bounded. So, there exists a constant C9(δ)>0 such that

    ΩdΩd,2δ2D(ˇud):D(ˇud)C9(δ) (3.67)

    for d<˜l.

    Transitional zone Ωd,2δΩd,δ. Since |χδ|1, using Young's inequality, we have

    ˇφd=χδ˜φd+(1χδ)ˆφd|ˇφd||˜φd|+|ˆφd||ˇφd|2C10(|˜φd|2+|ˆφd|2).

    So, the estimate of the transitional term can simply be obtained by combing the bounds (3.66) and (3.67).

    Finally, (3.26) then follows by combining the previous bounds with (3.60), which concludes the proof of Theorem 3.

    Remark 11. The analysis above can be adapted straightforwardly to the case in which the pure slip condition (2.5)3 is replaced by the BJS law (2.6). Indeed, in this case it suffices to add the following tangential friction term

    αεpηδδ|2φ|2,

    to the relaxed energy (3.49). By building on similar arguments than those used for the derivation of (3.59), one can show that the associated drag force can be bounded by C(εpη)1, with C>0 a constant independent of the friction coefficient α. Similarly, the extension contributions can also be bounded irrespectively of α. In particular, this means that the drag force does not blows when α, as uτ0 and the fluid can escape through the porous layer along the normal direction.

    In this section we provide numerical evidence of the theoretical results obtained in the previous sections by performing a series of numerical experiments. To this purpose, we consider finite element approximations of the solution to the static system (2.9), for different values of d (viz., the distance between the disk and the bottom wall), and of κ and η (the normal and tangential conductivity parameters, respectively). We recover numerically the theoretical upper bound of the drag force of Theorem 3, and the weak convergence results stated in Theorem 2 when either κ or η vanishes. Then, we numerically investigate the ability of the porous layer to enable collision as stated Corollary 1. Finally, we provide numerical insight on the validity of the fluid-reduced Darcy model (2.9) by comparing it against the full model.

    In what follows, we consider a disk of radius r=1 centered at position (40,d+r), immersed in a rectangular box of width l=80 and of height ˜l=40, whose bottom-left corner lies at position (0,0). We numerically solve the system (2.9) for different values of the gap d between the disk and the bottom wall, decreasing from 1 to 103. We do the same with system (2.9a)–(2.9d) supplemented with Navier boundary conditions (3.19). The values for d are chosen to be equidistant in log scale

    d{0.001,0.002,0.004,0.008,0.016,0.032,0.064,0.126,0.252,0.502,1}. (4.1)

    This experiment setup is very similar to the one considered in [18]. Therein the authors retrieve the d32 asymptotic behavior of the drag force for the problem with Dirichlet boundary conditions (on both the disk and the bottom wall).

    For a more intuitive visualization and without loss of generality, we choose the disk to move down with unit velocity, instead of going up as originally in system (2.9). One can straightforwardly check that this only changes the sign of the computed drag force, so it does not affect the expected theoretical results.

    For a given gap distance d, let Tfd denote a triangulation of the fluid domain Ωd and let Tpd be the 1D triangulation of the porous mid-surface Σ, given simply as the edges of Tfd lying on Σ. For the Navier problem, we consider the standard space of continuous piecewise polynomial functions of degree two for the fluid velocity, namely,

    UNad,hdef={vhC0(¯Ωd)2:vh|KP2(K)2KTfd,vh=e2on Sd,vh=0onΓ,vhn=0onΣ},

    and the space of continuous piecewise linear functions for the fluid pressure

    Qd,hdef={qhC0(¯Ωd):qh|KP1(K)  KTfd}.

    We also introduce the homogeneous counterpart of UNad,h

    U0,Nad,hdef={vhC0(¯Ωd)2:vh|KP2(K)2KTfd,vh=0 on ΓSd, vhn=0  on Σ}.

    For any d>0, the corresponding finite element approximation of (2.9a)–(2.9d) with the Navier boundary conditions writes

    Find(uh,ph)UNad,h ×Qhsuchthat2(D(uh),D(vh))Ωd+(divuh,qh)Ωd(ph,divvh)Ωd+εs(ph,qh)Ωd=0,(vh,qh)U0,Nad,h×Qd,h. (4.2)

    Here, the small perturbation of the incompressibility constraint εs(ph,qh)Ωd, with εs=1010, serves to enforce zero mean on the discrete pressure ph (this can be inferred by taking (vh,qh)=(0,1) in (4.2) and by integrating by parts in the divergence term). This is a standard approach for fixing the constant of the pressure for Stokes problems with only Dirichlet boundary conditions (see, e.g., [4,16]).

    To formulate the finite element approximation of (3.2), we consider a slightly different trial space for the fluid velocity, without any constraint on Σ,

    Ud,hdef={vhC0(¯Ωd)2: vh|KP2(K)2,KTfd, vh=e2 on Sd,vh=0 on Γ},

    with its homogeneous counterpart

    U0d,hdef={vhC0(¯Ωd)2: vh|KP2(K)2  KTfd, vh=0 on ΓSd}.

    For the Darcy pressure, we choose the space of continuous piecewise polynomial functions of degree 2 on Σ, namely,

    Dd,hdef={ˆqhC0(Σ): ˆqh|KP2(K) KTpd}.

    The finite element approximation of (3.2) finally writes

    Find(uh,ˆph,ph)Ud,h×Dd,h×Qd,h such that 2(D(ud,h),D(vd,h))Ωd+εp4κ(ud,hn,vhn)Σ+(divud,h,qh)Ωd(pd,h,divvh)Ωd+εpη(τˆpd,h,τˆqh)Σ+(ˆph,vhn)Σ(uhn,ˆqh)Σ+ε1sΣˆphΣˆqh=0,(vh,ˆqh,qh)U0d,h×Dd,h×Qd,h. (4.3)

    The last term enforces the mean value of the Darcy pressure ˆph to be 0 by penalization, with again εs=1010.

    For each value of the gap (4.1), a mesh of Ωd has been generated. The mesh size between the disk and the bottom wall depends on the gap d: We have at least ten triangles between the disk and the bottom wall in order to properly catch the fluid dynamics in the contact zone. The mesh size below the disk is defined by

    hmin=min{d10,0.01}.

    The meshes are not structured uniformly to avoid unnecessary refinement outside of the contact zone. Away from the disk, the mesh size is hmax=0.1. In order to ensure a smooth transition between the small mesh elements below the disk and the bigger mesh elements away from the gap, we first build the meshes using the FreeFem++ (see [16]) integrated mesh generator then we optimize them with the Mmg remeshing software (see [8]). The resulting meshes have between 104 and 105 nodes (see Figure 6).

    Figure 6.  Mesh for the smallest gap distance d=103. Zooms on the disk and on the contact zone.

    First, we numerically validate the porous layer model. We then investigate that we can numerically recover the main theoretical results of the paper, that is, the asymptotic analysis in κ and η of Theorem 2 and the estimate of the drag force stated in Theorem 3. Finally, we use the numerical simulations to bring insight on the way the porous layer model influences the fluid and solid dynamics, by reporting the fluid velocity and pressure fields and also by simulating the disk fall towards the wall.

    Validation of the porous layer model. The surface porous model (2.4)-(2.5) is derived in [7] by averaging the original Darcy system across the thickness. In order to investigate the validity of this approximation, we simulate the fall of the disk over the 2D porous layer Ωpdef=(l,l)×(0,εp). The porous fluid velocity in Ωp is given by the Darcy relation updef=Kpp (see, e.g., [2,25]), where the porous fluid pressure pp:ΩpR is described by the divergence free constraint:

    {div(Kpp)=0inΩp,Kppn=0onΩpΣ, (4.4)

    with the conductivity matrix Kdef=diag(κ,η). The coupling between (4.4) and the Stokes system (2.9a)–(2.9d) is then operated by the following kinematic and dynamic interface conditions on Σ:

    {udn=KppnonΣ,σ(ud,pd)n=ppnonΣ.

    Figures 7 and 8 report the velocity and pressure fields obtained in both cases and indicate that the behavior of the solution of the 2D Darcy problem (4.4) is very similar to the solution of the Darcy surface model (2.4) considered in this paper. In both cases, the physical parameters are κ=η=1, εp=0.01 and d=0.001. In Figures 7(b) and 8(b), the porous layer thickness has been amplified by a factor 25 to better visualize the Darcy pressure.

    Figure 7.  Comparison of velocity field with the porous layer model versus 2D Darcy.
    Figure 8.  Comparison of pressure field with the porous layer model versus 2D Darcy.

    Drag force asymptotics. In Figure 9, we report the drag force Fd for d[103,1] arising from system (2.9) where the fluid is coupled to the porous layer model. For the sake of simplicity, we call it the Darcy drag force. We also plot the drag force asymptotic arising from system (2.9a)–(2.9d) with Navier boundary conditions (3.19), that we similarly call Navier drag force asymptotic.

    Figure 9.  Drag force asymptotic when d0 and (κ,η)0.

    We test various set of values for the conductivity parameters κ and η, notably to investigate the behavior when the conductivity parameters gets smaller. Thus, we can compare with the asymptotic results of Theorem 2 and the estimate of the Darcy drag force (3.26) stated in Theorem 3.

    Away from the disk at d=1, the wall does not have any influence on the disk, so there is practically no difference between Navier and Darcy drag force. When the disk gets closer of the wall, the Darcy drag force asymptotic reaches a plateau which is consistent with the result of Theorem 3, whereas for Navier the drag force explodes. The very good recovery of the theoretical -1.5 slope for Navier drag force is clearly visible.

    As expected from the estimate on the drag force (3.26), the plateau value of the Darcy asymptotic increases when κ and/or η goes to 0. We can also note that the conductivity parameters κ and η play a role in the switch from the Navier asymptotic to a plateau, smaller conductivity values leading to a switch to a plateau closer to the wall. The combination of both effects make that the Darcy drag force asymptotic converges towards the Navier one, when η and/or κ goes to 0, which is consistent with the asymptotic analysis results of Theorem 2. It is worth noting that the convergence is faster when the tangential conductivity η vanishes than when the normal one κ vanishes.

    It is interesting to see that the porous layer simply removes the singularity of the drag force arising in the case of Navier boundary conditions. The introduction of a porous modeling of the asperities solves the no-collision paradox by enabling contact, but it also changes the contact dynamics with respect to Navier.

    Navier versus Darcy velocity and pressure fields. In order to better understand the fundamental difference between the Navier and Darcy drag force asymptotics, Figures 10 and 11 show the velocity and pressure fields, respectively obtained with the Navier and Darcy models, when the disk is very close to the wall (d=0.032).

    Figure 10.  Velocity and pressure fields for a gap distance d=0.032 obtained with Navier boundary conditions on the below wall.
    Figure 11.  Velocity and pressure fields for a gap distance d=0.032 and a porous layer model with conductivity parameters κ=1 and η=1 on the bottom wall.

    In the case of the coupling with a porous layer, Figure 11, we can clearly see that the fluid velocity on the wall right below the disk is vertical and of magnitude 1, which yields a plateau of fluid entering into the porous layer below the disk. In this new regime where the disk is in the close vicinity of the wall, all the fluid escapes vertically through the porous layer without any tangential leak. Thus, reducing the gap does not affect the flow of the fluid. On the contrary, with Navier boundary conditions on the wall, Figure 10 indicates that size of the escaping jet can only increase whenever d gets smaller, as only tangential escape is allowed.

    Porous layer dynamics when η0 or κ0. We now focus on the influence of κ and η on the porous layer dynamics. For that purpose, we fix the disk at a tiny distance d=4103 from the wall and we look at the evolution of the incoming fluid flow uhn on Σ and of the Darcy pressure ˆph when κ or η vanishes, see Figures 12 and 13. We zoom around the contact point at x=40.

    Figure 12.  Zoom below the disk of udn and ˆpd when κ0.
    Figure 13.  Zoom below the disk of udn and ˆpd1|Ωd|Ωdpd when η0.

    For κ=η=1, a plateau that stagnates at 1 is observed, which corresponds to the velocity of the disk. When the disk is in the close vicinity of the wall, all the fluid leaks into the porous medium without any tangential escape, as we have already seen in Figure 11. Reducing the conductivity hinders the fluid escape: the plateau first narrows then it shrinks to 0. We recover in both cases κ0 and η0 the convergence of uhn to 0, which is consistent with weak convergence results of Theorem 2.

    It is interesting to highlight the differences between the two cases. Let us look at the behavior of the Darcy pressure. When η0, it becomes more and more singular below the disk, converging towards the Navier normal stresses accordingly to the proved weak convergence of Theorem 2, with a huge peak of the Darcy pressure up to 8103 at η=106, see Figure 13. This singularity makes it difficult to run the simulation for smaller values of η because of numerical instabilities. It also gives numerical evidence on why in Figure 9 the Darcy drag force assymptotic seems to converge much faster towards the Navier one for η0 than κ0. On the contrary, reducing κ makes ˆph flatter, consistently with the H1 strong convergence of ˆph towards 0 when κ0 proved in Theorem 2.

    The difference between the two asymptotics κ0 and η0 is also noticeable if we look at the way the fluid escapes in Figures 12 and 13: in the first asymptotic we observe that the (reentrant) fluid velocity on both sides of the disk tends to flatten as κ tends to 0, while for the second asymptotic, two narrow peaks start to appear before flattening out.

    Disk trajectories with Darcy layer on the wall versus with Navier boundary conditions. The numerical approximations obtained for the drag force, can be combined with the ODE (2.2) to simulate the evolution of the gap distance d(t). We recall that the disk dynamics are given by the relation

    ¨d(t)+˙d(t)Fd(t)=0, (4.5)

    where the drag force depends on the roughness model considered for the bottom wall (here, Navier boundary conditions or the porous layer model).

    In order to approximate d, we first consider a linear interpolation of log(Fd) that we then insert in an explicit time-stepping of (4.5). In the following, we fix the conductivity parameters κ and η to be equal to 1. In Figure 9, we see that the bound of the Darcy drag force is around C=100. We first consider d0=1 and then start a downward motion with |˙d0| going from 1 to 100, see Figure 14. As expected from Corollary 1, we have collision for |˙d0|Cd0.

    Figure 14.  Evolution of the gap distance for a ball starting at distance d0=1 with reduced Darcy on the wall and for various initial condition.

    We now compare the effect of Darcy and Navier on the ball dynamics in Figure 15, which provides the trajectories obtained in both cases for different initial velocities. For ˙d0=1, the trajectories are indistinguishable. The porous layer has no significant effect on the dynamics whenever the disk stays far away from the wall. For ˙d0=30, we see that the disk reaches the wall in the case of the Darcy layer, whereas with Navier boundary conditions the disk maintains a distance d=0.1 from the wall. The porous layer completely modifies the disk dynamics at contact.

    Figure 15.  Comparison of the evolution of the gap distance between Navier and Darcy porous layer for a ball starting at distance d0=1.

    In this paper, we have analyzed the contact capabilities of a fluid-structure interaction model with seepage reported in [7]. The key feature of this model lies in taking into account the surface roughness of the contacting wall in terms of a reduced Darcy model. The analysis shows that this modeling approach removes the no-collision paradox. The contact dynamics are also modified with respect to more standard boundary conditions, such as Navier, since contact is allowed with a non-zero velocity. A non penetration condition must hence be added in order to prevent the solid to go through the contacting wall.

    Extensions of this work can explore several directions. From the mathematical analysis point of view, one interesting question would be the study of the complete fluid-structure-contact interaction model, with appropriate non penetration conditions. Another interesting question would be the formulation of the model in the case of multiple solids getting into contact.

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    MF and MV were partially supported by the French National Research Agency (ANR), through the SIMR project (ANR-19-CE45-0020). The authors thank the anonymous reviewers for their constructive comments that helped to improve the manuscript.

    The authors declare no conflicts of interest.



    [1] C. Ager, B. Schott, A. Vuong, A. Popp, W. A. Wall, A consistent approach for fluid-structure-contact interaction based on a porous flow model for rough surface contact, Int. J. Numer. Methods Eng., 119 (2019), 1345–1378. https://doi.org/10.1002/nme.6094 doi: 10.1002/nme.6094
    [2] M. B. Allen, The mathematics of fluid flow through porous media, John Wiley & Sons, 2021. https://doi.org/10.1002/9781119663881
    [3] G. S. Beavers, D. D. Joseph, Boundary conditions at a naturally permeable wall, J. Fluid Mech., 30 (1967), 197–207. https://doi.org/10.1017/S0022112067001375 doi: 10.1017/S0022112067001375
    [4] M. Bercovier, Perturbation of mixed variational problems. Application to mixed finite element methods, RAIRO. Anal. Numér., 12 (1978), 211–236. https://doi.org/10.1051/m2an/1978120302111 doi: 10.1051/m2an/1978120302111
    [5] M. Bogovskii, Solution of the first boundary value problem for an equation of continuity of an incompressible medium, Soviet Math. Dokl., 20 (1979), 1094–1098.
    [6] F. Boyer, P. Fabrie, Mathematical tools for the study of the incompressible Navier-Stokes equations and related models, Vol. 183, Springer Science & Business Media, 2012.
    [7] E. Burman, M. A. Fernández, S. Frei, F. M. Gerosa, A mechanically consistent model for fluid-structure interactions with contact including seepage, Comput. Methods Appl. Mech. Eng., 392 (2022), 114637. https://doi.org/10.1016/j.cma.2022.114637 doi: 10.1016/j.cma.2022.114637
    [8] C. Dobrzynski, P. Frey, Anisotropic delaunay mesh adaptation for unsteady simulations, In: R. V. Garimella, Proceedings of the 17th international Meshing Roundtable, Springer, 2008,177–194. https://doi.org/10.1007/978-3-540-87921-3_11
    [9] Q. Du, M. D. Gunzburger, L. S. Hou, J. Lee, Analysis of a linear fluid-structure interaction problem, Discrete Cont. Dyn. Syst., 9 (2003), 633–650. https://doi.org/10.3934/dcds.2003.9.633 doi: 10.3934/dcds.2003.9.633
    [10] A. Ern, J. L. Guermond, Theory and practice of finite elements, Vol. 159, Springer, 2004. https://doi.org/10.1007/978-1-4757-4355-5
    [11] D. Gérard-Varet, M. Hillairet, Regularity issues in the problem of fluid structure interaction, Arch. Rational Mech. Anal., 195 (2010), 375–407. https://doi.org/10.1007/s00205-008-0202-9 doi: 10.1007/s00205-008-0202-9
    [12] D. Gérard-Varet, M. Hillairet, Computation of the drag force on a sphere close to a wall: the roughness issue, ESAIM: Math. Modell. Numer. Anal., 46 (2012), 1201–1224. https://doi.org/10.1051/m2an/2012001 doi: 10.1051/m2an/2012001
    [13] D. Gérard-Varet, M. Hillairet, C. Wang, The influence of boundary conditions on the contact problem in a 3D Navier-Stokes flow, J. Math. Pures Appl., 103 (2015), 1–38. https://doi.org/10.1016/j.matpur.2014.03.005 doi: 10.1016/j.matpur.2014.03.005
    [14] V. Girault, P. A. Raviart, Finite element approximation of the Navier-Stokes equations, Vol. 749, Springer Berlin, 1979. https://doi.org/10.1007/BFb0063447
    [15] D. Gérard-Varet, M. Hillairet, Existence of weak solutions up to collision for viscous fluid-solid systems with slip, Commun. Pure Appl. Math., 67 (2014), 2022–2075. https://doi.org/10.1002/cpa.21523 doi: 10.1002/cpa.21523
    [16] F. Hecht, New development in freefem++, J. Numer. Math., 20 (2012), 251–265. https://doi.org/10.1515/jnum-2012-0013
    [17] M. Hillairet, Lack of collision between solid bodies in a 2D incompressible viscous flow, Commun. Partial Differ. Eq., 32 (2007), 1345–1371. https://doi.org/10.1080/03605300601088740 doi: 10.1080/03605300601088740
    [18] M. Hillairet, A. Lozinski, M. Szopos, On discretization in time in simulations of particulate flows, Discrete Cont. Dyn. Syst.-Ser. B, 15 (2010), 935–956. https://doi.org/10.3934/dcdsb.2011.15.935 doi: 10.3934/dcdsb.2011.15.935
    [19] M. Hillairet, T. Takahashi, Collisions in three-dimensional fluid structure interaction problems, SIAM J. Math. Anal., 40 (2009), 2451–2477. https://doi.org/10.1137/080716074 doi: 10.1137/080716074
    [20] M. Hillairet, T. Takahashi, Existence of contacts for the motion of a rigid body into a viscous incompressible fluid with the tresca boundary conditions, Tunis. J. Math., 3 (2021), 447–468. https://doi.org/10.2140/tunis.2021.3.447 doi: 10.2140/tunis.2021.3.447
    [21] M. Hillairet, T. Takahashi, Blow up and grazing collision in viscous fluid solid interaction systems, Ann. Inst. Henri Poincaré C, Anal. non linéaire, 27 (2010), 291–313. https://doi.org/10.1016/j.anihpc.2009.09.007 doi: 10.1016/j.anihpc.2009.09.007
    [22] D. Kamensky, F. Xu, C. H. Lee, J. Yan, Y. Bazilevs, M. C. Hsu, A contact formulation based on a volumetric potential: application to isogeometric simulations of atrioventricular valves, Comput. Methods Appl. Mech. Eng., 330 (2018), 522–546. https://doi.org/10.1016/j.cma.2017.11.007 doi: 10.1016/j.cma.2017.11.007
    [23] N. Khaledian, P. F. Villard, M. O. Berger, Capturing contact in mitral valve dynamic closure with fluid-structure interaction simulation, Int. J. Comput. Assisted Radiol. Surg., 17 (2022), 1391–1398. https://doi.org/10.1007/s11548-022-02674-4 doi: 10.1007/s11548-022-02674-4
    [24] J. L. Lions, E. Magenes, Non-homogeneous boundary value problems and applications: Vol. 1, Vol. 181, Springer Science & Business Media, 2012. https://doi.org/10.1007/978-3-642-65161-8
    [25] V. Martin, J. Jaffré, J. E. Roberts, Modeling fractures and barriers as interfaces for flow in porous media, SIAM J. Sci. Comput., 26 (2005), 1667–1691. https://doi.org/10.1137/S1064827503429363 doi: 10.1137/S1064827503429363
    [26] A. Mikelic, W. Jäger, On the interface boundary condition of Beavers, Joseph, and Saffman, SIAM J. Appl. Math., 60 (2000), 1111–1127. https://doi.org/10.1137/S003613999833678X doi: 10.1137/S003613999833678X
    [27] P. G. Saffman, On the boundary condition at the surface of a porous medium, Stud. Appl. Math., 50 (1971), 93–101. https://doi.org/10.1002/sapm197150293 doi: 10.1002/sapm197150293
    [28] L. Tartar, The Lions-Magenes space H1/200(Ω), In: An introduction to Sobolev spaces and interpolation Spaces, Lecture Notes of the Unione Matematica Italiana, Springer, 3 (2007), 159–161. https://doi.org/10.1007/978-3-540-71483-5_33
    [29] C. Wang, Strong solutions for the fluid-solid systems in a 2-D domain, Asymptotic Anal., 89 (2014), 263–306. https://doi.org/10.3233/ASY-141230 doi: 10.3233/ASY-141230
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1682) PDF downloads(195) Cited by(0)

Figures and Tables

Figures(15)  /  Tables(1)

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog