Introduction
This paper is motivated by our interest in modeling aspects of brain physiology. Poroelasticity can be effective in modeling interstitial fluid (IF) flow through brain parenchyma and its interaction with the cerebrospinal fluid (CSF) circulation. While this paper is not about brain physiology, we briefly touch upon the problem of identifying the pathways and drivers of waste clearance in the brain to provide some context for what we present herein.
The accumulation of potentially neurotoxic waste products, such as amyloid-β \beta β , in the brain extracellular space (ECS) has been linked to the pathogenesis of neurodegenerative disorders like Alzheimer’s disease.[47 ] The brain does not have a conventional lymphatic system1 {}^{1} 1 responsible for clearing products of metabolism and neurotransmission,[41 , 43 ] and the identification of clearance pathways and driving mechanisms remains a poorly understood aspect of brain physiology and pathology.[25 , 26 , 2 ]
The brain and spinal cord are surrounded by CSF, which flows through a system of fluid-filled communicating spaces including the subarachnoid space (SAS), ventricular system, central canal of the spinal cord, and the perivascular spaces (PVS) surrounding cerebral vessels. Transport of waste by CSF flow is acknowledged as a key process in maintaining brain function. Transport of metabolites through parenchyma has traditionally been attributed to Fickian diffusion,[49 , 50 , 52 , 53 ] eventually leading to mixing with CSF and draining into the peripheral lymphatics in the neck.[10 , 11 ] More recently, the work by Iliff, Nedergaard, and co-workers[31 , 29 , 32 , 30 ] has challenged this understanding, pointing to the existence of a directional flow of fluid into and out of the central nervous system (CNS). This circulation has been dubbed the glymphatic system and is characterized by a significant convective transport and exchange between IF and CSF along the PVS. However, several aspects of the glymphatic hypothesis remain controversial[25 , 26 , 2 ] : little is known about the mechanisms and forces driving fluid movement. Various drivers of the glymphatic flow have been proposed in the literature, such as arterial pulsation,[32 , 40 , 7 ] functional hyperemia,[34 , 33 ] respiration,[35 ] and body posture.[36 ]
The transport phenomena at hand are characterized by flow through parenchyma that has both convective and diffusive features. Furthermore, this flow can cross a variety of permeable membranes, such as the glia limitans and the pia mater (cf. Ref. 26). These membranes have different cellular structures and can play different roles, including that of regulating osmotic pressures. To simultaneously account for both diffusive and convective flows, we had adopted,[16 , 33 ] a poroelasticity theory based on mixture theory.[8 , 9 ] In this paper, we focus on interface conditions at permeable interfaces with an eye toward the possibility of explicitly reflecting interfacial properties in our model.
There is a vast literature on the conditions at the interface between a porous medium and a fluid (see, e.g., in Refs. 27, 48, and 18). This literature concerns both modeling aspects (e.g., Refs. 27, 48, and 18) as well as the numerical implementation of such conditions (e.g., Refs. 37, 14, and 45). It is well-known that the jump conditions of the balance laws do not imply or demand that the fluid velocity field or the pore pressure be continuous across the interface.[8 ] It is also well-known that a boundary value problem (BVP) without sufficient constraints on said discontinuities is generally ill-posed (e.g., see discussion in Ref. 27) and such that the ill-posedness has different manifestations depending on the fluid’s constitutive behavior on the two sides of the interface. Therefore, complementing the jump conditions of the balance laws with some additional interface conditions is often essential for the formulation of a well-posed BVP. Provided that several approaches have been proposed in the literature, by and large, the most commonly adopted interface conditions (cf. Ref. 48) are kinematic in nature as they prescribe continuity for the pore pressure field and the tangential component of the filtration velocity. For mixtures of incompressible materials, the continuity of the tangential component of the filtration velocity, along with the requirement that its normal component be continuous as a consequence of the balance of mass, yields a continuity condition on the filtration velocity as a whole, leaving the fluid velocity field discontinuous. Such continuity requirement is fairly straightforward to implement in practice whenever the field equations are written in terms of filtration velocity, as opposed to the fluid velocity. However, this type of formulation may not always be appropriate for the modeling of a particular problem. This said, and much more importantly, prescribing the continuity of certain fields across the interface is tantamount to prescribing an implicit constitutive response to the interface. This idea is brought out by the analysis in Ref. 18 where the authors use Hamilton’s principle2 to derive the problem’s governing equations and interface conditions. In this approach, one must define an action potential and, when dissipation is present, a Rayleigh pseudo-potential. In turn, this requires an explicit model of all energy storage and dissipation mechanisms in the system. As such, the interface constitutive behavior is explicitly modeled and the associated interface conditions are derived by a choice of test functions that appropriately respects the kinematic assumptions underlying the model. In doing so, dell’Isola et al. {sup}18` obtained interface conditions that had not previously appeared in the literature but that transparently translated the constitutive behavior of the interface and its relation to the constitutive behavior of the surrounding materials, regardless of whether they are both poroelastic, or one being poroelastic and the other being pure fluid.
We found the work presented in Ref. 18 to provide the most appropriate setting for our interest. This said, we could not find a numerical implementation of such framework, especially for the case in which the constituents are incompressible in their pure form. Furthermore, to obtain the interface conditions pertaining to the motion of the fluid relative to the solid, the authors of Ref. 18 designed two ad hoc test functions for each of two different components of the interface conditions and distinguishing the cases with and without mass transport across the interface. While we find said construction rather clever, we also find it somewhat difficult to implement it practically within common finite element method (FEM) platforms. With this in mind, the objectives of this work are (i ) to propose an adaptation of the variational approach in Ref. 18 to the case in which each of the mixture constituents is incompressible in its pure form, and (ii) to propose a practical implementation of the approach using a FEM that allows the use of commonly available discontinuous representations of fluid velocity fields.
The paper is organized as follows. First, in § 2 , we present the essential notation and basic kinematics needed to describe our model. In § 3 , we then discuss the relevant balance laws following the traditional approach in Ref. 8 and we derive a corresponding energy estimate. In § 4 , we derive the field equations using Hamilton’s principle and offer a comparison with the previously derived governing equations. In this section we also propose a practical FEM implementation of our equations. In § 5 , we discuss some examples. In § 6 , we conclude the paper with a summary and discussion.
Basic definitions and kinematics
Here we describe the motion of our system. Our treatment is rooted in Refs. 8 and 9 and we adopt the kinematic and analytical framework in Ref. 18, which we adapt to the case of a biphasic mixture of incompressible constituents. For completeness, basic definitions have been placed in Appendix § 7 .
The physical system and its kinematics
We consider a mixture with a solid and a fluid phase. The mixture’s current configuration is B ( t ) ⊂ E 3 \mathcal{B}(t) \subset\mathcal{E}^{3} B ( t ) ⊂ E 3 , while B s ⊂ E 3 \mathcal{B}_{s} \subset\mathcal{E}^{3} B s ⊂ E 3 and B f ⊂ E 3 \mathcal{B}_{f} \subset\mathcal{E}^{3} B f ⊂ E 3 are the reference configurations of the solid and fluid phases, respectively. The symbols x \mathbf{x} x , X s \mathbf{X}_{s} X s , and X f \mathbf{X}_{f} X f denote points in B ( t ) \mathcal{B}(t) B ( t ) , B s \mathcal{B}_{s} B s , and B f \mathcal{B}_{f} B f , respectively. Referring to the notation in Eq. (A.1), E \mathcal{E} E , S \mathcal{S} S , and F \mathcal{F} F indicate B ( t ) \mathcal{B}(t) B ( t ) , B s \mathcal{B}_{s} B s , and B f \mathcal{B}_{f} B f , respectively.
B ( t ) \mathcal{B}(t) B ( t ) is the common image of the solid and fluid motions, respectively
χ s : B s × R 0 + → B ( t ) and χ f : B f × R 0 + → B ( t ) , (1) \boldsymbol{\chi}_{s}:\mathcal{B}_{s} \times\mathbb{R}_{0}^{+} \to\mathcal{B}(t) \quad\text{and} \quad\boldsymbol{\chi}_{f}:\mathcal{B}_{f} \times\mathbb{R}_{0}^{+} \to\mathcal{B}(t), \tag*{(1)} χ s : B s × R 0 + → B ( t ) and χ f : B f × R 0 + → B ( t ) , (1)
which are expected to be diffeomorphisms except, possibly, across a permeable interface S ( t ) ⊂ B ( t ) \mathcal{S}(t) \subset\mathcal{B}(t) S ( t ) ⊂ B ( t ) separating two (or more) porous domains and/or a porous and a purely fluid domain.
If S ( t ) \mathcal{S}(t) S ( t ) separates two distinct porous domains, S ( t ) \mathcal{S}(t) S ( t ) is assumed to be a coherent interface convected by a globally continuous solid motion. If S ( t ) \mathcal{S}(t) S ( t ) separates a porous domain from a purely fluid domain, χ s \boldsymbol{\chi}_{s} χ s is understood to be continuously extended in the fluid domain. In all cases, S ( t ) \mathcal{S}(t) S ( t ) is the image under χ s \boldsymbol{\chi}_{s} χ s of a smooth not self-intersecting fixed submanifold S s ⊂ B s \mathcal{S}_{s} \subset\mathcal{B}_{s} S s ⊂ B s such that S s \mathcal{S}_{s} S s either cuts through B s \mathcal{B}_{s} B s intersecting the boundary of B s \mathcal{B}_{s} B s or is closed and completely contained in B s \mathcal{B}_{s} B s . S f ( t ) \mathcal{S}_{f}(t) S f ( t ) is the inverse image of S ( t ) \mathcal{S}(t) S ( t ) under χ f \boldsymbol{\chi}_{f} χ f .
Next, away from S s \mathcal{S}_{s} S s and S f ( t ) \mathcal{S}_{f}(t) S f ( t ) , the gradients (the meaning of ∇ \nabla ∇ is context dependent, cf. Appendix § 7 ) of χ s \boldsymbol{\chi}_{s} χ s and χ f \boldsymbol{\chi}_{f} χ f are
F s ( X s , t ) : = ∇ χ s ( X s , t ) and F f ( X f , t ) : = ∇ χ f ( X f , t ) , (2) \mathbf{F}_{s}(\mathbf{X}_{s},t) := \nabla\boldsymbol{\chi}_{s}(\mathbf{X}_{s},t) \quad\text{and} \quad\mathbf{F}_{f}(\mathbf{X}_{f},t) := \nabla\boldsymbol{\chi}_{f}(\mathbf{X}_{f},t), \tag*{(2)} F s ( X s , t ) := ∇ χ s ( X s , t ) and F f ( X f , t ) := ∇ χ f ( X f , t ) , (2)
with determinants J s J_s J s and J f J_f J f , respectively, which are taken to be strictly positive. Similarly, the solid and fluid material velocity fields are, respectively,
v s ( X s , t ) : = ∂ t χ s ( X s , t ) and v f ( X f , t ) : = ∂ t χ f ( X f , t ) . (3) \mathbf{v}_s(\mathbf{X}_s,t) := \partial_t\boldsymbol{\chi}_s(\mathbf{X}_s,t) \quad\text{and} \quad\mathbf{v}_f(\mathbf{X}_f,t) := \partial_t\boldsymbol{\chi}_f(\mathbf{X}_f,t). \tag*{(3)} v s ( X s , t ) := ∂ t χ s ( X s , t ) and v f ( X f , t ) := ∂ t χ f ( X f , t ) . (3)
In Ref. 18, it is posited that the motions χ s \boldsymbol{\chi}_s χ s and χ f \boldsymbol{\chi}_f χ f are diffeomorphisms away from the interface, but can suffer jump discontinuities in their space–time gradients so as to have the Hadamard properties.[18 ] A general discussion of the subject can be found in Refs. 18, 51, and 8. Here, we start with following the same kinematic assumption for χ s \boldsymbol{\chi}_s χ s as in Ref. 18, so that χ s \boldsymbol{\chi}_s χ s is continuous across the interface. We will also assume that, away from the interface, χ f \boldsymbol{\chi}_f χ f is a diffeomorphism. However, we will allow the fluid material velocity field to be discontinuous across the interface while only restricted to the requirements imposed by the jump condition of the mass balance. The kinematic constraint originating with the latter will be imposed via a Lagrange multiplier in § 4.1 .
We denote by m s \mathbf{m}_s m s the unit normal orienting S s S_s S s . The corresponding normal orienting S ( t ) S(t) S ( t ) is m \mathbf{m} m , which is given by
m : = ( J s F s − T m s ∥ J s F s − T m s ∥ ) ⊙ , (4) \mathbf{m} := \left(\frac{J_s\mathbf{F}_s^{-\mathsf{T}}\mathbf{m}_s}{\left\lVert J_s\mathbf{F}_s^{-\mathsf{T}}\mathbf{m}_s\right\rVert}\right)^{\odot}, \tag*{(4)} m := ( ∥ J s F s − T m s ∥ J s F s − T m s ) ⊙ , (4)
and the normal velocity of S ( t ) S(t) S ( t ) is the unique scalar field
c S ( t ) : S ( t ) → R , c S ( t ) : = v s ⊙ ⋅ m . (5) c_{S(t)}: S(t) \to\mathbb{R}, \quad c_{S(t)} := \mathbf{v}_s^{\odot}\cdot\mathbf{m}. \tag*{(5)} c S ( t ) : S ( t ) → R , c S ( t ) := v s ⊙ ⋅ m . (5)
In addition to the maps χ s \boldsymbol{\chi}_s χ s and χ f \boldsymbol{\chi}_f χ f , following Ref. 18, we introduce the following map
χ s f : B s → B f , χ s f : = χ f − 1 ∘ χ s . (6) \boldsymbol{\chi}_{sf}:B_s\to B_f,\quad\boldsymbol{\chi}_{sf}:=\boldsymbol{\chi}_f^{-1}\circ\boldsymbol{\chi}_s. \tag*{(6)} χ s f : B s → B f , χ s f := χ f − 1 ∘ χ s . (6)
The function χ s f ( X s , t ) \boldsymbol{\chi}_{sf}(\mathbf{X}_s,t) χ s f ( X s , t ) , which is well-defined away from the interface, identifies the fluid particle X f \mathbf{X}_f X f that, at time t t t , occupies the same position in the mixture as the solid material particle X s \mathbf{X}_s X s . This map encodes the relative motion of the fluid with respect to the solid. Letting
F s f : = ∇ χ s f , J s f = det F s f , and v s f : = ∂ t χ s f , (7) \mathbf{F}_{sf}:=\nabla\boldsymbol{\chi}_{sf},\quad J_{sf}=\det\mathbf{F}_{sf},\quad\text{and}\quad\mathbf{v}_{sf}:=\partial_t\boldsymbol{\chi}_{sf}, \tag*{(7)} F s f := ∇ χ s f , J s f = det F s f , and v s f := ∂ t χ s f , (7)
it can be shown that
F f ⊙ = F s F s f − 1 , J f ⊙ = J s / J s f , and v f ⊙ = v s − F s F s f − 1 v s f . (8) \mathbf{F}_f^{\odot}=\mathbf{F}_s\mathbf{F}_{sf}^{-1},\quad J_f^{\odot}=J_s/J_{sf},\quad\text{and}\quad\mathbf{v}_f^{\odot}=\mathbf{v}_s-\mathbf{F}_s\mathbf{F}_{sf}^{-1}\mathbf{v}_{sf}. \tag*{(8)} F f ⊙ = F s F s f − 1 , J f ⊙ = J s / J s f , and v f ⊙ = v s − F s F s f − 1 v s f . (8)
If the solid and fluid are submanifolds of an all-containing Euclidean manifold, the displacement of the solid and fluid phases, denoted by u s \mathbf{u}_s u s and u f \mathbf{u}_f u f , respectively, are defined as
u s ( X s , t ) : = χ s ( X s , t ) − X s and u f ( X f , t ) : = χ f ( X f , t ) − X f . (9) \mathbf{u}_s(\mathbf{X}_s,t):=\boldsymbol{\chi}_s(\mathbf{X}_s,t)-\mathbf{X}_s\quad\text{and}\quad\mathbf{u}_f(\mathbf{X}_f,t):=\boldsymbol{\chi}_f(\mathbf{X}_f,t)-\mathbf{X}_f. \tag*{(9)} u s ( X s , t ) := χ s ( X s , t ) − X s and u f ( X f , t ) := χ f ( X f , t ) − X f . (9)
We note that
u f ⊙ = χ s − χ s f . (10) \mathbf{u}_f^{\odot}=\boldsymbol{\chi}_s-\boldsymbol{\chi}_{sf}. \tag*{(10)} u f ⊙ = χ s − χ s f . (10)
The material accelerations of the solid and fluid phases, denoted by a s \mathbf{a}_{s} a s and a f \mathbf{a}_{f} a f , respectively, are
a s ( X s , t ) : = ∂ t v s ( X s , t ) and a f ( X f , t ) : = ∂ t v f ( X f , t ) . (11) \mathbf{a}_{s}(\mathbf{X}_{s},t) := \partial_{t}\mathbf{v}_{s}(\mathbf{X}_{s},t) \quad\text{and} \quad\mathbf{a}_{f}(\mathbf{X}_{f},t) := \partial_{t}\mathbf{v}_{f}(\mathbf{X}_{f},t). \tag*{(11)} a s ( X s , t ) := ∂ t v s ( X s , t ) and a f ( X f , t ) := ∂ t v f ( X f , t ) . (11)
Balance of mass and momentum
Here we summarize the presentation in Refs. 8 and 9.
Balance of mass
In their pure form, the fluid and the solid phases are assumed to be incompressible with mass densities given by the constants ρ f ∗ \rho_{f}^{*} ρ f ∗ and ρ s ∗ \rho_{s}^{*} ρ s ∗ , respectively. Let
ϕ s : B ( t ) → [ 0 , 1 ] and ϕ f : B ( t ) → [ 0 , 1 ] (12) \phi_{s}: B(t) \to[0,1] \quad\text{and} \quad\phi_{f}: B(t) \to[0,1] \tag*{(12)} ϕ s : B ( t ) → [ 0 , 1 ] and ϕ f : B ( t ) → [ 0 , 1 ] (12)
denote the volume fractions of the solid and fluid, respectively, subject to the saturation condition
ϕ s + ϕ f = 1. (13) \phi_{s}+\phi_{f}=1. \tag*{(13)} ϕ s + ϕ f = 1. (13)
Then, the mass densities of the solid and the fluid over B ( t ) B(t) B ( t ) (i.e., in the mixture) are, respectively,
ρ s = ρ s ∗ ϕ s and ρ f = ρ f ∗ ϕ f . (14) \rho_{s}=\rho_{s}^{*}\phi_{s} \quad\text{and} \quad\rho_{f}=\rho_{f}^{*}\phi_{f}. \tag*{(14)} ρ s = ρ s ∗ ϕ s and ρ f = ρ f ∗ ϕ f . (14)
In the absence of chemical reactions, the mass balance in B ( t ) B(t) B ( t ) requires that
∂ t ρ a + ∇ ⋅ ( ρ a v a e ) = 0 ⇒ ∂ t ϕ a + ∇ ⋅ ( ϕ a v a e ) = 0 , (15) \partial_{t}\rho_{a}+\nabla\cdot(\rho_{a}\mathbf{v}_{a}^{\mathrm{e}})=0 \quad\Rightarrow\quad\partial_{t}\phi_{a}+\nabla\cdot(\phi_{a}\mathbf{v}_{a}^{\mathrm{e}})=0, \tag*{(15)} ∂ t ρ a + ∇ ⋅ ( ρ a v a e ) = 0 ⇒ ∂ t ϕ a + ∇ ⋅ ( ϕ a v a e ) = 0 , (15)
for each species a = s , f a=s,f a = s , f . In view of (13) , the motion of the mixture is governed by the following (single) constraint:
∇ ⋅ ( ϕ s v s e + ϕ f v f e ) = 0. (16) \nabla\cdot(\phi_{s}\mathbf{v}_{s}^{\mathrm{e}}+\phi_{f}\mathbf{v}_{f}^{\mathrm{e}})=0. \tag*{(16)} ∇ ⋅ ( ϕ s v s e + ϕ f v f e ) = 0. (16)
We now define the following fields over B ( t ) B(t) B ( t ) :
v v a : = ϕ s v s e + ϕ f v f e and v f l t : = ϕ f ( v f e − v s e ) . (17) \mathbf{v}_{\mathrm{va}}:=\phi_{s}\mathbf{v}_{s}^{\mathrm{e}}+\phi_{f}\mathbf{v}_{f}^{\mathrm{e}} \quad\text{and} \quad\mathbf{v}_{\mathrm{flt}}:=\phi_{f}(\mathbf{v}_{f}^{\mathrm{e}}-\mathbf{v}_{s}^{\mathrm{e}}). \tag*{(17)} v va := ϕ s v s e + ϕ f v f e and v flt := ϕ f ( v f e − v s e ) . (17)
The field v v a \mathbf{v}_{\mathrm{va}} v va is the volume average velocity of the mixture, whereas v f l t \mathbf{v}_{\mathrm{flt}} v flt is the filtration velocity. Using (17) , (16) can also be written as
∇ ⋅ v v a = 0 or ∇ ⋅ v s e + ∇ ⋅ v f l t = 0. (18) \nabla\cdot\mathbf{v}_{\mathrm{va}}=0 \quad\text{or} \quad\nabla\cdot\mathbf{v}_{s}^{\mathrm{e}}+\nabla\cdot\mathbf{v}_{\mathrm{flt}}=0. \tag*{(18)} ∇ ⋅ v va = 0 or ∇ ⋅ v s e + ∇ ⋅ v flt = 0. (18)
As v f \mathbf{v}_{f} v f may not be continuous across S ( t ) S(t) S ( t ) , we also need to state the jump condition of the balance of mass. We assume that the interface does not support the production or removal of mass. Hence, using traditional arguments (cf. Refs. 8 and 24), at every point on S ( t ) S(t) S ( t ) we must have
⟦ v f l t ⟧ ⋅ m = 0 ⇒ ⟦ v v a ⟧ ⋅ m = 0 , (19) \llbracket\mathbf{v}_{\mathrm{flt}}\rrbracket\cdot\mathbf{m}=0 \quad\Rightarrow\quad\llbracket\mathbf{v}_{\mathrm{va}}\rrbracket\cdot\mathbf{m}=0, \tag*{(19)} [ [ v flt ] ] ⋅ m = 0 ⇒ [ [ v va ] ] ⋅ m = 0 , (19)
where the second of (19) is a consequence of the continuity of v s \mathbf{v}_{s} v s .
The second of (15) and the assumed regularity of the motions away from the interface imply the existence of referential forms of the volume fraction distributions for the solid and fluid phases, which we denote by ϕ R s \phi_{R_s} ϕ R s and ϕ R f \phi_{R_f} ϕ R f , respectively:
\phi_{R_s}=J_s\phi_s^{\circledS}\quad\text{and}\quad\phi_{R_f}=J_f\phi_f^{\circledF}. \tag*{(20)}
Remark 3.1 (Solid immersed in fluid). In the present kinematic framework, it is admissible for ϕ R s \phi_{R_s} ϕ R s to vanish over parts of B s B_s B s . In such cases, the support of ϕ R s \phi_{R_s} ϕ R s will be assumed to be a closed manifold that supports the physics of a solid body immersed in a fluid.
Balance of momentum
The mixture is assumed to be isothermal. Using Refs. 8 and 9 as guides, the balance of momentum for phase a = s , f a=s,f a = s , f in a mixture requires that in B ( t ) B(t) B ( t ) away from S ( t ) S(t) S ( t ) we have
ρ a a ∘ a − ∇ ⋅ T a − b a − p a = 0 , (21) \rho_a\overset{\circ}{\mathbf{a}}_a-\nabla\cdot\mathbf{T}_a-\mathbf{b}_a-\mathbf{p}_a=0, \tag*{(21)} ρ a a ∘ a − ∇ ⋅ T a − b a − p a = 0 , (21)
where T a \mathbf{T}_a T a is the Cauchy stress of species a a a , b a \mathbf{b}_a b a is an external force density (per unit volume of B ( t ) B(t) B ( t ) ), and p a \mathbf{p}_a p a is the force density (per unit volume of B ( t ) B(t) B ( t ) ) acting on a a a because of its interaction with the rest of the mixture. The interaction forces must satisfy the constraint
∑ a p a = 0. (22) \sum_a\mathbf{p}_a=0. \tag*{(22)} a ∑ p a = 0. (22)
For the momentum balance jump condition over S ( t ) S(t) S ( t ) we posit that
[ [ ρ a ( v a ∘ − v s ∘ ) ⊗ ( v a ∘ − v s ∘ ) − T a ] ] m + ξ a = 0 , (23) \left[\!\left[\rho_a(\mathbf{v}_a^\circ-\mathbf{v}_s^\circ)\otimes(\mathbf{v}_a^\circ-\mathbf{v}_s^\circ)-\mathbf{T}_a\right]\!\right]\mathbf{m}+\boldsymbol{\xi}_a=0, \tag*{(23)} [ [ ρ a ( v a ∘ − v s ∘ ) ⊗ ( v a ∘ − v s ∘ ) − T a ] ] m + ξ a = 0 , (23)
where ξ a \boldsymbol{\xi}_a ξ a is a force density per unit area of S ( t ) S(t) S ( t ) analogous to p a \mathbf{p}_a p a , and for which we posit a similar constraint
∑ a ξ a = 0. (24) \sum_a\boldsymbol{\xi}_a=0. \tag*{(24)} a ∑ ξ a = 0. (24)
Equation (3.12) differs from what can be found in Ref. 8 (cf. Eq. (2.10.22) in Ref. 8) because of ξ a \boldsymbol{\xi}_a ξ a . In Remark 4.5 of Section 4.1, we argue that ξ a \boldsymbol{\xi}_a ξ a plays a role in modeling dissipation at the interface and in the satisfaction of the first of Eqs. (3.7).
Equations (3.12) and (3.13) yield the following conditions for the solid, fluid, and mixture, respectively:
− [ [ T s ] ] m + ξ s = 0 , d [ [ v f ∘ ] ] − [ [ T f ] ] m + ξ f = 0 , (25) -\left[\!\left[\mathbf{T}_s\right]\!\right]\mathbf{m}+\boldsymbol{\xi}_s=0,\quad d\left[\!\left[\mathbf{v}_f^\circ\right]\!\right]-\left[\!\left[\mathbf{T}_f\right]\!\right]\mathbf{m}+\boldsymbol{\xi}_f=0, \tag*{(25)} − [ [ T s ] ] m + ξ s = 0 , d [ [ v f ∘ ] ] − [ [ T f ] ] m + ξ f = 0 , (25)
and
d [ [ v f ∘ ] ] − [ [ T f + T s ] ] m = 0. (26) d\left[\!\left[\mathbf{v}_f^\circ\right]\!\right]-\left[\!\left[\mathbf{T}_f+\mathbf{T}_s\right]\!\right]\mathbf{m}=0. \tag*{(26)} d [ [ v f ∘ ] ] − [ [ T f + T s ] ] m = 0. (26)
where d d d is
d : = ρ f ( v f ⊛ − v s ⊛ ) ⋅ m ⇒ ⟦ d ⟧ = 0. (27) d := \rho_f\left(\mathbf{v}_f^{\circledast}-\mathbf{v}_s^{\circledast}\right)\cdot\mathbf{m}\quad\Rightarrow\quad\llbracket d\rrbracket=0. \tag*{(27)} d := ρ f ( v f ⊛ − v s ⊛ ) ⋅ m ⇒ [ [ d ] ] = 0. (27)
Remark 3.2. The jump condition in Eq. (26) is generally agreed upon in the literature (cf. Eq. (11b) in Ref. 27, Eq. (2.73) in Ref. 48, and Eq. (30) in Ref. 18). However, the same cannot be said for Eqs. (25) . The tangential projection of the second of Eqs. (25) is similar to Eq. (32) in Ref. 18.
Constitutive response functions
We assume that the pure solid phase is incompressible and hyperelastic. The strain energy of the pure solid phase is assumed to be a non-negative scalar function of the form
Ψ ^ s = Ψ ^ s ( C ‾ s , X s ) , (28) \widehat{\Psi}_s=\widehat{\Psi}_s\left(\overline{\mathbf{C}}_s,\mathbf{X}_s\right), \tag*{(28)} Ψ s = Ψ s ( C s , X s ) , (28)
where
C s : = F s T F s and C ‾ s = J s − 2 / 3 C s , (29) \mathbf{C}_s:=\mathbf{F}_s^{\mathsf{T}}\mathbf{F}_s\quad\text{and}\quad\overline{\mathbf{C}}_s=J_s^{-2/3}\mathbf{C}_s, \tag*{(29)} C s := F s T F s and C s = J s − 2/3 C s , (29)
so that C ‾ s \overline{\mathbf{C}}_s C s is the isochoric part of C s \mathbf{C}_s C s . The strain energy of the solid phase in the mixture is simply assumed to that of the pure solid scaled by its volume fraction:
Ψ s : = ϕ R s Ψ ^ s . (30) \Psi_s:=\phi_{R_s}\widehat{\Psi}_s. \tag*{(30)} Ψ s := ϕ R s Ψ s . (30)
It is convenient to introduce a symbol to refer to the solid’s strain energy per unit volume of the current configuration:
ψ s : = ( J s − 1 Ψ s ) ⊛ . (31) \psi_s:=\left(J_s^{-1}\Psi_s\right)^{\circledast}. \tag*{(31)} ψ s := ( J s − 1 Ψ s ) ⊛ . (31)
The contributions to the first Piola-Kirchhoff and Cauchy stress tensors originating from the elastic strain energy are, respectively,
P s e = 2 ϕ R s F s ∂ Ψ ^ s ∂ C s and T s e = ( 2 ϕ R s J s − 1 F s ∂ Ψ ^ s ∂ C s F s T ) ⊛ . (32) \mathbf{P}_s^{e}=2\phi_{R_s}\mathbf{F}_s\frac{\partial\widehat{\Psi}_s}{\partial\mathbf{C}_s}
\quad\text{and}\quad
\mathbf{T}_s^{e}=\left(2\phi_{R_s}J_s^{-1}\mathbf{F}_s\frac{\partial\widehat{\Psi}_s}{\partial\mathbf{C}_s}\mathbf{F}_s^{\mathsf{T}}\right)^{\circledast}. \tag*{(32)} P s e = 2 ϕ R s F s ∂ C s ∂ Ψ s and T s e = ( 2 ϕ R s J s − 1 F s ∂ C s ∂ Ψ s F s T ) ⊛ . (32)
In addition to a hyperelastic contribution, we allow for a dissipative stress contribution due to the interaction of the solid with the fluid phase, analogous in form to what is found in Eq. (2.1.3) in Ref. 8. With this in mind, for a = s , f a=s,f a = s , f , we define
D a : = 1 2 ( L a + L a T ) , with L a : = ∇ ( v a ⊛ ) . (33) \mathbf{D}_a:=\frac{1}{2}\left(\mathbf{L}_a+\mathbf{L}_a^{\mathsf{T}}\right),\quad\text{with}\quad\mathbf{L}_a:=\nabla\left(\mathbf{v}_a^{\circledast}\right). \tag*{(33)} D a := 2 1 ( L a + L a T ) , with L a := ∇ ( v a ⊛ ) . (33)
Then, the viscous contribution to the solid’s stress in the mixture will be
T s v = 2 μ B ( D s − D f ) . (34) \mathbf{T}_s^{v}=2\mu_B\left(\mathbf{D}_s-\mathbf{D}_f\right). \tag*{(34)} T s v = 2 μ B ( D s − D f ) . (34)
This type of contribution is common in poroelasticity, although it is usually discussed in relation to the fluid’s behavior where it is referred to as a Brinkman term.3 The dynamic viscosity μ B \mu_B μ B is generally a function of ϕ s \phi_s ϕ s with support identical to that of ϕ s \phi_s ϕ s .
As far as the pure fluid phase is concerned, having already assumed that it is incompressible, we further assume that it is linear viscous, i.e., with viscous stress equal to 2 μ ^ f D f 2\hat{\mu}_{\mathrm{f}}\mathbf{D}_{\mathrm{f}} 2 μ ^ f D f , where μ ^ f ∈ R 0 + \hat{\mu}_{\mathrm{f}} \in\mathbb{R}^{+}_{0} μ ^ f ∈ R 0 + is a constant dynamic shear viscosity. As for the stress response of the fluid in the mixture, we will assume that it consists of two terms: one with form 2 μ f D f 2\mu_{\mathrm{f}}\mathbf{D}_{\mathrm{f}} 2 μ f D f , where μ f \mu_{\mathrm{f}} μ f is a dynamic viscosity that depends on ϕ f \phi_{\mathrm{f}} ϕ f , and the other being the counterpart of the Brinkman term introduced earlier (again, cf. Eq. (2.1.3) in Ref. 8):
T f v = 2 μ f D f + 2 μ B ( D f − D s ) . (35) \mathbf{T}_{\mathrm{f}}^{\mathrm{v}} = 2\mu_{\mathrm{f}}\mathbf{D}_{\mathrm{f}} + 2\mu_{\mathrm{B}}(\mathbf{D}_{\mathrm{f}}-\mathbf{D}_{\mathrm{s}}). \tag*{(35)} T f v = 2 μ f D f + 2 μ B ( D f − D s ) . (35)
The dynamic viscosity μ f \mu_{\mathrm{f}} μ f will be subject to intuitive constraints such as μ f → 0 \mu_{\mathrm{f}} \to0 μ f → 0 for ϕ f → 0 \phi_{\mathrm{f}} \to0 ϕ f → 0 and μ f → μ ^ f \mu_{\mathrm{f}} \to\hat{\mu}_{\mathrm{f}} μ f → μ ^ f for ϕ f → 1 \phi_{\mathrm{f}} \to1 ϕ f → 1 .
A pressure field (cf., e.g., Ref. 9) denoted by p p p and called the pore pressure is introduced over B ( t ) B(t) B ( t ) to enforce Eq. (16) . Then, the full forms of T s \mathbf{T}_{\mathrm{s}} T s and T f \mathbf{T}_{\mathrm{f}} T f are, respectively,
T s = − ϕ s p I + T s e + T s v and T f = − ϕ f p I + T f v , (36) \mathbf{T}_{\mathrm{s}}=-\phi_{\mathrm{s}}p\mathbf{I}+\mathbf{T}_{\mathrm{s}}^{\mathrm{e}}+\mathbf{T}_{\mathrm{s}}^{\mathrm{v}}
\quad\text{and}\quad
\mathbf{T}_{\mathrm{f}}=-\phi_{\mathrm{f}}p\mathbf{I}+\mathbf{T}_{\mathrm{f}}^{\mathrm{v}}, \tag*{(36)} T s = − ϕ s p I + T s e + T s v and T f = − ϕ f p I + T f v , (36)
where I \mathbf{I} I is the identity tensor.
The constitutive equations for a mixture must also include expressions for p a \mathbf{p}_{a} p a and ξ a \boldsymbol{\xi}_{a} ξ a (a = s , f a=\mathrm{s},\mathrm{f} a = s , f ). In both cases, we will assume that these forces originate from a viscous resistance to the sliding of the fluid relative to the solid, i.e., they originate from the kinematic quantity ( v f − v s ) (\mathbf{v}_{\mathrm{f}}-\mathbf{v}_{\mathrm{s}}) ( v f − v s ) .
For p f \mathbf{p}_{\mathrm{f}} p f (by Eq. (22) , p s = − p f \mathbf{p}_{\mathrm{s}}=-\mathbf{p}_{\mathrm{f}} p s = − p f ), we follow Ref. 9 and write
p f = p ∇ ϕ f + f f , with f f = − μ D ϕ f 2 κ s ( v f ∘ − v s ∘ ) , (37) \mathbf{p}_{\mathrm{f}}=p\nabla\phi_{\mathrm{f}}+\mathbf{f}_{\mathrm{f}},
\quad\text{with}\quad
\mathbf{f}_{\mathrm{f}}=-\frac{\mu_{\mathrm{D}}\phi_{\mathrm{f}}^{2}}{\kappa_{\mathrm{s}}}(\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{v}_{\mathrm{s}}^{\circ}), \tag*{(37)} p f = p ∇ ϕ f + f f , with f f = − κ s μ D ϕ f 2 ( v f ∘ − v s ∘ ) , (37)
where κ s \kappa_{\mathrm{s}} κ s is the solid’s permeability, while μ D \mu_{\mathrm{D}} μ D is a dynamic viscosity that, just like μ B \mu_{\mathrm{B}} μ B , has support identical to that of ϕ s \phi_{\mathrm{s}} ϕ s . We defer the modeling of ξ a \boldsymbol{\xi}_{a} ξ a to later in the paper.
Additional consequences of momentum balance
Here we derive an energy estimate following a process similar to that used in the theorem of power expended (TPE).[22 , 24 ]
We consider a time independent domain Ω ⊂ B ( t ) \Omega\subset B(t) Ω ⊂ B ( t ) such that it is crossed by the singularity surface S ( t ) S(t) S ( t ) . Letting S ( t ) Ω : = Ω ∩ S ( t ) S(t)_{\Omega}:=\Omega\cap S(t) S ( t ) Ω := Ω ∩ S ( t ) , it is assumed that S ( t ) Ω S(t)_{\Omega} S ( t ) Ω is an open surface that separates Ω \Omega Ω into two domains Ω + ( t ) \Omega^{+}(t) Ω + ( t ) and Ω − ( t ) \Omega^{-}(t) Ω − ( t ) , such that the unit normal m \mathbf{m} m is outward relative to Ω − ( t ) \Omega^{-}(t) Ω − ( t ) . Next, we take the inner product between Eq. (21) and v a ∘ \mathbf{v}_{a}^{\circ} v a ∘ , and integrate over Ω \Omega Ω . Accounting for mass balance, and using integration by parts as well as the transport theorem, for a = s , f a=s,f a = s , f we have
d d t ∫ Ω k a + ∫ Ω ( T a : L a − p a ⋅ v a ∘ ) + ∫ S ( t ) Ω [ [ T a v a − k a ( v a ∘ − v s ∘ ) ] ] ⋅ m = − ∫ ∂ Ω e x t ( k a v a ∘ ⋅ n − T a n ⋅ v a ∘ ) + ∫ Ω b a ⋅ v a ∘ , (38) \frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega} k_a+\int_{\Omega}\left(\mathbf{T}_a:\mathbf{L}_a-p_a\cdot\mathbf{v}_a^{\circ}\right)+\int_{S(t)_{\Omega}}\left[\!\left[\mathbf{T}_a\mathbf{v}_a-k_a\left(\mathbf{v}_a^{\circ}-\mathbf{v}_s^{\circ}\right)\right]\!\right]\cdot\mathbf{m}
=-\int_{\partial\Omega_{\mathrm{ext}}}\left(k_a\mathbf{v}_a^{\circ}\cdot\mathbf{n}-\mathbf{T}_a\mathbf{n}\cdot\mathbf{v}_a^{\circ}\right)+\int_{\Omega}\mathbf{b}_a\cdot\mathbf{v}_a^{\circ}, \tag*{(38)} d t d ∫ Ω k a + ∫ Ω ( T a : L a − p a ⋅ v a ∘ ) + ∫ S ( t ) Ω [ [ T a v a − k a ( v a ∘ − v s ∘ ) ] ] ⋅ m = − ∫ ∂ Ω ext ( k a v a ∘ ⋅ n − T a n ⋅ v a ∘ ) + ∫ Ω b a ⋅ v a ∘ , (38)
where n \mathbf{n} n is the boundary outward unit normal and k a k_a k a is the kinetic energy density of species a a a per unit volume of the mixture, i.e.,
k a : = 1 2 ρ a ∥ v a ∘ ∥ 2 . (39) k_a:=\frac{1}{2}\rho_a\left\lVert\mathbf{v}_a^{\circ}\right\rVert^2. \tag*{(39)} k a := 2 1 ρ a ∥ v a ∘ ∥ 2 . (39)
To express our energy estimate in a compact manner, we first define the quantities W e x t ( Ω ) \mathcal{W}_{\mathrm{ext}}(\Omega) W ext ( Ω ) , D \mathcal{D} D , and I \mathcal{I} I as follows:
W e x t ( Ω ) : = − ∫ ∂ Ω e x t [ ( k s + ψ s ) v s ∘ + k f v f ∘ ] ⋅ n + ∫ ∂ Ω e x t ( T s n ⋅ v s ∘ + T f n ⋅ v f ∘ ) + ∫ Ω ( b s ⋅ v s ∘ + b f ⋅ v f ∘ ) , (40) \mathcal{W}_{\mathrm{ext}}(\Omega):=-\int_{\partial\Omega_{\mathrm{ext}}}\left[(k_s+\psi_s)\mathbf{v}_s^{\circ}+k_f\mathbf{v}_f^{\circ}\right]\cdot\mathbf{n}
+\int_{\partial\Omega_{\mathrm{ext}}}\left(\mathbf{T}_s\mathbf{n}\cdot\mathbf{v}_s^{\circ}+\mathbf{T}_f\mathbf{n}\cdot\mathbf{v}_f^{\circ}\right)
+\int_{\Omega}\left(\mathbf{b}_s\cdot\mathbf{v}_s^{\circ}+\mathbf{b}_f\cdot\mathbf{v}_f^{\circ}\right), \tag*{(40)} W ext ( Ω ) := − ∫ ∂ Ω ext [ ( k s + ψ s ) v s ∘ + k f v f ∘ ] ⋅ n + ∫ ∂ Ω ext ( T s n ⋅ v s ∘ + T f n ⋅ v f ∘ ) + ∫ Ω ( b s ⋅ v s ∘ + b f ⋅ v f ∘ ) , (40)
D : = 2 μ f ∥ D f ∥ 2 + 2 μ B ∥ D f − D s ∥ 2 + μ D ϕ f 2 κ s ∥ v f ∘ − v s ∘ ∥ 2 , (41) \mathcal{D}:=2\mu_f\left\lVert\mathbf{D}_f\right\rVert^2+2\mu_B\left\lVert\mathbf{D}_f-\mathbf{D}_s\right\rVert^2+\frac{\mu_D\phi_f^2}{\kappa_s}\left\lVert\mathbf{v}_f^{\circ}-\mathbf{v}_s^{\circ}\right\rVert^2, \tag*{(41)} D := 2 μ f ∥ D f ∥ 2 + 2 μ B ∥ D f − D s ∥ 2 + κ s μ D ϕ f 2 v f ∘ − v s ∘ 2 , (41)
and
I : = [ [ T f m ⋅ ( v f ∘ − v s ∘ ) ] ] − 1 2 d [ [ ∥ v f ∘ − v s ∘ ∥ 2 ] ] . (42) \mathcal{I}:=\left[\!\left[\mathbf{T}_f\mathbf{m}\cdot\left(\mathbf{v}_f^{\circ}-\mathbf{v}_s^{\circ}\right)\right]\!\right]-\frac{1}{2}d\left[\!\left[\left\lVert\mathbf{v}_f^{\circ}-\mathbf{v}_s^{\circ}\right\rVert^2\right]\!\right]. \tag*{(42)} I := [ [ T f m ⋅ ( v f ∘ − v s ∘ ) ] ] − 2 1 d [ [ v f ∘ − v s ∘ 2 ] ] . (42)
The quantities W e x t ( Ω ) \mathcal{W}_{\mathrm{ext}}(\Omega) W ext ( Ω ) , D \mathcal{D} D , and I \mathcal{I} I denote, respectively, the external supply of mechanical power to Ω \Omega Ω , the volume density of the dissipation rate in Ω \Omega Ω , and the surface density of the power expenditure on the interface. Then, summing the contributions in Eq. (38) for the fluid and the solid phases, and using the constitutive response functions for the stress in both the solid and fluid phases, for the mixture, we have (see Proposition B.3 in the Appendix)
W e x t ( Ω ) = d d t ∫ Ω ( k s + k f + ψ s ) + ∫ Ω D + ∫ S ( t ) I . (43) \mathcal{W}_{\mathrm{ext}}(\Omega)=\frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega}\left(k_s+k_f+\psi_s\right)+\int_{\Omega}\mathcal{D}+\int_{S(t)}\mathcal{I}. \tag*{(43)} W ext ( Ω ) = d t d ∫ Ω ( k s + k f + ψ s ) + ∫ Ω D + ∫ S ( t ) I . (43)
Remark 3.4 (Theorem of power expended). (43) is the version of the theorem of power expended (cf. Ref. 22) for the present context. It presents a consequence of the momentum balance laws as opposed to, say, the balance of energy. In a way, the result in Eq. (43) is intuitive: the right-hand side tells us what happens to the power input to Ω \Omega Ω , which is the left-hand side. It also allows to consider the power expended at the interface in a transparent manner, as argued in the following remarks.
where we have used the fact that the normal component of the filtration velocity is continuous as a consequence of the balance of mass. If ( v f l t ⋅ m ) ≠ 0 (\mathbf{v}_{\mathrm{flt}}\cdot\mathbf{m})\ne0 ( v flt ⋅ m ) = 0 , we would then conclude that ⟦ p ⟧ = 0 \llbracket p\rrbracket=0 [ [ p ] ] = 0 . This result is a “classic” result (cf., e.g., Ref. 27) that is recovered in the current treatment without enforcing it strongly by representing p p p as a globally continuous field.
A Variational Formulation
Here we derive the field equations using Hamilton’s principle (cf. Ref. 6) following the approach in Ref. 18. The use of Hamilton’s principle in poroelasticity is not new (cf. Ref. 6). Our purpose in somewhat duplicating the derivation of the field equations is (i) to provide a connection between the Newtonian approach (cf. Ref. 23) to derive the field equations and a variational one; and (ii) to find a simple path to their numerical implementation using FEM. As mentioned in Remark 3.6, we wish to translate assumptions on the interface constitutive response to energy estimates for our FEM formulation in a direct way.
Hamilton-Rayleigh Principle
In the chosen kinematic framework, the system’s configuration can be equivalently described by the pair { χ s , χ f } \{\chi_s,\chi_f\} { χ s , χ f } (cf. (1) ) or the pair { χ s , χ s f } \{\chi_s,\chi_{sf}\} { χ s , χ s f } (cf. (6) ). Recalling that the inverse image of the interface S ( t ) \mathcal{S}(t) S ( t ) under χ s ( X s , t ) \chi_s(\mathbf{X}_s,t) χ s ( X s , t ) is time independent, and observing that each element of { χ s , χ s f } \{\chi_s,\chi_{sf}\} { χ s , χ s f } has the same spatial domain B s B_s B s and is continuous over it, we choose
q ( t ) : = { χ s , χ s f } (46) q(t):=\{\chi_s,\chi_{sf}\} \tag*{(46)} q ( t ) := { χ s , χ s f } (46)
as the (abstract) motion of our system for the purpose of applying Hamilton’s principle.
Letting q ˙ : = ∂ t q \dot{q}:=\partial_tq q ˙ := ∂ t q , and interpreting differentiation with respect to q q q and q ˙ \dot{q} q ˙ in a variational sense, for any sufficiently smooth q q q in time and δ q \delta q δ q with compact support in the time interval [ 0 , T ] [0,T] [ 0 , T ] , we require
∫ 0 T ∂ ( K − U ) ∂ q δ q + ∫ 0 T ∂ ( C + C f l t − R ) ∂ q ˙ δ q = 0 , (47) \int_0^T\frac{\partial(\mathcal{K}-\mathcal{U})}{\partial q}\delta q+\int_0^T\frac{\partial(\mathcal{C}+\mathcal{C}_{\mathrm{flt}}-\mathcal{R})}{\partial\dot{q}}\delta q=0, \tag*{(47)} ∫ 0 T ∂ q ∂ ( K − U ) δ q + ∫ 0 T ∂ q ˙ ∂ ( C + C flt − R ) δ q = 0 , (47)
where, assuming for simplicity (and without loss of generality) that the boundary conditions consist of homogeneous Dirichlet data for both components of the motion, K \mathcal{K} K , U \mathcal{U} U , C \mathcal{C} C , C f l t \mathcal{C}_{\mathrm{flt}} C flt , and R \mathcal{R} R are the following functionals:
K : = ∫ B s ( k s + k f ) s J s , (48) \mathcal{K}:=\int_{B_s}(k_s+k_f)^{\mathrm{s}}J_s, \tag*{(48)} K := ∫ B s ( k s + k f ) s J s , (48)
U : = ∫ B s ( Ψ s − u s ⋅ b s s J s − u f ⋅ b f s J s ) , (49) \mathcal{U}:=\int_{B_s}\left(\Psi_s-\mathbf{u}_s\cdot\mathbf{b}_s^{\mathrm{s}}J_s-\mathbf{u}_f\cdot\mathbf{b}_f^{\mathrm{s}}J_s\right), \tag*{(49)} U := ∫ B s ( Ψ s − u s ⋅ b s s J s − u f ⋅ b f s J s ) , (49)
C : = ∫ B s p s J s [ ∇ ⋅ ( v s e + v f l t ) ] s , (50) \mathcal{C}:=\int_{B_s}p^{\mathrm{s}}J_s[\nabla\cdot(\mathbf{v}_s^{\mathrm{e}}+\mathbf{v}_{\mathrm{flt}})]^{\mathrm{s}}, \tag*{(50)} C := ∫ B s p s J s [ ∇ ⋅ ( v s e + v flt ) ] s , (50)
C f l t : = ∫ S s p ⊛ [ [ v f l t ⋅ J s F s − T m s ] ] (51) \mathcal{C}_{\mathrm{flt}} := \int_{S_s} p^{\circledast} [\![\mathbf{v}_{\mathrm{flt}}\cdot J_s\mathbf{F}_s^{-\mathrm{T}}\mathbf{m}_s]\!] \tag*{(51)} C flt := ∫ S s p ⊛ [ [ v flt ⋅ J s F s − T m s ] ] (51)
2 R : = ∫ B s D s ⊛ J s + ∫ S s ( D S ( t ) ) ⊛ J s ∥ F s − T m s ∥ . (52) 2\mathcal{R} := \int_{B_s}\mathcal{D}_s^{\circledast}J_s+\int_{S_s}\left(\mathcal{D}_{S(t)}\right)^{\circledast}J_s\left\|\mathbf{F}_s^{-\mathrm{T}}\mathbf{m}_s\right\|. \tag*{(52)} 2 R := ∫ B s D s ⊛ J s + ∫ S s ( D S ( t ) ) ⊛ J s F s − T m s . (52)
K \mathcal{K} K and U \mathcal{U} U are the system’s kinetic and potential energy, respectively. C \mathcal{C} C encodes constituent’s (individual) incompressibility, whereas letting
p : S ( t ) → R (53) \mathbf{p}:S(t)\to\mathbb{R} \tag*{(53)} p : S ( t ) → R (53)
be a Lagrange multiplier, C f l t \mathcal{C}_{\mathrm{flt}} C flt encodes the requirement that the normal component of the filtration velocity be continuous across S ( t ) S(t) S ( t ) . R \mathcal{R} R is the time rate of energy dissipation. The term D S ( t ) \mathcal{D}_{S(t)} D S ( t ) represents the interface dissipation mentioned in Remark 3.5, for which we adopt the following (simple) quadratic form:
D S ( t ) : = 1 2 μ S ∥ [ [ v f l t ] ] ∥ 2 , (54) \mathcal{D}_{S(t)}:=\frac{1}{2}\mu_S\left\|[\![\mathbf{v}_{\mathrm{flt}}]\!]\right\|^2, \tag*{(54)} D S ( t ) := 2 1 μ S ∥ [ [ v flt ] ] ∥ 2 , (54)
with μ S \mu_S μ S a positive constant with dimensions of a dynamic viscosity per unit length.
Applying standard methods, under the stated assumptions (cf. also Ref. 18),
Eqs. (4.2)–(4.9) yield the following result (cf. Proposition B.5 in the Appendix):
0 = ∫ B s [ J s ( ρ s a s ⊚ + ρ f a f ⊚ − b s − b f − ∇ ⋅ ( T s + T f ) ) S ⋅ δ χ s ] − ∫ B s f [ J s ( ρ f a f ⊚ − b f − ∇ ⋅ T f − p f ) S ⋅ F s F s f − 1 δ χ s f ] + ∫ S s [ [ ρ f ( v f ⊚ − v s ⊚ ) ⊗ ( v f ⊚ − v s ⊚ ) − T s − T f ] S J s F s − T m s ⋅ δ χ s ] + ∫ S s [ [ ( k f I − ρ f v f ⊚ ⊗ ( v f ⊚ − v s ⊚ ) ) S J s F s − T m s ⋅ F s F s f − 1 δ χ s f ] ] + ∫ S s [ [ ( ϕ f p I + T f − 1 2 ϕ f μ S [ v f ∥ ] ⊗ m ) S J s F s − T m s ⋅ F s F s f − 1 δ χ s f ] ] . (55) \begin{aligned}
0={}&\int_{B_s}\left[J_s\left(\rho_s\mathbf{a}_s^{\circledcirc}+\rho_f\mathbf{a}_f^{\circledcirc}-\mathbf{b}_s-\mathbf{b}_f-\nabla\cdot(\mathbf{T}_s+\mathbf{T}_f)\right)^{\mathsf{S}}\cdot\delta\boldsymbol{\chi}_s\right]\\
&-\int_{B_{sf}}\left[J_s\left(\rho_f\mathbf{a}_f^{\circledcirc}-\mathbf{b}_f-\nabla\cdot\mathbf{T}_f-\mathbf{p}_f\right)^{\mathsf{S}}\cdot\mathbf{F}_s\mathbf{F}_{sf}^{-1}\delta\boldsymbol{\chi}_{sf}\right]\\
&+\int_{S_s}\left[\left[\rho_f\left(\mathbf{v}_f^{\circledcirc}-\mathbf{v}_s^{\circledcirc}\right)\otimes\left(\mathbf{v}_f^{\circledcirc}-\mathbf{v}_s^{\circledcirc}\right)-\mathbf{T}_s-\mathbf{T}_f\right]^{\mathsf{S}}\mathbf{J}_s\mathbf{F}_s^{-\mathsf{T}}\mathbf{m}_s\cdot\delta\boldsymbol{\chi}_s\right]\\
&+\int_{S_s}\left[\left[\left(\mathbf{k}_f\mathbf{I}-\rho_f\mathbf{v}_f^{\circledcirc}\otimes\left(\mathbf{v}_f^{\circledcirc}-\mathbf{v}_s^{\circledcirc}\right)\right)^{\mathsf{S}}\mathbf{J}_s\mathbf{F}_s^{-\mathsf{T}}\mathbf{m}_s\cdot\mathbf{F}_s\mathbf{F}_{sf}^{-1}\delta\boldsymbol{\chi}_{sf}\right]\right]\\
&+\int_{S_s}\left[\left[\left(\phi_f p\mathbf{I}+\mathbf{T}_f-\frac{1}{2}\phi_f\mu_S\left[\mathbf{v}_{f\parallel}\right]\otimes\mathbf{m}\right)^{\mathsf{S}}\mathbf{J}_s\mathbf{F}_s^{-\mathsf{T}}\mathbf{m}_s\cdot\mathbf{F}_s\mathbf{F}_{sf}^{-1}\delta\boldsymbol{\chi}_{sf}\right]\right]. \tag*{(55)}
\end{aligned} 0 = ∫ B s [ J s ( ρ s a s ⊚ + ρ f a f ⊚ − b s − b f − ∇ ⋅ ( T s + T f ) ) S ⋅ δ χ s ] − ∫ B s f [ J s ( ρ f a f ⊚ − b f − ∇ ⋅ T f − p f ) S ⋅ F s F s f − 1 δ χ s f ] + ∫ S s [ [ ρ f ( v f ⊚ − v s ⊚ ) ⊗ ( v f ⊚ − v s ⊚ ) − T s − T f ] S J s F s − T m s ⋅ δ χ s ] + ∫ S s [ [ ( k f I − ρ f v f ⊚ ⊗ ( v f ⊚ − v s ⊚ ) ) S J s F s − T m s ⋅ F s F s f − 1 δ χ s f ] ] + ∫ S s [ [ ( ϕ f p I + T f − 2 1 ϕ f μ S [ v f ∥ ] ⊗ m ) S J s F s − T m s ⋅ F s F s f − 1 δ χ s f ] ] . (55)
By taking admissible variations of the fields δ χ s \delta\boldsymbol{\chi}_s δ χ s and δ χ s f \delta\boldsymbol{\chi}_{sf} δ χ s f , we then conclude that
ρ s a s ⊚ + ρ f a f ⊚ − b s − b f − ∇ ⋅ ( T s + T f ) = 0 in B ( t ) , (56) \rho_s\mathbf{a}_s^{\circledcirc}+\rho_f\mathbf{a}_f^{\circledcirc}-\mathbf{b}_s-\mathbf{b}_f-\nabla\cdot(\mathbf{T}_s+\mathbf{T}_f)=\mathbf{0}\quad\text{in }B(t), \tag*{(56)} ρ s a s ⊚ + ρ f a f ⊚ − b s − b f − ∇ ⋅ ( T s + T f ) = 0 in B ( t ) , (56)
ρ f a f ⊚ − b f − ∇ ⋅ T f − p f = 0 in B ( t ) , (57) \rho_f\mathbf{a}_f^{\circledcirc}-\mathbf{b}_f-\nabla\cdot\mathbf{T}_f-\mathbf{p}_f=\mathbf{0}\quad\text{in }B(t), \tag*{(57)} ρ f a f ⊚ − b f − ∇ ⋅ T f − p f = 0 in B ( t ) , (57)
[ [ ρ f ( v f ⊚ − v s ⊚ ) ⊗ ( v f ⊚ − v s ⊚ ) − T s − T f ] ] m = 0 on S ( t ) , (58) \left[\left[\rho_f\left(\mathbf{v}_f^{\circledcirc}-\mathbf{v}_s^{\circledcirc}\right)\otimes\left(\mathbf{v}_f^{\circledcirc}-\mathbf{v}_s^{\circledcirc}\right)-\mathbf{T}_s-\mathbf{T}_f\right]\right]\mathbf{m}=\mathbf{0}\quad\text{on }S(t), \tag*{(58)} [ [ ρ f ( v f ⊚ − v s ⊚ ) ⊗ ( v f ⊚ − v s ⊚ ) − T s − T f ] ] m = 0 on S ( t ) , (58)
and
[ [ F s f − T F s T ( k f I − ρ f v f ⊚ ⊗ ( v f ⊚ − v s ⊚ ) + ϕ f p I + T f − 1 2 ϕ f μ S [ v f ∥ ] ⊗ m ) S ] m S = 0 on S s . (59) [[\mathbf{F}_{sf}^{-\mathsf{T}}\mathbf{F}_s^{\mathsf{T}}(\mathbf{k}_f\mathbf{I}-\rho_f\mathbf{v}_f^{\circledcirc}\otimes(\mathbf{v}_f^{\circledcirc}-\mathbf{v}_s^{\circledcirc})+\phi_f p\mathbf{I}+\mathbf{T}_f-\frac{1}{2}\phi_f\mu_S[\mathbf{v}_{f\parallel}]\otimes\mathbf{m})^{\mathsf{S}}]\mathbf{m}^{\mathsf{S}}=\mathbf{0}\quad\text{on }S_s. \tag*{(59)} [[ F s f − T F s T ( k f I − ρ f v f ⊚ ⊗ ( v f ⊚ − v s ⊚ ) + ϕ f p I + T f − 2 1 ϕ f μ S [ v f ∥ ] ⊗ m ) S ] m S = 0 on S s . (59)
Remark 4.2 (Concerning Eqs. (4.11) and (4.12)). Referring to Eqs. (3.10) and (3.11), one can see that Eq. (4.11) expresses the balance of momentum for the mixture, the latter obtained by summing Eq. (3.10) for the two constituents and enforcing the constraint in Eq. (3.11); whereas Eq. (4.12) is directly the expression of the balance of momentum for the fluid phase. These equations correspond to Eqs. (28) and (29) in Ref. 18.
Remark 4.3 (Concerning (58) ). (58) expresses the jump condition of the balance of momentum for the mixture as a whole. This equation coincides with (26) .
Remark 4.4 (Concerning Eq. (4.14)). Equation (4.14) corresponds to Eq. (31) in Ref. 18 and it can be treated following the arguments presented therein to break up the equation into components. The object on the left-hand side of Eq. (4.14) takes values in the cotangent space of points on S f ( t ) S_f(t) S f ( t ) . One component of this quantity can be obtained by taking a convenient tangent vector to S f ( t ) S_f(t) S f ( t ) acting on it. To select such a vector, we recall that the continuity of χ s \boldsymbol{\chi}_s χ s implies:
[ [ F s ] ] τ s 1 = 0 , [ [ F s ] ] τ s 2 = 0 , [ [ J s F s − T m s ] ] = 0 , and [ [ v s ] ] = 0. (60) \left[\left[\mathbf{F}_s\right]\right]\boldsymbol{\tau}_{s_1}=\mathbf{0},\quad
\left[\left[\mathbf{F}_s\right]\right]\boldsymbol{\tau}_{s_2}=\mathbf{0},\quad
\left[\left[\mathbf{J}_s\mathbf{F}_s^{-\mathsf{T}}\mathbf{m}_s\right]\right]=\mathbf{0},\quad
\text{and}\quad
\left[\left[\mathbf{v}_s\right]\right]=\mathbf{0}. \tag*{(60)} [ [ F s ] ] τ s 1 = 0 , [ [ F s ] ] τ s 2 = 0 , [ [ J s F s − T m s ] ] = 0 , and [ [ v s ] ] = 0 . (60)
Furthermore, under the kinematic framework in Ref. 18, the following conditions also hold:
⟦ F f − 1 F s τ S 1 ⟧ = 0 , ⟦ F f − 1 F s τ S 2 ⟧ = 0 , and ⟦ J f − 1 J s ( F f − 1 F s ) − T m s ⟧ = 0. (61) \llbracket\mathbf{F}_{f}^{-1}\mathbf{F}_{s}\boldsymbol{\tau}_{S1}\rrbracket= \mathbf{0}, \qquad\llbracket\mathbf{F}_{f}^{-1}\mathbf{F}_{s}\boldsymbol{\tau}_{S2}\rrbracket= \mathbf{0}, \qquad\text{and} \qquad\llbracket\mathbf{J}_{f}^{-1}\mathbf{J}_{s}(\mathbf{F}_{f}^{-1}\mathbf{F}_{s})^{-T}\mathbf{m}_{s}\rrbracket= \mathbf{0}. \tag*{(61)} [ [ F f − 1 F s τ S 1 ] ] = 0 , [ [ F f − 1 F s τ S 2 ] ] = 0 , and [ [ J f − 1 J s ( F f − 1 F s ) − T m s ] ] = 0 . (61)
Then, letting τ \boldsymbol{\tau} τ be a tangent vector at x ^ ∈ S ( t ) \hat{\mathbf{x}} \in S(t) x ^ ∈ S ( t ) , a suitable tangent vector to S f ( t ) S_f(t) S f ( t ) would be F s f F s − 1 τ \mathbf{F}_{sf}\mathbf{F}_{s}^{-1}\boldsymbol{\tau} F s f F s − 1 τ and its action on the left-hand side of Eq. (59) , once remapped on S ( t ) S(t) S ( t ) , yields
τ ⋅ ( d ⟦ v f ⊖ ⟧ − ⟦ T f ⟧ m + 1 2 ⟦ ϕ f ⟧ μ S ⟦ v f t ⟧ ) = 0 on S ( t ) . \boldsymbol{\tau}\cdot\left(d\llbracket\mathbf{v}_{f}^{\ominus}\rrbracket-\llbracket\mathbf{T}_{f}\rrbracket\mathbf{m}+\frac{1}{2}\llbracket\phi_{f}\rrbracket\mu_{S}\llbracket\mathbf{v}_{ft}\rrbracket\right)=0 \quad\text{on } S(t). τ ⋅ ( d [ [ v f ⊖ ] ] − [ [ T f ] ] m + 2 1 [ [ ϕ f ] ] μ S [ [ v f t ] ] ) = 0 on S ( t ) .
The above equation coincides with the projection of the second of Eqs. (25) along τ \boldsymbol{\tau} τ . Letting P τ = I − m ⊗ m \mathbf{P}_{\tau}=\mathbf{I}-\mathbf{m}\otimes\mathbf{m} P τ = I − m ⊗ m denote the projection tangent to S ( t ) S(t) S ( t ) , the above equation can then be given in the more general form
P τ ( d ⟦ v f ⊖ ⟧ − ⟦ T f ⟧ m + 1 2 ⟦ ϕ f ⟧ μ S ⟦ v f t ⟧ ) = 0 on S ( t ) . (62) \mathbf{P}_{\tau}\left(d\llbracket\mathbf{v}_{f}^{\ominus}\rrbracket-\llbracket\mathbf{T}_{f}\rrbracket\mathbf{m}+\frac{1}{2}\llbracket\phi_{f}\rrbracket\mu_{S}\llbracket\mathbf{v}_{ft}\rrbracket\right)=\mathbf{0} \quad\text{on } S(t). \tag*{(62)} P τ ( d [ [ v f ⊖ ] ] − [ [ T f ] ] m + 2 1 [ [ ϕ f ] ] μ S [ [ v f t ] ] ) = 0 on S ( t ) . (62)
Equation (62) is useful in providing clarity as to the nature of the surface force ξ f \boldsymbol{\xi}_{f} ξ f introduced earlier. Clearly 1 2 ⟦ ϕ f ⟧ μ S P τ ⟦ v f t ⟧ \frac{1}{2}\llbracket\phi_{f}\rrbracket\mu_{S}\mathbf{P}_{\tau}\llbracket\mathbf{v}_{ft}\rrbracket 2 1 [ [ ϕ f ] ] μ S P τ [ [ v f t ] ] is only a component of ξ f \boldsymbol{\xi}_{f} ξ f . As pointed out in Ref. 18, if d ≠ 0 d\ne0 d = 0 , another (independent) component can be derived by considering the action of the vector F s f F s − 1 ( v f ⊖ − v s ⊖ ) \mathbf{F}_{sf}\mathbf{F}_{s}^{-1}(\mathbf{v}_{f}^{\ominus}-\mathbf{v}_{s}^{\ominus}) F s f F s − 1 ( v f ⊖ − v s ⊖ ) (cf. the last of Eqs. (8) ) onto the left-hand side of Eq. (59) . The vector in question is continuous across S s S_s S s and not tangent to S f S_f S f when d ≠ 0 d\ne0 d = 0 .
With our choice of test functions, then the following result is obtained:
( k f m − d v f ⊕ + ϕ f p m + T f m − 1 2 ϕ f μ S ⟦ v f t ⟧ ) ± = 0 on S ( t ) . (63) \left(\mathbf{k}_{f}\mathbf{m}-\mathrm{d}\mathbf{v}_{f}^{\oplus}+\phi_{f}\mathbf{p}\mathbf{m}+\mathbf{T}_{f}\mathbf{m}-\frac{1}{2}\phi_{f}\mu_{S}\llbracket\mathbf{v}_{f\mathrm{t}}\rrbracket\right)^{\pm}=0\quad\text{on }\mathcal{S}(t). \tag*{(63)} ( k f m − d v f ⊕ + ϕ f pm + T f m − 2 1 ϕ f μ S [ [ v f t ] ] ) ± = 0 on S ( t ) . (63)
We note that, under the assumption that d = 0 d=0 d = 0 , Eqs. (4.18) have similarities with Eqs. (35) in Ref. 18. Next, referring to the second of Eqs. (3.14), using Eqs. (4.18), we would conclude that
ξ f = 1 2 μ S ⟦ ϕ f ⟧ ⟦ v f t ⟧ − ⟦ k f + ϕ f p ⟧ m . (64) \boldsymbol{\xi}_{f}=\frac{1}{2}\mu_{S}\llbracket\phi_{f}\rrbracket\llbracket\mathbf{v}_{f\mathrm{t}}\rrbracket-\llbracket\mathbf{k}_{f}+\phi_{f}\mathbf{p}\rrbracket\mathbf{m}. \tag*{(64)} ξ f = 2 1 μ S [ [ ϕ f ] ] [ [ v f t ] ] − [ [ k f + ϕ f p ] ] m . (64)
The above form of ξ f \boldsymbol{\xi}_{f} ξ f is such that the first term on the right-hand side of Eq. (4.19) is tangent to the interface whereas the second term is normal to it.
A practical weak formulation
Here we propose a weak formulation of the problem based on Eq. (4.10) and the kinematic framework adopted in Remark 4.5. For the purpose of this paper, we will use the same simple constitutive assumptions made in the discussion of Hamilton’s principle. Clearly, these assumptions can be changed to fit the modeling needs of a particular problem.
With the above in mind, the principal unknowns of our problem are selected as follows:
u s , v f ∘ S , p ∘ S , and p ∘ S . (65) \mathbf{u}_{s},\quad\mathbf{v}_{f}^{\circ\mathrm{S}},\quad p^{\circ\mathrm{S}},\quad\text{and}\quad\mathbf{p}^{\circ\mathrm{S}}. \tag*{(65)} u s , v f ∘ S , p ∘ S , and p ∘ S . (65)
This choice of principal unknowns is meant to produce a formulation that is close to familiar FEM formulations of poroelasticity. That is, we do not perceive that a FEM in which the primary unknowns are χ s \chi_{s} χ s and χ s f \chi_{sf} χ s f would be adaptable to most applications. As for the particular form of the list, Eq. (4.20) is meant to indicate that we are proposing an arbitrary Lagrangian–Eulerian (ALE) formulation where the field variables are understood to be defined over B s \mathcal{B}_{s} B s , the latter domain coinciding with our computational domain. At the same time, to avoid unnecessary repetition, we will keep using the same symbols as in the formulation section of the paper. Also, to avoid proliferation of symbols, we denote the test functions of said fields by a tilde.
While the regularity of these fields has been discussed earlier in the paper, we recognize that the formal properties of these fields must be examined on a case by case basis depending on the specific set of constitutive equations that are selected for a corresponding specific problem. In fact, the constitutive equations also have repercussions to the type of boundary conditions that one might impose, as it is clear from the comparison between, say, the traditional Stokes and Darcy problems. With this in mind, and referring to Eq. (4.10), we operate the following substitution of test functions:
δ χ s → u ~ s / θ and − F s F s f − 1 δ χ s f → v ~ f , (66) \delta\boldsymbol{\chi}_{s}\to\widetilde{\mathbf{u}}_{s}/\theta\quad\text{and}\quad-\mathbf{F}_{s}\mathbf{F}_{sf}^{-1}\delta\boldsymbol{\chi}_{sf}\to\widetilde{\mathbf{v}}_{f}, \tag*{(66)} δ χ s → u s / θ and − F s F s f − 1 δ χ s f → v f , (66)
where θ \theta θ is a unit time constant whose only purpose is to achieve dimensional consistency. We then propose the following weak formulation of the problem, which incorporates a weak enforcement of the constraints in the second of Eqs. (18) and the first of Eqs. (19) . For simplicity of presentation, we assume that Dirichlet boundary conditions are prescribed for us and vf on the external boundary of the mixture. Such weak formulation is then given by
0 = ∫ B s J s ( ρ s a s ⊕ + ρ f a f ⊕ − b s − b f + ∇ p ) s ⋅ u ~ s / θ + ∫ B s J s ( T s e + T s v + T f v ) F s − T : ∇ u ~ s / θ + ∫ B s J s [ ( ρ f a f ⊕ − b f + ϕ f ∇ p − f f ) ⋅ v ~ f s + ( T f v ) s F s − T : ∇ v ~ f s ] + ∫ S s [ ϕ f ( v f ⊕ − v s ⊕ ) ⊗ ( v f ⊕ − v s ⊕ ) + p I ] s J s F s − T m s ⋅ u ~ s / θ − ∫ S s [ ( k f I − ρ f v f ⊕ ⊗ ( v f ⊕ − v s ⊕ ) ) s J s F s − T m s ⋅ v ~ f s ] − ∫ S s [ ( ϕ f ( p − p ) I − 1 2 ϕ f μ S [ ∇ v f l t ] ⊗ m ) s J s F s − T m s ⋅ v ~ f s ] − ∫ B s J s ( v s + v f l t s ) ⋅ F s − T ∇ p ~ s − ∫ S s [ ( v s + v f l t s ) p ~ s ] ⋅ J s F s − T m s + ∫ S s [ ∇ v f l t s ] p ~ s ⋅ J s F s − T m s + ∫ ∂ B s e x t p ~ s ( v s + v f l t s ) ⋅ J s F s − T n s . (67) \begin{aligned}
0={}&\int_{B_s}J_s\left(\rho_s a_s^{\oplus}+\rho_f a_f^{\oplus}-\mathbf b_s-\mathbf b_f+\nabla p\right)^{\mathsf{s}}\cdot\widetilde{\mathbf u}_s/\theta\\
&+\int_{B_s}J_s\left(\mathbf T_s^{\mathsf e}+\mathbf T_s^{\mathsf v}+\mathbf T_f^{\mathsf v}\right)\mathbf F_s^{-\mathsf T}:\nabla\widetilde{\mathbf u}_s/\theta\\
&+\int_{B_s}J_s\left[\left(\rho_f a_f^{\oplus}-\mathbf b_f+\phi_f\nabla p-\mathbf f_f\right)\cdot\widetilde{\mathbf v}_f^{\mathsf{s}}+\left(\mathbf T_f^{\mathsf v}\right)^{\mathsf{s}}\mathbf F_s^{-\mathsf T}:\nabla\widetilde{\mathbf v}_f^{\mathsf{s}}\right]\\
&+\int_{S_s}\left[\phi_f\left(\mathbf v_f^{\oplus}-\mathbf v_s^{\oplus}\right)\otimes\left(\mathbf v_f^{\oplus}-\mathbf v_s^{\oplus}\right)+p\mathbf I\right]^{\mathsf{s}}J_s\mathbf F_s^{-\mathsf T}\mathbf m_s\cdot\widetilde{\mathbf u}_s/\theta\\
&-\int_{S_s}\left[\left(k_f\mathbf I-\rho_f\mathbf v_f^{\oplus}\otimes\left(\mathbf v_f^{\oplus}-\mathbf v_s^{\oplus}\right)\right)^{\mathsf{s}}J_s\mathbf F_s^{-\mathsf T}\mathbf m_s\cdot\widetilde{\mathbf v}_f^{\mathsf{s}}\right]\\
&-\int_{S_s}\left[\left(\phi_f(\mathbf p-p)\mathbf I-\frac{1}{2}\phi_f\mu_S[\nabla\mathbf v_{\mathrm{flt}}]\otimes\mathbf m\right)^{\mathsf{s}}J_s\mathbf F_s^{-\mathsf T}\mathbf m_s\cdot\widetilde{\mathbf v}_f^{\mathsf{s}}\right]\\
&-\int_{B_s}J_s(\mathbf v_s+\mathbf v_{\mathrm{flt}}^{\mathsf{s}})\cdot\mathbf F_s^{-\mathsf T}\nabla\widetilde p^{\mathsf{s}}-\int_{S_s}\left[(\mathbf v_s+\mathbf v_{\mathrm{flt}}^{\mathsf{s}})\widetilde p^{\mathsf{s}}\right]\cdot J_s\mathbf F_s^{-\mathsf T}\mathbf m_s\\
&+\int_{S_s}[\nabla\mathbf v_{\mathrm{flt}}^{\mathsf{s}}]\widetilde p^{\mathsf{s}}\cdot J_s\mathbf F_s^{-\mathsf T}\mathbf m_s+\int_{\partial B_s^{\mathrm{ext}}}\widetilde p^{\mathsf{s}}(\mathbf v_s+\mathbf v_{\mathrm{flt}}^{\mathsf{s}})\cdot J_s\mathbf F_s^{-\mathsf T}\mathbf n_s . \tag*{(67)}
\end{aligned} 0 = ∫ B s J s ( ρ s a s ⊕ + ρ f a f ⊕ − b s − b f + ∇ p ) s ⋅ u s / θ + ∫ B s J s ( T s e + T s v + T f v ) F s − T : ∇ u s / θ + ∫ B s J s [ ( ρ f a f ⊕ − b f + ϕ f ∇ p − f f ) ⋅ v f s + ( T f v ) s F s − T : ∇ v f s ] + ∫ S s [ ϕ f ( v f ⊕ − v s ⊕ ) ⊗ ( v f ⊕ − v s ⊕ ) + p I ] s J s F s − T m s ⋅ u s / θ − ∫ S s [ ( k f I − ρ f v f ⊕ ⊗ ( v f ⊕ − v s ⊕ ) ) s J s F s − T m s ⋅ v f s ] − ∫ S s [ ( ϕ f ( p − p ) I − 2 1 ϕ f μ S [ ∇ v flt ] ⊗ m ) s J s F s − T m s ⋅ v f s ] − ∫ B s J s ( v s + v flt s ) ⋅ F s − T ∇ p s − ∫ S s [ ( v s + v flt s ) p s ] ⋅ J s F s − T m s + ∫ S s [ ∇ v flt s ] p s ⋅ J s F s − T m s + ∫ ∂ B s ext p s ( v s + v flt s ) ⋅ J s F s − T n s . (67)
where ∂ B s e x t \partial B_s^{\mathrm{ext}} ∂ B s ext denotes the external boundary of B s B_s B s .
Proposition 4.1 (Formal stability estimate). Setting to zero the body forces and the boundary data, if, for almost every t ∈ [ 0 , T ] t\in[0,T] t ∈ [ 0 , T ] , u s \mathbf u_s u s , v f s \mathbf v_f^{\mathsf{s}} v f s , p s p^{\mathsf{s}} p s , and ψ s \psi^{\mathsf{s}} ψ s solve Eq. (55) , then the following energy estimate holds:
d d t ∫ B ( t ) ( k s + k f + ψ s ) + ∫ B ( t ) D + ∫ S ( t ) D S ( t ) = 0. (68) \frac{\mathrm d}{\mathrm dt}\int_{B(t)}(k_s+k_f+\psi_s)+\int_{B(t)}\mathcal D+\int_{S(t)}\mathcal D_{S(t)}=0. \tag*{(68)} d t d ∫ B ( t ) ( k s + k f + ψ s ) + ∫ B ( t ) D + ∫ S ( t ) D S ( t ) = 0. (68)
Proof. Set
u ~ s / θ = v s , v ~ f s = v f s − v s , p ~ s = p s , and p ~ s = p s . \widetilde{\mathbf u}_s/\theta=\mathbf v_s,\qquad
\widetilde{\mathbf v}_f^{\mathsf{s}}=\mathbf v_f^{\mathsf{s}}-\mathbf v_s,\qquad
\widetilde p^{\mathsf{s}}=p^{\mathsf{s}},\qquad\text{and}\qquad
\widetilde p^{\mathsf{s}}=p^{\mathsf{s}}. u s / θ = v s , v f s = v f s − v s , p s = p s , and p s = p s .
This substitution yields
0 = ∫ B s J s ( ρ s a ˚ s e ˚ + ρ f a ˚ f e ˚ ) S ˚ ⋅ v s + ∫ B s J s ( T s e + T s v + T f v ) S ˚ F s − T : ∇ v s + ∫ B s [ J s ( ρ f a f e ˚ − f f ) S ˚ ⋅ ( v f S ˚ − v s ) + ( T f v ) S ˚ F s − T : ∇ ( v f S ˚ − v s ) ] + ∫ S s [ [ ρ f ( v f e ˚ − v s e ˚ ) ⊗ ( v f e ˚ − v s e ˚ ) ] S ˚ J s F s − T m s ⋅ v s ] − ∫ S s [ ( k f I − ρ f v f e ˚ ⊗ ( v f e ˚ − v s e ˚ ) ) S ˚ J s F s − T m s ⋅ ( v f S ˚ − v s ) ] + ∫ S ( t ) 1 2 μ S ∣ v f Π t ∣ ⋅ ∣ v f Π t ∣ . (69) \begin{aligned}
0 ={}& \int_{B_s} J_s\left(\rho_s \mathring{a}_s^{\mathring{e}}+\rho_f \mathring{a}_f^{\mathring{e}}\right)^{\mathring{S}}\cdot v_s \\
&+\int_{B_s} J_s\left(T_s^{e}+T_s^{v}+T_f^{v}\right)^{\mathring{S}}F_s^{-T}:\nabla v_s \\
&+\int_{B_s}\left[J_s\left(\rho_f a_f^{\mathring{e}}-f_f\right)^{\mathring{S}}\cdot\left(v_f^{\mathring{S}}-v_s\right)+\left(T_f^{v}\right)^{\mathring{S}}F_s^{-T}:\nabla\left(v_f^{\mathring{S}}-v_s\right)\right] \\
&+\int_{S_s}\left[\left[\rho_f\left(v_f^{\mathring{e}}-v_s^{\mathring{e}}\right)\otimes\left(v_f^{\mathring{e}}-v_s^{\mathring{e}}\right)\right]^{\mathring{S}}J_sF_s^{-T}m_s\cdot v_s\right] \\
&-\int_{S_s}\left[\left(k_f I-\rho_f v_f^{\mathring{e}}\otimes\left(v_f^{\mathring{e}}-v_s^{\mathring{e}}\right)\right)^{\mathring{S}}J_sF_s^{-T}m_s\cdot\left(v_f^{\mathring{S}}-v_s\right)\right] \\
&+\int_{S(t)}\frac{1}{2}\mu_S\left|v_{f\Pi t}\right|\cdot\left|v_{f\Pi t}\right|. \tag*{(69)}
\end{aligned} 0 = ∫ B s J s ( ρ s a ˚ s e ˚ + ρ f a ˚ f e ˚ ) S ˚ ⋅ v s + ∫ B s J s ( T s e + T s v + T f v ) S ˚ F s − T : ∇ v s + ∫ B s [ J s ( ρ f a f e ˚ − f f ) S ˚ ⋅ ( v f S ˚ − v s ) + ( T f v ) S ˚ F s − T : ∇ ( v f S ˚ − v s ) ] + ∫ S s [ [ ρ f ( v f e ˚ − v s e ˚ ) ⊗ ( v f e ˚ − v s e ˚ ) ] S ˚ J s F s − T m s ⋅ v s ] − ∫ S s [ ( k f I − ρ f v f e ˚ ⊗ ( v f e ˚ − v s e ˚ ) ) S ˚ J s F s − T m s ⋅ ( v f S ˚ − v s ) ] + ∫ S ( t ) 2 1 μ S ∣ v f Π t ∣ ⋅ ∣ v f Π t ∣ . (69)
Substituting Eqs. (36) and (37) in the above expression, and using the definitions in Eqs. (41) and (54) , we have
0 = ∫ B ( t ) ( ρ s a s e ˚ ⋅ v s e ˚ + ρ f a f e ˚ ⋅ v f e ˚ ) + ∫ B ( t ) J s ( T s e ) S ˚ F s − T : ∇ v s + ∫ B ( t ) D S ˚ + ∫ S ( t ) D S ( t ) . (70) \begin{aligned}
0={}&\int_{B(t)}\left(\rho_s a_s^{\mathring{e}}\cdot v_s^{\mathring{e}}+\rho_f a_f^{\mathring{e}}\cdot v_f^{\mathring{e}}\right)+\int_{B(t)}J_s\left(T_s^{e}\right)^{\mathring{S}}F_s^{-T}:\nabla v_s\\
&+\int_{B(t)}D^{\mathring{S}}+\int_{S(t)}D_{S(t)}. \tag*{(70)}
\end{aligned} 0 = ∫ B ( t ) ( ρ s a s e ˚ ⋅ v s e ˚ + ρ f a f e ˚ ⋅ v f e ˚ ) + ∫ B ( t ) J s ( T s e ) S ˚ F s − T : ∇ v s + ∫ B ( t ) D S ˚ + ∫ S ( t ) D S ( t ) . (70)
Using the results in Propositions B.1 and B.3 in Appendix B, Eq. (68) follows. □
Remark 4.6 (Consistency of the energy estimate). The energy estimate just proven is consistent with that in Eq. (43) when Eq. (45) is used and every supply of power to the system is suppressed. We note that, for any physically admissible formulation, and excluding the trivial case of rigid body motions, the last two terms on the left-hand side of Eq. (68) are positive for any motion of the fluid relative to the solid, thereby ensuring a monotonic decline of the sum of the kinetic, strain, and stored energies of the system.
Remark 4.7 (Analysis and FEM). We are not aware of a formal analysis of this specific problem articulated in the language of Sobolev spaces. Nor are we aware of an analysis within the context of the FEM. We have implemented Eq. (67) using standard FEM approaches typical of ALE formulations. The interface S s S_s S s was simply modeled in a conformal way, namely, it consisted of inter-element boundaries. The chosen interpolation functions are as follows:
2nd order Lagrange polynomials for the (globally continuous) displacement field.
2nd order Lagrange polynomials for the fluid velocity field. This field is not continuous across the interface.
1st order Lagrange polynomials for the field p p p , which is not continuous across the interface.
The appropriateness of this selection of finite element spaces is supported by a recent work by Barnafi et al., [4 ] which, along with Ref. 3, offers some guidance for a rigorous analysis of the present work.
Results
In this section, we present three sets of results. We begin by verifying our FEM implementation using the method of manufactured solutions.[46 ] We then present our analytical solution to a classic problem, to compare our results with existing ones. Finally, we solve an example problem to investigate the effect of the interface viscosity μ S \mu_S μ S on the jumps in the tangential component of filtration velocity and the pore pressure.
The numerical implementation of the formulation presented in this paper was done in COMSOL Multiphysics®,[15 ] which has been used as a FEM programming environment. We used a monolithic solver. For time integration we used the available variable time step and variable order backward differentiation formula (BDF) method.[42 ] We have already mentioned the specific interpolation used. Additional details for the results presented will be indicated in context.
Verification
To verify the formulation, we employed the method of manufactured solutions.[46 ] We consider the planar geometry displayed in Figure 1 , which consists of two contiguous
Figure 1. Two contiguous square poroelastic domains A and B used for the verification of the FEM implementation of our formulation through the method of manufactured solutions.
poroelastic domains with reference configurations A and B, respectively, aligned vertically and whose material properties are presented in Table 1 . Referring to
Table 1. Summary of parameter values used in the convergence rate study and in the last example.
Figure 1 , the position X \mathbf{X} X of points in the reference configuration is expressed through the coordinates X X X and Y Y Y . The position x \mathbf{x} x of points in the deformed configuration is expressed through the coordinates x = X + u s X ( X , Y , t ) x = X + u_{sX}(X,Y,t) x = X + u s X ( X , Y , t ) and y = Y + u s Y ( X , Y , t ) y = Y + u_{sY}(X,Y,t) y = Y + u s Y ( X , Y , t ) , where u s X u_{sX} u s X and u s Y u_{sY} u s Y are the horizontal and vertical components of the displacement field. The images of A and B under the motion will be denoted by A( t ) (t) ( t ) and B( t ) (t) ( t ) , respectively. The interface’s reference configuration S s \mathcal{S}_s S s is the segment with X ∈ [ 0 , L ] X \in[0,L] X ∈ [ 0 , L ] and Y = L Y = L Y = L . For both domains, the neo-Hookean hyperelastic material model
Ψ ^ s = 1 2 μ R e ( I ˉ 1 − 3 ) \widehat{\Psi}_s = \frac{1}{2}\mu_{R_e}\left(\bar{I}_1 - 3\right) Ψ s = 2 1 μ R e ( I ˉ 1 − 3 )
is chosen for the strain energy density of the pure solid phase, where I ˉ 1 \bar{I}_1 I ˉ 1 is the first principal invariant of C ˉ s = J s − 2 / 3 C s \bar{\mathbf{C}}_s = J_s^{-2/3}\mathbf{C}_s C ˉ s = J s − 2/3 C s , namely the isochoric component of the right Cauchy-Green strain tensor C s \mathbf{C}_s C s . μ R e \mu_{R_e} μ R e is the referential solid’s elastic shear modulus. Moreover, it is assumed that μ f = μ ˉ f ϕ f 2 \mu_f = \bar{\mu}_f\phi_f^2 μ f = μ ˉ f ϕ f 2 and μ B = μ ˉ B ϕ f 2 \mu_B = \bar{\mu}_B\phi_f^2 μ B = μ ˉ B ϕ f 2 .
The manufactured solution consists of the following fields:
p = p o ( 1 − cos ( π t t o ) ) cos [ π ( x + y ) L ] in A ( t ) , (71) p = p_o\left(1-\cos\left(\frac{\pi t}{t_o}\right)\right)\cos\left[\frac{\pi(x+y)}{L}\right] \qquad\text{in } A(t), \tag*{(71)} p = p o ( 1 − cos ( t o π t ) ) cos [ L π ( x + y ) ] in A ( t ) , (71)
p = p o ( 1 − cos ( π t t o ) ) sin [ π ( x + y ) L ] in B ( t ) , (72) p = p_o\left(1-\cos\left(\frac{\pi t}{t_o}\right)\right)\sin\left[\frac{\pi(x+y)}{L}\right] \qquad\text{in } B(t), \tag*{(72)} p = p o ( 1 − cos ( t o π t ) ) sin [ L π ( x + y ) ] in B ( t ) , (72)
u s X = u o ( 1 − cos ( π t t o ) ) cos [ π ( X + Y ) L ] in A ∪ B , (73) u_{sX} = u_o\left(1-\cos\left(\frac{\pi t}{t_o}\right)\right)\cos\left[\frac{\pi(X+Y)}{L}\right] \qquad\text{in } A\cup B, \tag*{(73)} u s X = u o ( 1 − cos ( t o π t ) ) cos [ L π ( X + Y ) ] in A ∪ B , (73)
u s Y = u o ( 1 − cos ( π t t o ) ) sin [ π ( X + Y ) L ] in A ∪ B , (74) u_{sY} = u_o\left(1-\cos\left(\frac{\pi t}{t_o}\right)\right)\sin\left[\frac{\pi(X+Y)}{L}\right] \qquad\text{in } A\cup B, \tag*{(74)} u s Y = u o ( 1 − cos ( t o π t ) ) sin [ L π ( X + Y ) ] in A ∪ B , (74)
v f ⊖ = ( ∂ t u s ) ⊖ − ( ϕ f γ ) ∇ p in A ( t ) ∪ B ( t ) , (75) \mathbf{v}_f^\ominus= (\partial_t\mathbf{u}_s)^\ominus- \left(\frac{\phi_f}{\gamma}\right)\nabla p \qquad\text{in } A(t)\cup B(t), \tag*{(75)} v f ⊖ = ( ∂ t u s ) ⊖ − ( γ ϕ f ) ∇ p in A ( t ) ∪ B ( t ) , (75)
p = p o ( 1 − cos ( π t t o ) ) cos ( π x L ) on S ( t ) , (76) p = p_o\left(1-\cos\left(\frac{\pi t}{t_o}\right)\right)\cos\left(\frac{\pi x}{L}\right) \qquad\text{on } \mathcal{S}(t), \tag*{(76)} p = p o ( 1 − cos ( t o π t ) ) cos ( L π x ) on S ( t ) , (76)
where γ = ( μ D ϕ f 2 ) / k s \gamma= (\mu_D\phi_f^2)/k_s γ = ( μ D ϕ f 2 ) / k s , u o = 1 m m u_o = 1\ \mathrm{mm} u o = 1 mm , p o = 2.5 k P a p_o = 2.5\ \mathrm{kPa} p o = 2.5 kPa , p o = 1 k P a p_o = 1\ \mathrm{kPa} p o = 1 kPa , L = 10 m m L = 10\ \mathrm{mm} L = 10 mm , and t o = 1 s t_o = 1\ \mathrm{s} t o = 1 s . These fields are substituted in the strong form of the governing equations to calculate the source terms in the interiors of their domains of definition as well as on the interface.
The initial conditions are those dictated by the manufactured solution for t = 0 t = 0 t = 0 . For the solid part of the problem, the boundary conditions consist of prescribed displacements on the entirety of the external boundary; for the fluid part, traction boundary conditions are prescribed on the very top and bottom edges of the external boundary; the fluid’s velocity is prescribed on the lateral surfaces.
We have already indicated the choice of finite element spaces for our solutions. As far as the time integration part of the problem, we employed a BDF method of order 2 with a constant time step of 0.001 s.
The selected manufactured solution in terms of solid displacement, fluid velocity, and pore pressure is depicted in Figure 2 at t = 0.1 t = 0.1 t = 0.1 s. This figure shows that indeed
Figure 2. Panels (a) and (b) show the manufactured solutions for components of the displacement of the solid phase. Panels (c) and (d) show the components of the velocity of the fluid phase. Panel (e) shows the pore pressure. All plots are at time t = 0.1 t = 0.1 t = 0.1 s. For simplicity, all the plots are over the reference configuration of the solid phase.
the selected solution is globally continuous in the displacement field, whereas it is discontinuous in the fluid velocity field and in the pore pressure field.
The mesh employed was made up of identical triangles and was uniformly refined through four cycles. The initial mesh diameter was h = 1.25 h = 1.25 h = 1.25 mm. Figure 3 presents the convergence rates under uniform refinement. Specifically,
Figure 3. Convergence rates for (a) the L 2 L^2 L 2 -norm and (b) the H 1 H^1 H 1 -semi-norm of the relative error e e e between numerical and exact solution in all variables, at time t = 0.1 t = 0.1 t = 0.1 s, obtained using second-order Lagrange polynomials for the solid displacement and fluid velocity and first-order Lagrange polynomials for the pressure and the interface pressure. The computation of the norms of the relative errors is done via integrals that respect the domain of definition of said errors (e.g., the norms of the relative errors pertaining to the ( v f A ) ∘ (v_{\mathrm{f}}^{\mathrm{A}})^{\circ} ( v f A ) ∘ are computed over A ( t ) A(t) A ( t ) ).
we report the L 2 L^2 L 2 -norm and the H 1 H^1 H 1 -semi-norm of the relative errors for each of the quantities specified in the manufactured solution. As the functions indicated in Eqs. (71) –(76) each have their own domain of definition, the relative errors for these functions are computed via integrals that are consistent with said domains.
Provided that a formal analysis is not available so as to assess the optimality of the convergence rates, the convergence rates estimated here are rather good when compared to those typically obtained for the Navier-Stokes problem. We find this result rather encouraging, especially in view of the fact that we have relied on readily available numerical tools as opposed to devise custom-made algorithms for this problem.
A classic problem
To test our mathematical framework, we analytically solved a benchmark problem involving Poiseuille flow over a porous medium. In the Beavers-Joseph problem, a viscous fluid flows under a constant pressure gradient in a channel where the wall at the fluid-porous medium interface is permeable,[5 , 19 , 48 ] as illustrated in Figure 4 .
Figure 4. Schematic view of viscous fluid flow over a rigid porous medium, driven by a constant pressure gradient.
and the fluid velocity at these boundaries is set to zero. The flow inside the channel and the porous medium is only in the x x x direction. Among others, this problem was previously considered by Hou et al.^{27} in formulating boundary conditions at the interface between the pure fluid and the porous medium. In their study, they introduced a “pseudo-no-slip” kinematic interface condition based on the idea that the conditions at the interface between mixtures, or between mixtures and fluids, should reduce to those of single-phase continuum mechanics. They mentioned that, to formulate a well-posed mathematical problem for each subdomain of the fluid-solid interaction, it is necessary not only to satisfy the jump conditions for mass and momentum conservation, but also to impose additional kinematic constraints. They chose to enforce the continuity of the pore pressure and the full continuity of the filtration velocity.
Here we solved the same problem but enforcing the jump conditions that were obtained in Eqs. (63) .
We begin by observing that the unidirectional nature of the fluid flow and the rigidity of the porous medium imply that the mass balance is trivially satisfied. The dimensionless form of the balance of linear momentum can then be derived for both the fluid channel and the porous medium, and is given by, respectively,
∂ 2 V u ∂ Z 2 + 1 = 0 , for 0 < Z < H , (77) \frac{\partial^2 V_u}{\partial Z^2} + 1 = 0,\qquad\text{for } 0 < Z < H, \tag*{(77)} ∂ Z 2 ∂ 2 V u + 1 = 0 , for 0 < Z < H , (77)
and
δ 2 ∂ 2 V l ∂ Z 2 − V l + η δ 2 = 0 , for − 1 < Z < 0 , (78) \delta^2 \frac{\partial^2 V_l}{\partial Z^2} - V_l + \eta\delta^2 = 0,\qquad\text{for } -1 < Z < 0, \tag*{(78)} δ 2 ∂ Z 2 ∂ 2 V l − V l + η δ 2 = 0 , for − 1 < Z < 0 , (78)
where the latter is the simplified version of Eq. (57) and the former is the dimensionless form of the Navier-Stokes equation in the absence of inertia and body forces. Here, V u ( Z ) V_u(Z) V u ( Z ) and V l ( Z ) V_l(Z) V l ( Z ) are the dimensionless fluid velocities in the channel and the porous medium, respectively. Additionally, Z = z / h 2 Z = z/h_2 Z = z / h 2 , H = h 1 / h 2 H = h_1/h_2 H = h 1 / h 2 , η = ϕ f 2 μ f / μ a \eta= \phi_f^2\mu_f/\mu_a η = ϕ f 2 μ f / μ a , and δ = μ a / ( h 2 2 κ s ) \delta= \sqrt{\mu_a/(h_2^2\kappa_s)} δ = μ a / ( h 2 2 κ s ) are dimensionless parameters, where μ a \mu_a μ a represents the apparent viscosity of the fluid in the porous medium. Since the top and bottom boundaries are fixed and impermeable, the fluid velocity is zero at these boundaries. For the interface between the porous medium and the viscous fluid, the conditions represented in Eqs. (63) for the current problem can be simplified as
V u ′ − 1 2 γ ( V u − V l ) = 0 , for Z → 0 + , (79) V_u' - \frac{1}{2}\gamma\left(V_u - V_l\right) = 0,\quad\text{for } Z \to0^+, \tag*{(79)} V u ′ − 2 1 γ ( V u − V l ) = 0 , for Z → 0 + , (79)
V l ′ − 1 2 γ η ( V u − V l ) = 0 , for Z → 0 − , (80) V_l' - \frac{1}{2}\gamma\eta\left(V_u - V_l\right) = 0,\quad\text{for } Z \to0^-, \tag*{(80)} V l ′ − 2 1 γ η ( V u − V l ) = 0 , for Z → 0 − , (80)
where γ = μ s h 2 μ f \gamma= \frac{\mu_s h_2}{\mu_f} γ = μ f μ s h 2 is a dimensionless parameter representing the ratio of the interface viscosity to the fluid viscosity. Solving the governing equations along with the boundary conditions yields an analytical solution for the fluid velocity in both the channel and the porous medium.
Figure 5 and Figure 6 compare our analytical solution and the analytical solution in Hou et al.[27 ] . Our solution is predicated on the use of a relatively large value of γ \gamma γ , which can be interpreted as representing a strongly dissipative interface. In this case, we show a strong agreement with the results reported by Hou et al.[27 ] , who had enforced continuity of both the pore pressure and the entire filtration velocity across the interface as the implementation of what these authors call a “pseudo-noslip” interface condition. With this result we show that the pseudo-no-slip condition, which is a kinematic constraint, can be viewed as the manifestation of a constitutive assumption on the interface behavior. Furthermore, we show that it is not necessary the continuity of the pressure in a strict sense to obtain the above results.
Figure 5. Comparison of profiles of fluid flux in the porous medium and the fluid channel with η = 1 \eta= 1 η = 1 and γ = 1000 \gamma= 1000 γ = 1000 , for different values of δ \delta δ , and for (a) H = 1 H = 1 H = 1 and (b) H = 0.25 H = 0.25 H = 0.25 . The lines represent results according to the formulation presented in this paper whereas the hollow circles represent the solution in Ref. 27.
Figure 6. Comparison of profiles of fluid flux in the porous medium and the fluid channel with η = 2 \eta= 2 η = 2 and γ = 1000 \gamma= 1000 γ = 1000 , for different values of δ \delta δ , and for (a) H = 1 H = 1 H = 1 and (b) H = 0.25 H = 0.25 H = 0.25 . The lines represent results according to the formulation presented in this paper whereas the hollow circles represent the solution in Ref. 27.
Role of interface viscosity
To investigate the role of the interface viscosity, we set up a numerical example with the geometry shown in Figure 7 . The parameters used are reported in Table 1 , with the permeability of both subdomains increased by two orders of magnitude.
Figure 7. The example problem setup to study the effect of interface viscosity on the discontinuities of filtration velocity and pore pressure. The radius of domain B is L / 4 L/4 L /4 , where L L L has the same value as used in the verification problem.
In this example problem, the top and bottom boundaries are impermeable, while the lateral boundaries are permeable and traction-free. The bottom boundary is fixed in the vertical direction, and a prescribed vertical displacement u S y = u o ( t / t o ) 2 u_{S y} = u_o(t/t_o)^2 u S y = u o ( t / t o ) 2 is applied to the top boundary, with u o u_o u o and t o t_o t o having the same values used in the manufactured solution considered earlier. Both the top and bottom boundaries are free to move in the horizontal direction. We suppress the rigid body motions of the domain A. Moreover, we adopt quiescent initial conditions.
Figure 8 (a) shows the effect of interface viscosity on the jump in the tangential component of the filtration velocity. As the interface viscosity increases, the norm of the jump of the tangential component of the filtration velocity decreases. Figure 8 (b) shows the jump in the pressure for different values of the interface viscosity. In this case, the jump in pressure does not seem to vanish. Furthermore, these observations are in agreement with the results in Figure 9 , displaying the L 2 L^2 L 2 -norm of the two jumps as a function of the interface viscosity μ S \mu_S μ S . Again, as the interface viscosity increases, the jump in the tangential component of the filtration velocity tends to zero, while the jump in the pressure converges to a non-zero value.
Figure 8. The effect of the interface viscosity μ S \mu_S μ S on (a) the jump in the tangential component of the filtration velocity ⟦ v f l t ⟧ \llbracket v_{\mathrm{fl}t}\rrbracket [ [ v fl t ] ] , and (b) the jump in the pressure ⟦ p ⟧ \llbracket p\rrbracket [ [ p ] ] , at time t = 0.1 s t = 0.1\,\mathrm{s} t = 0.1 s . Both jump variables are plotted as a function of the interface arc length for several increasing values of μ S \mu_S μ S .
Figure 9. The L 2 L^2 L 2 -norm of the jump in (a) the tangential component of the filtration velocity ⟦ v f l t t ⟧ \llbracket\mathbf{v}_{\mathrm{flt}t}\rrbracket [ [ v flt t ] ] , and (b) the pressure ⟦ p ⟧ \llbracket p\rrbracket [ [ p ] ] , as a function of the interface viscosity μ S \mu_S μ S , at time t = 0.1 t = 0.1 t = 0.1 s.
Summary, discussion, and conclusions
In this work we have proposed an approach to a variational formulation of the conditions at the interface separating two contiguous poroelastic domains. The formulation stems from mixture theory. We focus our attention to mixture of constituents that are incompressible in their pure form. This said, the present work is applicable to problems with large deformations. Our approach follows very closely that by dell’Isola [18 ] and adapts it to the constitutive class considered herein. We also offer a connection to a treatment of the subject according to a traditional continuum mechanics approach.[8 ]
From a modeling perspective, our first goal was to make a more direct link between the constitutive assumptions pertaining to the behavior of the interface and the corresponding interface conditions. This is in contrast to interface conditions that rely on kinematic constraints such as the full continuity of the filtration velocity and/or of the pore pressure. The second goal of our work was to produce a variational formulation of the problem easily implementable as a FEM able to respect, at least formally, the energetics of the problem. In such an attempt, we chose test functions for the fluid velocity to have arbitrary jumps at the interface, while the constraint originating with the jump condition of the balance of mass is imposed by way of a Lagrange multiplier.
The results from the examples we have included are encouraging in that we have empirically shown convergence for our formulation using the method of manufactured solutions. In doing so, we have identified what seems to be a practical choice of stable finite element spaces.
We have also solved a classic problem of flow through a rigid porous medium showing that we can recover some well-established results by controlling the constitutive response of the interface. In our case, the formulation contains a dissipative term associated to the norm of the jump of the filtration velocity. This term can be interpreted as a penalization of the jump of the filtration velocity. As the value of the parameter μ S \mu_S μ S is increased, our solution appears to approach a solution with continuous filtration velocity. The last set of results confirms this behavior, this time in a full poroelastic setting. However, it also shows that the jump of the pore pressure field across the interface is not directly controlled by the parameter that controls the jump of the filtration velocity. This conclusion was to be expected as these fields are indeed independent.
In conclusion, we believe that the proposed formulation has some encouraging results that warrant further development and analysis. As mentioned in the introduction, our interest lies in brain physiology and the modeling of membranes in the central nervous system and we plan future applications in this arena. These developments will require more sophisticated models to regulate the pore pressure jump across the interfaces under consideration. Another line of inquiry concerns the adaptation of the proposed FEM formulation to more general mixtures, say with compressible constituents, and with the kinematics considered in Ref. 18. From a numerical perspective, a formal investigation on stability issues associated to the compatibility of spaces of the Lagrange multipliers in relation to the other fields is necessary. An important area of study is also stability relative to constitutive parameters as discussed, for example, by Rodrigo et al.[44 ] .
Notational conventions
Preliminaries
R \mathbb{R} R will denote the real number set, with R + \mathbb{R}^{+} R + and R 0 + \mathbb{R}^{+}_{0} R 0 + denoting its positive and non-negative subsets, respectively. E 3 \mathcal{E}^{3} E 3 will denote the three-dimensional Euclidean point space and t ∈ R 0 + t \in\mathbb{R}^{+}_{0} t ∈ R 0 + time. The translation (vector) space of E 3 \mathcal{E}^{3} E 3 will be denoted by V \mathcal{V} V . Given vector spaces U \mathcal{U} U and W \mathcal{W} W , the vector space of linear operators with domain U \mathcal{U} U and range in W \mathcal{W} W will be denoted by L ( U , W ) \mathcal{L}(\mathcal{U},\mathcal{W}) L ( U , W ) . The expression L ( V ) \mathcal{L}(\mathcal{V}) L ( V ) is shorthand for L ( V , V ) \mathcal{L}(\mathcal{V},\mathcal{V}) L ( V , V ) .
Physical quantities are described by means of functions whose domain is a specific manifold. Often, the physical meaning of a quantity remains unchanged when the function representing it is composed with a map between a manifold and another. This fact is at the root of the distinction between Eulerian, Lagrangian, and Eulerian–Lagrangian frameworks (cf. Refs. 24 and 20) in Continuum Mechanics. To indicate the domain of a function representing one of these physical quantities, we adopt the shorthand notation in Ref. 18. For any t t t , let Ω a ( t ) ⊂ E 3 \Omega_a(t) \subset\mathcal{E}^{3} Ω a ( t ) ⊂ E 3 and Ω b ( t ) ⊂ E 3 \Omega_b(t) \subset\mathcal{E}^{3} Ω b ( t ) ⊂ E 3 be two possibly time-dependent smooth manifolds whose points are denoted by X a \mathbf{X}_a X a and X b \mathbf{X}_b X b , respectively. Let χ t : Ω a ( t ) → Ω b ( t ) \boldsymbol{\chi}_t:\Omega_a(t)\to\Omega_b(t) χ t : Ω a ( t ) → Ω b ( t ) such that χ t ( X a ) = X b ∈ Ω b ( t ) \boldsymbol{\chi}_t(\mathbf{X}_a)=\mathbf{X}_b\in\Omega_b(t) χ t ( X a ) = X b ∈ Ω b ( t ) be a homeomorphism. Let ϕ : Ω a ( t ) → U \phi:\Omega_a(t)\to\mathcal{U} ϕ : Ω a ( t ) → U and ψ : Ω b ( t ) → W \psi:\Omega_b(t)\to\mathcal{W} ψ : Ω b ( t ) → W be continuous functions taking values in the vector spaces U \mathcal{U} U and W \mathcal{W} W , respectively. Then, we denote by ϕ b \phi^{\mathrm{b}} ϕ b and ψ a \psi^{\mathrm{a}} ψ a the following functions
ϕ b : = ϕ ∘ χ t − 1 and ψ a : = ψ ∘ χ t . (81) \phi^{\mathrm{b}} := \phi\circ\boldsymbol{\chi}_t^{-1}
\quad\text{and}\quad
\psi^{\mathrm{a}} := \psi\circ\boldsymbol{\chi}_t. \tag*{(81)} ϕ b := ϕ ∘ χ t − 1 and ψ a := ψ ∘ χ t . (81)
As in Ref. 24, we view second-order tensors as linear operators from a space to another. Given a second-order tensor A \mathrm{A} A , its transpose will be denoted by A T \mathrm{A}^{\mathsf{T}} A T . If A \mathrm{A} A is invertible, its inverse and its inverse transpose will be denoted by A − 1 \mathrm{A}^{-1} A − 1 and A − T \mathrm{A}^{-\mathsf{T}} A − T , respectively. Given a vector-valued quantity u \mathbf{u} u in the domain of A \mathrm{A} A , the action of A \mathrm{A} A on u \mathbf{u} u will be denoted by A u \mathrm{A}\mathbf{u} A u .
Given a smooth manifold A \mathcal{A} A and a point P ∈ A P \in\mathcal{A} P ∈ A , T P A T_{P}\mathcal{A} T P A denotes the tangent space at P P P . Clearly, in an Euclidean context, if A \mathcal{A} A is a three-dimensional submanifold of E 3 \mathcal{E}^{3} E 3 , its tangent spaces are isomorphic to V \mathcal{V} V . If a metric is defined over A \mathcal{A} A , the notation ∇ ψ \nabla\psi ∇ ψ will denote the gradient of the field ψ \psi ψ . We note that a gradient is not a stand-alone object in that it is inherently linked to the metric manifold over which it is defined — a gradient on a manifold is different from that on another manifold. The notation ∇ ⋅ □ \nabla\cdot\Box ∇ ⋅ □ will denote the divergence of the quantity □ \Box □ , assuming that □ \Box □ admits a divergence. As is the convention in elementary calculus, we will use the notation ∂ t □ \partial_t\Box ∂ t □ to denote the derivative of □ \Box □ with respect to t t t while holding all other arguments of □ \Box □ fixed.
Let a smooth surface Σ \Sigma Σ be a codimension-1 submanifold of a three-dimensional A ⊂ E 3 \mathcal{A} \subset\mathcal{E}^{3} A ⊂ E 3 . Let ν \boldsymbol{\nu} ν be a unit vector field orienting Σ \Sigma Σ . Let ϕ ( X ) \phi(\mathbf{X}) ϕ ( X ) , X ∈ A \mathbf{X} \in\mathcal{A} X ∈ A , be some field that is continuous away from Σ \Sigma Σ . Then, at X ^ ∈ Σ \widehat{\mathbf{X}} \in\Sigma X ∈ Σ ,
ϕ ± : = lim s → 0 + ϕ ( X ^ ± s ν ) , ⟦ ϕ ⟧ : = ϕ + − ϕ − , and ⟨ ϕ ⟩ : = 1 2 ( ϕ + + ϕ − ) . (82) \phi^{\pm} := \lim_{s\to0^{+}}\phi(\widehat{\mathbf{X}}\pm s\boldsymbol{\nu}),\quad\llbracket\phi\rrbracket:= \phi^{+}-\phi^{-},\quad\text{and}\quad\langle\phi\rangle:= \tfrac{1}{2}(\phi^{+}+\phi^{-}). \tag*{(82)} ϕ ± := s → 0 + lim ϕ ( X ± s ν ) , [ [ ϕ ] ] := ϕ + − ϕ − , and ⟨ ϕ ⟩ := 2 1 ( ϕ + + ϕ − ) . (82)
Appendix B
Here we provide explicit derivations for some relationships that were referenced in the main text.
Proposition B.1. Under the assumptions stated in the main paper, we have
v a ⋅ ρ a a a = ∂ t ( 1 2 ρ a ∥ v a ∥ 2 ) + ∇ ⋅ ( 1 2 ρ a ∥ v a ∥ 2 v a ) . (83) \mathbf{v}_{a}\cdot\rho_{a}\mathbf{a}_{a}=\partial_{t}\left(\tfrac{1}{2}\rho_{a}\lVert\mathbf{v}_{a}\rVert^{2}\right)+\nabla\cdot\left(\tfrac{1}{2}\rho_{a}\lVert\mathbf{v}_{a}\rVert^{2}\mathbf{v}_{a}\right). \tag*{(83)} v a ⋅ ρ a a a = ∂ t ( 2 1 ρ a ∥ v a ∥ 2 ) + ∇ ⋅ ( 2 1 ρ a ∥ v a ∥ 2 v a ) . (83)
Proof. We recall that a a \mathbf{a}_{a} a a is the material acceleration of species a a a . Hence, in Eulerian form we have
a a = ∂ t v a + L a v a = ∂ t v a + ∇ v a [ v a ] , \mathbf{a}_{a}=\partial_{t}\mathbf{v}_{a}+\mathcal{L}_{a}\mathbf{v}_{a}=\partial_{t}\mathbf{v}_{a}+\nabla\mathbf{v}_{a}[\mathbf{v}_{a}], a a = ∂ t v a + L a v a = ∂ t v a + ∇ v a [ v a ] ,
where we recall that L a = ∇ v a \mathcal{L}_{a}=\nabla\mathbf{v}_{a} L a = ∇ v a and we write ∇ □ [ ⋄ ] \nabla\Box[\diamond] ∇ □ [ ⋄ ] to denote the action of the gradient of a field □ \Box □ onto the field ⋄ \diamond ⋄ , assuming that the operation is meaningful for □ \Box □ and ⋄ \diamond ⋄ . We also recall the following two identities (cf. Ref. 24):
Given smooth vector fields w \mathbf{w} w and u \mathbf{u} u ,
∇ ( w ⋅ u ) = ( ∇ w ) T [ u ] + ( ∇ u ) T [ w ] . (84) \nabla(\mathbf{w}\cdot\mathbf{u})=(\nabla\mathbf{w})^{\mathsf{T}}[\mathbf{u}]+(\nabla\mathbf{u})^{\mathsf{T}}[\mathbf{w}]. \tag*{(84)} ∇ ( w ⋅ u ) = ( ∇ w ) T [ u ] + ( ∇ u ) T [ w ] . (84)
Given a smooth scalar field ϕ \phi ϕ and a smooth vector field w \mathbf{w} w ,
∇ ⋅ ( ϕ w ) = ∇ ϕ [ w ] + ϕ ∇ ⋅ w . (85) \nabla\cdot(\phi\mathbf{w})=\nabla\phi[\mathbf{w}]+\phi\nabla\cdot\mathbf{w}. \tag*{(85)} ∇ ⋅ ( ϕ w ) = ∇ ϕ [ w ] + ϕ ∇ ⋅ w . (85)
Finally, we recall that the balance of mass for species a a a in the absence of chemical reaction is
∂ t ρ a + ∇ ⋅ ( ρ a v a ) = 0 ⇒ ∂ t ρ a = − ∇ ⋅ ( ρ a v a ) . \partial_{t}\rho_{a}+\nabla\cdot(\rho_{a}\mathbf{v}_{a})=0\quad\Rightarrow\quad\partial_{t}\rho_{a}=-\nabla\cdot(\rho_{a}\mathbf{v}_{a}). ∂ t ρ a + ∇ ⋅ ( ρ a v a ) = 0 ⇒ ∂ t ρ a = − ∇ ⋅ ( ρ a v a ) .
With the above in mind, we have
v a ⋅ ρ a a a = ρ a ( v a ⋅ ∂ t v a + v a ⋅ L v a ) = ρ a ( v a ⋅ ∂ t v a + L a T v a ⋅ v a ) = 1 2 ρ a ( ∂ t ∥ v a ∥ 2 + grad ∥ v a ∥ 2 [ v a ] ) = ∂ t ( 1 2 ρ a ∥ v a ∥ 2 ) + 1 2 ρ a grad ∥ v a ∥ 2 [ v a ] − ( ∂ t ρ a ) 1 2 ∥ v a ∥ 2 = ∂ t ( 1 2 ρ a ∥ v a ∥ 2 ) + 1 2 ρ a grad ∥ v a ∥ 2 [ v a ] + ( ∇ ⋅ ρ a v a ) 1 2 ∥ v a ∥ 2 = ∂ t ( 1 2 ρ a ∥ v a ∥ 2 ) + ∇ ⋅ ( 1 2 ρ a ∥ v a ∥ 2 v a ) . \begin{aligned}
\mathbf{v}_a\cdot\rho_a\mathbf{a}_a
&=\rho_a(\mathbf{v}_a\cdot\partial_t\mathbf{v}_a+\mathbf{v}_a\cdot\mathbf{L}\mathbf{v}_a)\\
&=\rho_a(\mathbf{v}_a\cdot\partial_t\mathbf{v}_a+\mathbf{L}^{\mathsf{T}}_a\mathbf{v}_a\cdot\mathbf{v}_a)\\
&=\frac{1}{2}\rho_a\left(\partial_t\lVert\mathbf{v}_a\rVert^2+\operatorname{grad}\lVert\mathbf{v}_a\rVert^2[\mathbf{v}_a]\right)\\
&=\partial_t\left(\frac{1}{2}\rho_a\lVert\mathbf{v}_a\rVert^2\right)+\frac{1}{2}\rho_a\operatorname{grad}\lVert\mathbf{v}_a\rVert^2[\mathbf{v}_a]-(\partial_t\rho_a)\frac{1}{2}\lVert\mathbf{v}_a\rVert^2\\
&=\partial_t\left(\frac{1}{2}\rho_a\lVert\mathbf{v}_a\rVert^2\right)+\frac{1}{2}\rho_a\operatorname{grad}\lVert\mathbf{v}_a\rVert^2[\mathbf{v}_a]+(\nabla\cdot\rho_a\mathbf{v}_a)\frac{1}{2}\lVert\mathbf{v}_a\rVert^2\\
&=\partial_t\left(\frac{1}{2}\rho_a\lVert\mathbf{v}_a\rVert^2\right)+\nabla\cdot\left(\frac{1}{2}\rho_a\lVert\mathbf{v}_a\rVert^2\mathbf{v}_a\right).
\end{aligned} v a ⋅ ρ a a a = ρ a ( v a ⋅ ∂ t v a + v a ⋅ L v a ) = ρ a ( v a ⋅ ∂ t v a + L a T v a ⋅ v a ) = 2 1 ρ a ( ∂ t ∥ v a ∥ 2 + grad ∥ v a ∥ 2 [ v a ] ) = ∂ t ( 2 1 ρ a ∥ v a ∥ 2 ) + 2 1 ρ a grad ∥ v a ∥ 2 [ v a ] − ( ∂ t ρ a ) 2 1 ∥ v a ∥ 2 = ∂ t ( 2 1 ρ a ∥ v a ∥ 2 ) + 2 1 ρ a grad ∥ v a ∥ 2 [ v a ] + ( ∇ ⋅ ρ a v a ) 2 1 ∥ v a ∥ 2 = ∂ t ( 2 1 ρ a ∥ v a ∥ 2 ) + ∇ ⋅ ( 2 1 ρ a ∥ v a ∥ 2 v a ) .
Proposition B.2. Let Ω \Omega Ω be the control volume defined in § 3.4 . Recalling the definitions in § 3.3 , we have
∫ Ω T s e : L s = d d t ∫ Ω ψ s + ∫ ∂ Ω e x t ψ s v s ⊙ ⋅ n . (86) \int_{\Omega}\mathbf{T}^{\mathrm{e}}_s:\mathbf{L}_s
=\frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega}\psi_s+\int_{\partial\Omega_{\mathrm{ext}}}\psi_s\mathbf{v}^{\odot}_s\cdot\mathbf{n}. \tag*{(86)} ∫ Ω T s e : L s = d t d ∫ Ω ψ s + ∫ ∂ Ω ext ψ s v s ⊙ ⋅ n . (86) Proof. Let Ω s \Omega_s Ω s be the inverse image of Ω \Omega Ω under the solid’s motion χ s \boldsymbol{\chi}_s χ s . We recall that Ω \Omega Ω is partitioned into two domains by the moving surface S ( t ) S(t) S ( t ) and that ∂ Ω e x t \partial\Omega_{\mathrm{ext}} ∂ Ω ext , the external boundary of Ω \Omega Ω , is fixed. By contrast, in the solid’s reference configuration, the inverse image of S ( t ) S(t) S ( t ) , namely S s S_s S s , is fixed, whereas ∂ Ω s e x t \partial\Omega_{s\mathrm{ext}} ∂ Ω s ext , the inverse image of ∂ Ω e x t \partial\Omega_{\mathrm{ext}} ∂ Ω ext , is a moving surface with normal velocity − F s − 1 v s ⋅ n s -\mathbf{F}_s^{-1}\mathbf{v}_s\cdot\mathbf{n}_s − F s − 1 v s ⋅ n s , where n s \mathbf{n}_s n s is the unit normal orienting ∂ Ω s e x t \partial\Omega_{s\mathrm{ext}} ∂ Ω s ext , outward relative to Ω s \Omega_s Ω s . In fact, let X ^ ( t ) \widehat{\mathbf{X}}(t) X ( t ) be a point on ∂ Ω s e x t \partial\Omega_{s\mathrm{ext}} ∂ Ω s ext . This point is convected by the solid’s motion to the fixed point χ s ( X ^ ( t ) , t ) \boldsymbol{\chi}_s(\widehat{\mathbf{X}}(t),t) χ s ( X ( t ) , t ) on ∂ Ω e x t \partial\Omega_{\mathrm{ext}} ∂ Ω ext so that
d d t χ s ( X ^ ( t ) , t ) = 0 ⇒ ∂ t X ^ = − F s − 1 v s . \frac{\mathrm{d}}{\mathrm{d}t}\boldsymbol{\chi}_s(\widehat{\mathbf{X}}(t),t)=\mathbf{0}
\quad\Rightarrow\quad
\partial_t\widehat{\mathbf{X}}=-\mathbf{F}_s^{-1}\mathbf{v}_s. d t d χ s ( X ( t ) , t ) = 0 ⇒ ∂ t X = − F s − 1 v s . With the above in mind, we have
d d t ∫ Ω ψ s = d d t ∫ Ω s J s ψ s ⊙ = ∫ Ω s ∂ t Ψ s − ∫ ∂ Ω s e x t J s ψ s ⊙ F s − 1 v s ⋅ n s = ∫ Ω s ∂ t Ψ s − ∫ ∂ Ω s e x t ψ s ⊙ v s ⋅ ( J s F s − T n s ) . (87) \begin{aligned}
\frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega}\psi_s
&=\frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega_s}J_s\psi_s^{\odot}\\
&=\int_{\Omega_s}\partial_t\Psi_s-\int_{\partial\Omega_{s\mathrm{ext}}}J_s\psi_s^{\odot}\mathbf{F}_s^{-1}\mathbf{v}_s\cdot\mathbf{n}_s\\
&=\int_{\Omega_s}\partial_t\Psi_s-\int_{\partial\Omega_{s\mathrm{ext}}}\psi_s^{\odot}\mathbf{v}_s\cdot(J_s\mathbf{F}_s^{-\mathsf{T}}\mathbf{n}_s). \tag*{(87)}
\end{aligned} d t d ∫ Ω ψ s = d t d ∫ Ω s J s ψ s ⊙ = ∫ Ω s ∂ t Ψ s − ∫ ∂ Ω s ext J s ψ s ⊙ F s − 1 v s ⋅ n s = ∫ Ω s ∂ t Ψ s − ∫ ∂ Ω s ext ψ s ⊙ v s ⋅ ( J s F s − T n s ) . (87) Next, making use of the second of Eqs. (32) , we have
∂ t Ψ s = ∂ Ψ s ∂ C s : ∂ t C s = 1 2 J s F s − 1 ( T s e ) ⊙ F s − T : ∂ t C s . \partial_t\Psi_s=\frac{\partial\Psi_s}{\partial\mathbf{C}_s}:\partial_t\mathbf{C}_s
=\frac{1}{2}J_s\mathbf{F}_s^{-1}(\mathbf{T}^{\mathrm{e}}_s)^{\odot}\mathbf{F}_s^{-\mathsf{T}}:\partial_t\mathbf{C}_s. ∂ t Ψ s = ∂ C s ∂ Ψ s : ∂ t C s = 2 1 J s F s − 1 ( T s e ) ⊙ F s − T : ∂ t C s . Next, we recall that C s = F s T F s \mathbf{C}_s=\mathbf{F}_s^{\mathsf{T}}\mathbf{F}_s C s = F s T F s , so that ∂ t C s = 2 sym ( F s T ∂ t F s ) \partial_t\mathbf{C}_s=2\operatorname{sym}(\mathbf{F}_s^{\mathsf{T}}\partial_t\mathbf{F}_s) ∂ t C s = 2 sym ( F s T ∂ t F s ) . Hence, taking advantage of the symmetry of T s e \mathbf{T}^{\mathrm{e}}_s T s e , we can then write
∂ t Ψ s = J s F s − 1 ( T s e ) ⊙ F s − T : F s T ∂ t F s = J s ( T s e ) ⊙ : ( ∂ t F s ) F s − 1 = J s ( T s e ) ⊙ : L s ⊙ . (88) \begin{aligned}
\partial_t\Psi_s
&=J_s\mathbf{F}_s^{-1}(\mathbf{T}^{\mathrm{e}}_s)^{\odot}\mathbf{F}_s^{-\mathsf{T}}:\mathbf{F}_s^{\mathsf{T}}\partial_t\mathbf{F}_s\\
&=J_s(\mathbf{T}^{\mathrm{e}}_s)^{\odot}:(\partial_t\mathbf{F}_s)\mathbf{F}_s^{-1}\\
&=J_s(\mathbf{T}^{\mathrm{e}}_s)^{\odot}:\mathbf{L}^{\odot}_s. \tag*{(88)}
\end{aligned} ∂ t Ψ s = J s F s − 1 ( T s e ) ⊙ F s − T : F s T ∂ t F s = J s ( T s e ) ⊙ : ( ∂ t F s ) F s − 1 = J s ( T s e ) ⊙ : L s ⊙ . (88)
Substituting Eq. (B.6) into Eq. (B.5), pushing the integration to the domain Ω \Omega Ω , and rearranging terms, Eq. (B.4) follows.
Proposition B.3. Equation (43) holds true.
Proof. Summing Eqs. (38) for a = f a=f a = f and a = s a=s a = s , we have
d d t ∫ Ω ( k f + k s ) + ∫ Ω ( T f : L f + T s : L s − p f ⋅ ( v f ∘ − v s ∘ ) ) = − ∫ ∂ Ω e x t ( k f v f ∘ + k s v s ∘ ) ⋅ n + ∫ ∂ Ω e x t ( T f n ⋅ v f ∘ + T s n ⋅ v s ∘ ) + ∫ S ( t ) [ k f ( v f ∘ − v s ∘ ) ⋅ m − T f m ⋅ v f ∘ − T s m ⋅ v s ∘ ] + ∫ Ω ( b f ⋅ v f ∘ + b s ⋅ v s ∘ ) . (89) \begin{aligned}
\frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega}(k_{\mathrm{f}}+k_{\mathrm{s}})&+\int_{\Omega}\left(\mathbf{T}_{\mathrm{f}}:\mathbf{L}_{\mathrm{f}}+\mathbf{T}_{\mathrm{s}}:\mathbf{L}_{\mathrm{s}}-\mathbf{p}_{\mathrm{f}}\cdot\left(\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{v}_{\mathrm{s}}^{\circ}\right)\right)\\
&=-\int_{\partial\Omega_{\mathrm{ext}}}\left(k_{\mathrm{f}}\mathbf{v}_{\mathrm{f}}^{\circ}+k_{\mathrm{s}}\mathbf{v}_{\mathrm{s}}^{\circ}\right)\cdot\mathbf{n}
+\int_{\partial\Omega_{\mathrm{ext}}}\left(\mathbf{T}_{\mathrm{f}}\mathbf{n}\cdot\mathbf{v}_{\mathrm{f}}^{\circ}+\mathbf{T}_{\mathrm{s}}\mathbf{n}\cdot\mathbf{v}_{\mathrm{s}}^{\circ}\right)\\
&\quad+\int_{S(t)}\left[k_{\mathrm{f}}\left(\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{v}_{\mathrm{s}}^{\circ}\right)\cdot\mathbf{m}-\mathbf{T}_{\mathrm{f}}\mathbf{m}\cdot\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{T}_{\mathrm{s}}\mathbf{m}\cdot\mathbf{v}_{\mathrm{s}}^{\circ}\right]\\
&\quad+\int_{\Omega}\left(\mathbf{b}_{\mathrm{f}}\cdot\mathbf{v}_{\mathrm{f}}^{\circ}+\mathbf{b}_{\mathrm{s}}\cdot\mathbf{v}_{\mathrm{s}}^{\circ}\right). \tag*{(89)}
\end{aligned} d t d ∫ Ω ( k f + k s ) + ∫ Ω ( T f : L f + T s : L s − p f ⋅ ( v f ∘ − v s ∘ ) ) = − ∫ ∂ Ω ext ( k f v f ∘ + k s v s ∘ ) ⋅ n + ∫ ∂ Ω ext ( T f n ⋅ v f ∘ + T s n ⋅ v s ∘ ) + ∫ S ( t ) [ k f ( v f ∘ − v s ∘ ) ⋅ m − T f m ⋅ v f ∘ − T s m ⋅ v s ∘ ] + ∫ Ω ( b f ⋅ v f ∘ + b s ⋅ v s ∘ ) . (89)
where we have used the property that p s = − p f \mathbf{p}_{\mathrm{s}}=-\mathbf{p}_{\mathrm{f}} p s = − p f . Using Eqs. (37) , for the last term on the left-hand side of Eq. (B.7), we have
− p f ⋅ ( v f ∘ − v s ∘ ) = − p ∇ ϕ f ⋅ ( v f ∘ − v s ∘ ) − f f ⋅ ( v f ∘ − v s ∘ ) = − p ∇ ⋅ [ ϕ f ( v f ∘ − v s ∘ ) ] + p ϕ f ∇ ⋅ ( v f ∘ − v s ∘ ) − f f ⋅ ( v f ∘ − v s ∘ ) . (90) \begin{aligned}
-\mathbf{p}_{\mathrm{f}}\cdot\left(\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{v}_{\mathrm{s}}^{\circ}\right)
&=-p\nabla\phi_{\mathrm{f}}\cdot\left(\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{v}_{\mathrm{s}}^{\circ}\right)-\mathbf{f}_{\mathrm{f}}\cdot\left(\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{v}_{\mathrm{s}}^{\circ}\right)\\
&=-p\nabla\cdot\left[\phi_{\mathrm{f}}\left(\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{v}_{\mathrm{s}}^{\circ}\right)\right]+p\phi_{\mathrm{f}}\nabla\cdot\left(\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{v}_{\mathrm{s}}^{\circ}\right)\\
&\quad-\mathbf{f}_{\mathrm{f}}\cdot\left(\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{v}_{\mathrm{s}}^{\circ}\right). \tag*{(90)}
\end{aligned} − p f ⋅ ( v f ∘ − v s ∘ ) = − p ∇ ϕ f ⋅ ( v f ∘ − v s ∘ ) − f f ⋅ ( v f ∘ − v s ∘ ) = − p ∇ ⋅ [ ϕ f ( v f ∘ − v s ∘ ) ] + p ϕ f ∇ ⋅ ( v f ∘ − v s ∘ ) − f f ⋅ ( v f ∘ − v s ∘ ) . (90)
In view of Eqs. (32) –(36) , and recalling that the Cauchy stresses in these equations are symmetric, we also have
T f : L f + T s : L s = − p ϕ f I : L f ⏟ ∇ ⋅ v f ∘ + 2 μ f D f : D f + 2 μ B ( D f − D s ) : D f − p ϕ s I : L s ⏟ ∇ ⋅ v s ∘ + T s e : L s − 2 μ B ( D f − D s ) : D s = − p ϕ f ∇ ⋅ ( v f ∘ − v s ∘ ) − p ∇ ⋅ v s ∘ + 2 μ f ∥ D f ∥ 2 + 2 μ B ∥ D f − D s ∥ 2 + T s e : L s . (91) \begin{aligned}
\mathbf{T}_{\mathrm{f}}:\mathbf{L}_{\mathrm{f}}+\mathbf{T}_{\mathrm{s}}:\mathbf{L}_{\mathrm{s}}
&=-p\phi_{\mathrm{f}}\underbrace{\mathbf{I}:\mathbf{L}_{\mathrm{f}}}_{\nabla\cdot\mathbf{v}_{\mathrm{f}}^{\circ}}+2\mu_{\mathrm{f}}\mathbf{D}_{\mathrm{f}}:\mathbf{D}_{\mathrm{f}}+2\mu_{\mathrm{B}}\left(\mathbf{D}_{\mathrm{f}}-\mathbf{D}_{\mathrm{s}}\right):\mathbf{D}_{\mathrm{f}}\\
&\quad-p\phi_{\mathrm{s}}\underbrace{\mathbf{I}:\mathbf{L}_{\mathrm{s}}}_{\nabla\cdot\mathbf{v}_{\mathrm{s}}^{\circ}}+\mathbf{T}_{\mathrm{s}}^{\mathrm{e}}:\mathbf{L}_{\mathrm{s}}-2\mu_{\mathrm{B}}\left(\mathbf{D}_{\mathrm{f}}-\mathbf{D}_{\mathrm{s}}\right):\mathbf{D}_{\mathrm{s}}\\
&=-p\phi_{\mathrm{f}}\nabla\cdot\left(\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{v}_{\mathrm{s}}^{\circ}\right)-p\nabla\cdot\mathbf{v}_{\mathrm{s}}^{\circ}
+2\mu_{\mathrm{f}}\lVert\mathbf{D}_{\mathrm{f}}\rVert^{2}+2\mu_{\mathrm{B}}\lVert\mathbf{D}_{\mathrm{f}}-\mathbf{D}_{\mathrm{s}}\rVert^{2}\\
&\quad+\mathbf{T}_{\mathrm{s}}^{\mathrm{e}}:\mathbf{L}_{\mathrm{s}}. \tag*{(91)}
\end{aligned} T f : L f + T s : L s = − p ϕ f ∇ ⋅ v f ∘ I : L f + 2 μ f D f : D f + 2 μ B ( D f − D s ) : D f − p ϕ s ∇ ⋅ v s ∘ I : L s + T s e : L s − 2 μ B ( D f − D s ) : D s = − p ϕ f ∇ ⋅ ( v f ∘ − v s ∘ ) − p ∇ ⋅ v s ∘ + 2 μ f ∥ D f ∥ 2 + 2 μ B ∥ D f − D s ∥ 2 + T s e : L s . (91)
Summing Eqs. (B.8) and (B.9), and using Eqs. (13) along with the second of Eqs. (18) , we then have that
T f : L f + T s : L s − p f ⋅ ( v f ∘ − v s ∘ ) = T s e : L s + D , (92) \mathbf{T}_{\mathrm{f}}:\mathbf{L}_{\mathrm{f}}+\mathbf{T}_{\mathrm{s}}:\mathbf{L}_{\mathrm{s}}-\mathbf{p}_{\mathrm{f}}\cdot\left(\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{v}_{\mathrm{s}}^{\circ}\right)=\mathbf{T}_{\mathrm{s}}^{\mathrm{e}}:\mathbf{L}_{\mathrm{s}}+\mathcal{D}, \tag*{(92)} T f : L f + T s : L s − p f ⋅ ( v f ∘ − v s ∘ ) = T s e : L s + D , (92)
where we have used Eqs. (37) and the definition of D \mathcal{D} D in Eq. (41) . Integrating Eq. (B.10) over Ω \Omega Ω , and using Eq. (B.4), we can now write
∫ Ω ( T f : L f + T s : L s − p f ⋅ ( v f ∘ − v s ∘ ) ) = d d t ∫ Ω ψ s + ∫ ∂ Ω e x t ψ s v s ∘ ⋅ n + ∫ Ω D . (93) \int_{\Omega}\left(\mathbf{T}_{\mathrm{f}}:\mathbf{L}_{\mathrm{f}}+\mathbf{T}_{\mathrm{s}}:\mathbf{L}_{\mathrm{s}}-\mathbf{p}_{\mathrm{f}}\cdot\left(\mathbf{v}_{\mathrm{f}}^{\circ}-\mathbf{v}_{\mathrm{s}}^{\circ}\right)\right)
=\frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega}\psi_{\mathrm{s}}+\int_{\partial\Omega_{\mathrm{ext}}}\psi_{\mathrm{s}}\mathbf{v}_{\mathrm{s}}^{\circ}\cdot\mathbf{n}+\int_{\Omega}\mathcal{D}. \tag*{(93)} ∫ Ω ( T f : L f + T s : L s − p f ⋅ ( v f ∘ − v s ∘ ) ) = d t d ∫ Ω ψ s + ∫ ∂ Ω ext ψ s v s ∘ ⋅ n + ∫ Ω D . (93)
Using the definition in Eq. (40) , Eq. (B.7) can now be rewritten as
W e x t ( Ω ) = d d t ∫ Ω ( k s + k f + ψ s ) + ∫ Ω D − ∫ S ( t ) [ ⟦ k f ( v f e − v s e ) ⋅ m − T f m ⋅ v f e − T s m ⋅ v s e ⟧ ] . (94) \mathcal{W}_{\mathrm{ext}}(\Omega)=\frac{\mathrm{d}}{\mathrm{d}t}\int_{\Omega}(k_s+k_f+\psi_s)+\int_{\Omega}\mathcal{D}
-\int_{S(t)}\left[\llbracket k_f(\mathbf{v}_f^{\mathrm{e}}-\mathbf{v}_s^{\mathrm{e}})\cdot\mathbf{m}-\mathbf{T}_f\mathbf{m}\cdot\mathbf{v}_f^{\mathrm{e}}-\mathbf{T}_s\mathbf{m}\cdot\mathbf{v}_s^{\mathrm{e}}\rrbracket\right]. \tag*{(94)} W ext ( Ω ) = d t d ∫ Ω ( k s + k f + ψ s ) + ∫ Ω D − ∫ S ( t ) [ [ [ k f ( v f e − v s e ) ⋅ m − T f m ⋅ v f e − T s m ⋅ v s e ] ] ] . (94)
We observe that, using the definition in Eq. (27) , and recalling that both d d d and v s \mathbf{v}_s v s are continuous across S ( t ) \mathcal{S}(t) S ( t ) ,
⟦ k f ( v f e − v s e ) ⋅ m ⟧ = ⟦ 1 2 d ∥ v f e ∥ 2 ⟧ = d ⟦ 1 2 ∥ v f e ∥ 2 ⟧ = d ⟦ 1 2 ∥ v f e − v s e ∥ 2 + v f e ⋅ v s e − 1 2 ∥ v s e ∥ 2 ⟧ = d ⟦ 1 2 ∥ v f e − v s e ∥ 2 + v f e ⋅ v s e ⟧ . (95) \begin{aligned}
\llbracket k_f(\mathbf{v}_f^{\mathrm{e}}-\mathbf{v}_s^{\mathrm{e}})\cdot\mathbf{m}\rrbracket&= \llbracket\tfrac{1}{2}d\lVert\mathbf{v}_f^{\mathrm{e}}\rVert^2\rrbracket\\
&=d\llbracket\tfrac{1}{2}\lVert\mathbf{v}_f^{\mathrm{e}}\rVert^2\rrbracket\\
&=d\llbracket\tfrac{1}{2}\lVert\mathbf{v}_f^{\mathrm{e}}-\mathbf{v}_s^{\mathrm{e}}\rVert^2+\mathbf{v}_f^{\mathrm{e}}\cdot\mathbf{v}_s^{\mathrm{e}}-\tfrac{1}{2}\lVert\mathbf{v}_s^{\mathrm{e}}\rVert^2\rrbracket\\
&=d\llbracket\tfrac{1}{2}\lVert\mathbf{v}_f^{\mathrm{e}}-\mathbf{v}_s^{\mathrm{e}}\rVert^2+\mathbf{v}_f^{\mathrm{e}}\cdot\mathbf{v}_s^{\mathrm{e}}\rrbracket. \tag*{(95)}
\end{aligned} [ [ k f ( v f e − v s e ) ⋅ m ] ] = [ [ 2 1 d ∥ v f e ∥ 2 ] ] = d [ [ 2 1 ∥ v f e ∥ 2 ] ] = d [ [ 2 1 ∥ v f e − v s e ∥ 2 + v f e ⋅ v s e − 2 1 ∥ v s e ∥ 2 ] ] = d [ [ 2 1 ∥ v f e − v s e ∥ 2 + v f e ⋅ v s e ] ] . (95)
Next, again recalling that v s e \mathbf{v}_s^{\mathrm{e}} v s e is continuous across S ( t ) \mathcal{S}(t) S ( t ) , using Eq. (26) we have that
⟦ T s m ⋅ v s e ⟧ = d ⟦ v f e ⋅ v s e ⟧ − ⟦ T f m ⋅ v s e ⟧ . (96) \llbracket\mathbf{T}_s\mathbf{m}\cdot\mathbf{v}_s^{\mathrm{e}}\rrbracket=d\llbracket\mathbf{v}_f^{\mathrm{e}}\cdot\mathbf{v}_s^{\mathrm{e}}\rrbracket-\llbracket\mathbf{T}_f\mathbf{m}\cdot\mathbf{v}_s^{\mathrm{e}}\rrbracket. \tag*{(96)} [ [ T s m ⋅ v s e ] ] = d [ [ v f e ⋅ v s e ] ] − [ [ T f m ⋅ v s e ] ] . (96)
Substituting Eqs. (B.13) and (B.14) into Eq. (B.12), and recalling the definition in Eq. (42) , the claim follows.
Proposition B.4.
∂ ( v f s − v s ) ∂ q ˙ [ δ q ] = − F s F s f − 1 δ χ s f . (97) \frac{\partial(\mathbf{v}_f^{s}-\mathbf{v}_s)}{\partial\dot{\mathbf{q}}}[\delta\mathbf{q}]=-\mathbf{F}_s\mathbf{F}_{sf}^{-1}\delta\boldsymbol{\chi}_{sf}. \tag*{(97)} ∂ q ˙ ∂ ( v f s − v s ) [ δ q ] = − F s F s f − 1 δ χ s f . (97)
Proof. Equation (B.15) follows directly from Eq. (8) and the definition of the list q ( t ) \mathbf{q}(t) q ( t ) in Eq. (46) .
Proposition B.5. Under the assumptions stated in the paper, Eq. (55) holds.
Proof. The derivation presented herein makes repeated use of the standard identities such as those in (B.2) and (B.3).
The expression for the variation of the kinetic energy in this paper can be borrowed directly from Ref. 18. Adjusting for the differences in notation, this gives:
\begin{aligned}
\int_{0}^{T}\delta\mathcal{K}
={}&\int_{0}^{T}\left\{
\int_{B_s}-J_s\left(\rho_s^{\odot}\mathbf{a}_s^{\odot}+\rho_f^{\odot}\mathbf{a}_f^{\odot}\right)\cdot\delta\boldsymbol{\chi}_s\\
&-\int_{S_s}\left[\rho_f^{\odot}\mathbf{v}_f^{\odot}\otimes\left(\mathbf{v}_f^{\odot}-\mathbf{v}_s^{\odot}\right)\right]J_s\mathbf{F}_s^{-\mathrm{T}}\mathbf{m}_s\cdot\delta\boldsymbol{\chi}_s\\
&+\int_{B_s}J_s\rho_f^{\odot}\mathbf{a}_f^{\odot}\cdot\mathbf{F}_s\mathbf{F}_{sf}^{-1}\delta\boldsymbol{\chi}_{sf}\\
&-\int_{S_s}\left[\underbrace{\left(\frac{1}{2}\rho_f^{\odot}\lVert\mathbf{v}_f^{\odot}\rVert^2\right)}_{k_f^{\odot}}-\rho_f^{\odot}\mathbf{v}_f^{\odot}\cdot\left(\mathbf{v}_f^{\odot}-\mathbf{v}_s^{\odot}\right)\right]J_s\mathbf{F}_s^{-\mathrm{T}}\mathbf{m}_s\cdot\mathbf{F}_s\mathbf{F}_{sf}^{-1}\delta\boldsymbol{\chi}_{sf}
\right\}. \tag*{(98)}
\end{aligned}
where we have used the continuity of v s \mathbf{v}_s v s across S s \mathcal{S}_s S s to rewrite the term under bracket on the second line of the right-hand side.
To provide the form of the potential energy variation, we start with recalling that C s = F s T F s \mathbf{C}_s=\mathbf{F}_s^{\mathrm{T}}\mathbf{F}_s C s = F s T F s , with F s = ∇ χ s \mathbf{F}_s=\nabla\boldsymbol{\chi}_s F s = ∇ χ s . Hence, we have
δ Ψ s = ∂ Ψ s ∂ C s : δ C s ⇒ δ Ψ s = 2 F s ∂ Ψ s ∂ C s : ∇ δ χ s ⇒ δ Ψ s = P s : ∇ δ χ s , (99) \delta\Psi_s=\frac{\partial\Psi_s}{\partial\mathbf{C}_s}:\delta\mathbf{C}_s\Rightarrow\delta\Psi_s=2\mathbf{F}_s\frac{\partial\Psi_s}{\partial\mathbf{C}_s}:\nabla\delta\boldsymbol{\chi}_s\Rightarrow\delta\Psi_s=\mathbf{P}_s:\nabla\delta\boldsymbol{\chi}_s, \tag*{(99)} δ Ψ s = ∂ C s ∂ Ψ s : δ C s ⇒ δ Ψ s = 2 F s ∂ C s ∂ Ψ s : ∇ δ χ s ⇒ δ Ψ s = P s : ∇ δ χ s , (99)
where we have taken advantage of the symmetry of C s \mathbf{C}_s C s .
Recalling to Eqs. (9) , (10) , and (49) , and recalling that the terms J s b s J_s\mathbf{b}_s J s b s and J s b f J_s\mathbf{b}_f J s b f are treated as externally controlled entities at fixed X s \mathbf{X}_s X s and X f \mathbf{X}_f X f , respectively, the variation of the potential energy is as follows:
δ U = ∫ B s ( P s e : ∇ δ χ s − J s ( b s + b f ) ⊙ ⋅ δ χ s + J s b f ⊙ ⋅ F s F s f − 1 δ χ s f ) = ∫ B s ( ∇ ⋅ [ ( P s e ) T δ χ s ] − ( ∇ ⋅ P s e + J s ( b s + b f ) ⊙ ) ⋅ δ χ s + J s b f ⊙ ⋅ F s F s f − 1 δ χ s f ) = − ∫ S s [ T s e J s F s − T m s ⋅ δ χ s ] − ∫ B s ( J s ( ∇ ⋅ T s e + b s + b f ) ⊙ ⋅ δ χ s − J s b f ⊙ ⋅ F s F s f − 1 δ χ s f ) . (100) \begin{aligned}
\delta U
={}&\int_{B_s}\left(\mathbf{P}_s^{e}:\nabla\delta\boldsymbol{\chi}_s-J_s\left(\mathbf{b}_s+\mathbf{b}_f\right)^{\odot}\cdot\delta\boldsymbol{\chi}_s+J_s\mathbf{b}_f^{\odot}\cdot\mathbf{F}_s\mathbf{F}_{sf}^{-1}\delta\boldsymbol{\chi}_{sf}\right)\\
={}&\int_{B_s}\left(\nabla\cdot\left[(\mathbf{P}_s^{e})^{\mathrm{T}}\delta\boldsymbol{\chi}_s\right]-\left(\nabla\cdot\mathbf{P}_s^{e}+J_s\left(\mathbf{b}_s+\mathbf{b}_f\right)^{\odot}\right)\cdot\delta\boldsymbol{\chi}_s+J_s\mathbf{b}_f^{\odot}\cdot\mathbf{F}_s\mathbf{F}_{sf}^{-1}\delta\boldsymbol{\chi}_{sf}\right)\\
={}&-\int_{\mathcal{S}_s}\left[\mathbf{T}_s^{e}J_s\mathbf{F}_s^{-\mathrm{T}}\mathbf{m}_s\cdot\delta\boldsymbol{\chi}_s\right]\\
&-\int_{B_s}\left(J_s\left(\nabla\cdot\mathbf{T}_s^{e}+\mathbf{b}_s+\mathbf{b}_f\right)^{\odot}\cdot\delta\boldsymbol{\chi}_s-J_s\mathbf{b}_f^{\odot}\cdot\mathbf{F}_s\mathbf{F}_{sf}^{-1}\delta\boldsymbol{\chi}_{sf}\right). \tag*{(100)}
\end{aligned} δ U = = = ∫ B s ( P s e : ∇ δ χ s − J s ( b s + b f ) ⊙ ⋅ δ χ s + J s b f ⊙ ⋅ F s F s f − 1 δ χ s f ) ∫ B s ( ∇ ⋅ [ ( P s e ) T δ χ s ] − ( ∇ ⋅ P s e + J s ( b s + b f ) ⊙ ) ⋅ δ χ s + J s b f ⊙ ⋅ F s F s f − 1 δ χ s f ) − ∫ S s [ T s e J s F s − T m s ⋅ δ χ s ] − ∫ B s ( J s ( ∇ ⋅ T s e + b s + b f ) ⊙ ⋅ δ χ s − J s b f ⊙ ⋅ F s F s f − 1 δ χ s f ) . (100)
As C \mathcal{C} C has been introduced in rate form, its variation is computed as δ C = \delta\mathcal{C}= δ C = ( ∂ C / ∂ q ˙ ) [ δ q ] (\partial\mathcal{C}/\partial\dot{q})[\delta q] ( ∂ C / ∂ q ˙ ) [ δ q ] . This gives
\begin{aligned}
\delta\mathcal{C}
&=\int_{B_s} J_s^{\circledS}\left(\nabla\cdot\left(\delta\chi_s^{\circled e}-\phi_f\left(F_sF_{sf}^{-1}\delta\chi_{sf}^{\circled e}\right)\right)^{\circledS}\right)^{\circledS}\\
&=\int_{B_s}\left(\nabla\cdot\left(p^{\circledS}J_sF_s^{-1}\left(\delta\chi_s-\phi_f^{\circledS}F_sF_{sf}^{-1}\delta\chi_{sf}\right)\right)\right.\\
&\qquad\left.-\nabla\cdot\left(J_sp^{\circledS}F_s^{-\mathsf{T}}\right)\cdot\left(\delta\chi_s-\phi_f^{\circledS}F_sF_{sf}^{-1}\delta\chi_{sf}\right)\right)\\
&=-\int_{S_s}\left[\left(p^{\circledS}J_sF_s^{-1}\left(\delta\chi_s-\phi_f^{\circledS}F_sF_{sf}^{-1}\delta\chi_{sf}\right)\right)\right]\cdot m_s\\
&\qquad-\int_{B_s}\nabla\cdot\left(J_sp^{\circledS}F_s^{-\mathsf{T}}\right)\cdot\delta\chi_s\\
&\qquad+\int_{B_s}\underbrace{\phi_f^{\circledS}\nabla\cdot\left(J_sp^{\circledS}F_s^{-\mathsf{T}}\right)}_{\nabla\cdot\left(J_sp^{\circledS}\phi_f^{\circledS}F_s^{-\mathsf{T}}\right)-J_sp^{\circledS}F_s^{-\mathsf{T}}\nabla\phi_f^{\circledS}}\cdot F_sF_{sf}^{-1}\delta\chi_{sf}. \tag*{(101)}
\end{aligned}
Similarly, we have
δ C f l t = ∫ S s − p Ⓢ [ ϕ f Ⓢ F s F s f − 1 δ χ s f ⋅ J s F s − T m s ] . (102) \delta\mathcal{C}_{\mathrm{flt}}=\int_{S_s}-p^{\circledS}\left[\phi_f^{\circledS}F_sF_{sf}^{-1}\delta\chi_{sf}\cdot J_sF_s^{-\mathsf{T}}m_s\right]. \tag*{(102)} δ C flt = ∫ S s − p Ⓢ [ ϕ f Ⓢ F s F s f − 1 δ χ s f ⋅ J s F s − T m s ] . (102)
The variation of R \mathcal{R} R , again computed by taking into account that the dissipation is written in rate form, is
\begin{aligned}
\delta\mathcal{R}
&=\int_{B_s}J_s\left(2\mu_f\mathbf{D}_f:\nabla\left(\delta\chi_s-F_sF_{sf}^{-1}\delta\chi_{sf}\right)^{\circled e}\right.\\
&\qquad-2\mu_B(\mathbf{D}_f-\mathbf{D}_s):\nabla\left(F_sF_{sf}^{-1}\delta\chi_{sf}\right)^{\circled e}\\
&\qquad\left.-\frac{\mu_D\phi_f^2}{\kappa_s}\left(\mathbf{v}_f^{\circled e}-\mathbf{v}_s^{\circled e}\right)\cdot\left(F_sF_{sf}^{-1}\delta\chi_{sf}\right)^{\circledS}\right)\\
&\qquad-\int_{S_s}\frac{1}{2}\mu_S\left[\phi_f^{\circledS}\left(\mathbf{v}_f^{\circledS}-\mathbf{v}_s^{\circledS}\right)\right]\cdot\left[\phi_f^{\circledS}F_sF_{sf}^{-1}\delta\chi_{sf}\right]\left|J_s\right|\left|F_s^{-\mathsf{T}}m_s\right|. \tag*{(103)}
\end{aligned}
The above expression can be rewritten as
\begin{aligned}
\delta\mathcal{R}
&=\int_{B_s}J_s\left(2\mu_f\mathbf{D}_f:\nabla\delta\chi_s^{\circled e}-\mathbf{T}_f^{\mathsf{v}}:\nabla\left(F_sF_{sf}^{-1}\delta\chi_{sf}\right)^{\circled e}\right.\\
&\qquad\left.-\frac{\mu_D\phi_f^2}{\kappa_s}\left(\mathbf{v}_f^{\circled e}-\mathbf{v}_s^{\circled e}\right)\cdot\left(F_sF_{sf}^{-1}\delta\chi_{sf}\right)^{\circled e}\right)^{\circledS}\\
&\qquad-\int_{S_s}\left[\frac{1}{2}\mu_S\left[\phi_f^{\circledS}\left(\mathbf{v}_f^{\circledS}-\mathbf{v}_s\right)\right]\cdot\phi_f^{\circledS}F_sF_{sf}^{-1}\delta\chi_{sf}\right]\left|J_s\right|\left|F_s^{-\mathsf{T}}m_s\right|. \tag*{(104)}
\end{aligned}
Using the fact that for a field with domain in B ( t ) B(t) B ( t ) like, say, T f \mathbf{T}_f T f , we have
J s ( ∇ ⋅ T f ) Ⓢ = ∇ ⋅ ( J s T f Ⓢ F s − T ) , J_s\left(\nabla\cdot\mathbf{T}_f\right)^{\circledS}=\nabla\cdot\left(J_s\mathbf{T}_f^{\circledS}F_s^{-\mathsf{T}}\right), J s ( ∇ ⋅ T f ) Ⓢ = ∇ ⋅ ( J s T f Ⓢ F s − T ) ,
we can further rewrite Eq. (B.22) as
δ R = ∫ B s [ ∇ ⋅ ( ( 2 μ f J s D f ⊙ F s − T ) T δ χ s ) − ∇ ⋅ ( ( J s ( T f v ) ⊙ F s − T ) T F s F s f − 1 δ χ s f ) − ∇ ⋅ ( 2 μ f J s D f ⊙ F s − T ) ⋅ δ χ s + ∇ ⋅ ( J s ( T f v ) ⊙ F s − T ) ⋅ F s F s f − 1 δ χ s f − J s ( μ D ϕ f 2 κ s ( v f e − v s e ) ⋅ ( F s F s f − 1 δ χ s f ) e ) ⊙ ] − ∫ S s [ 1 2 μ S ∥ ϕ f ⊙ ( v f ⊙ − v s ) ∥ ⋅ ϕ f ⊙ F s F s f − 1 δ χ s f ] J s ∥ F s − T m s ∥ . (105) \begin{aligned}
\delta\mathcal{R}={}&\int_{B_s}\Bigl[
\nabla\cdot\Bigl((2\mu_fJ_sD_f^{\odot}F_s^{-\mathrm{T}})^{\mathrm{T}}\delta\chi_s\Bigr)\\
&-\nabla\cdot\Bigl((J_s(T_f^v)^{\odot}F_s^{-\mathrm{T}})^{\mathrm{T}}F_sF_{sf}^{-1}\delta\chi_{sf}\Bigr)\\
&-\nabla\cdot\Bigl(2\mu_fJ_sD_f^{\odot}F_s^{-\mathrm{T}}\Bigr)\cdot\delta\chi_s+\nabla\cdot\Bigl(J_s(T_f^v)^{\odot}F_s^{-\mathrm{T}}\Bigr)\cdot F_sF_{sf}^{-1}\delta\chi_{sf}\\
&-J_s\Bigl(\frac{\mu_D\phi_f^2}{\kappa_s}(v_f^e-v_s^e)\cdot(F_sF_{sf}^{-1}\delta\chi_{sf})^e\Bigr)^{\odot}
\Bigr]\\
&-\int_{S_s}\Bigl[\frac{1}{2}\mu_S\lVert\phi_f^{\odot}(v_f^{\odot}-v_s)\rVert\cdot\phi_f^{\odot}F_sF_{sf}^{-1}\delta\chi_{sf}\Bigr]J_s\lVert F_s^{-\mathrm{T}}m_s\rVert. \tag*{(105)}
\end{aligned} δ R = ∫ B s [ ∇ ⋅ ( ( 2 μ f J s D f ⊙ F s − T ) T δ χ s ) − ∇ ⋅ ( ( J s ( T f v ) ⊙ F s − T ) T F s F s f − 1 δ χ s f ) − ∇ ⋅ ( 2 μ f J s D f ⊙ F s − T ) ⋅ δ χ s + ∇ ⋅ ( J s ( T f v ) ⊙ F s − T ) ⋅ F s F s f − 1 δ χ s f − J s ( κ s μ D ϕ f 2 ( v f e − v s e ) ⋅ ( F s F s f − 1 δ χ s f ) e ) ⊙ ] − ∫ S s [ 2 1 μ S ∥ ϕ f ⊙ ( v f ⊙ − v s )∥ ⋅ ϕ f ⊙ F s F s f − 1 δ χ s f ] J s ∥ F s − T m s ∥ . (105)
Applying integration by parts, we then obtain
δ R = − ∫ S s [ 2 μ f D f ⊙ J s F s − T m s ⋅ δ χ s + J s ( T f v ) ⊙ F s − T m s ⋅ F s F s f − 1 δ χ s f ] + ∫ B s [ − ∇ ⋅ ( 2 μ f J s D f ⊙ F s − T ) ⋅ δ χ s + ∇ ⋅ ( J s ( T f v ) ⊙ F s − T ) ⋅ F s F s f − 1 δ χ s f − J s ( μ D ϕ f 2 κ s ( v f e − v s e ) ⋅ ( F s F s f − 1 δ χ s f ) e ) ⊙ ] − ∫ S s [ 1 2 μ S ∥ ϕ f ⊙ ( v f ⊙ − v s ) ∥ ⋅ ϕ f ⊙ F s F s f − 1 δ χ s f ] J s ∥ F s − T m s ∥ . (106) \begin{aligned}
\delta\mathcal{R}={}&-\int_{S_s}\bigl[2\mu_fD_f^{\odot}J_sF_s^{-\mathrm{T}}m_s\cdot\delta\chi_s\\
&\qquad+J_s(T_f^v)^{\odot}F_s^{-\mathrm{T}}m_s\cdot F_sF_{sf}^{-1}\delta\chi_{sf}\bigr]\\
&+\int_{B_s}\Bigl[-\nabla\cdot\Bigl(2\mu_fJ_sD_f^{\odot}F_s^{-\mathrm{T}}\Bigr)\cdot\delta\chi_s+\nabla\cdot\Bigl(J_s(T_f^v)^{\odot}F_s^{-\mathrm{T}}\Bigr)\cdot F_sF_{sf}^{-1}\delta\chi_{sf}\\
&\qquad-J_s\Bigl(\frac{\mu_D\phi_f^2}{\kappa_s}(v_f^e-v_s^e)\cdot(F_sF_{sf}^{-1}\delta\chi_{sf})^e\Bigr)^{\odot}\Bigr]\\
&-\int_{S_s}\Bigl[\frac{1}{2}\mu_S\lVert\phi_f^{\odot}(v_f^{\odot}-v_s)\rVert\cdot\phi_f^{\odot}F_sF_{sf}^{-1}\delta\chi_{sf}\Bigr]J_s\lVert F_s^{-\mathrm{T}}m_s\rVert. \tag*{(106)}
\end{aligned} δ R = − ∫ S s [ 2 μ f D f ⊙ J s F s − T m s ⋅ δ χ s + J s ( T f v ) ⊙ F s − T m s ⋅ F s F s f − 1 δ χ s f ] + ∫ B s [ − ∇ ⋅ ( 2 μ f J s D f ⊙ F s − T ) ⋅ δ χ s + ∇ ⋅ ( J s ( T f v ) ⊙ F s − T ) ⋅ F s F s f − 1 δ χ s f − J s ( κ s μ D ϕ f 2 ( v f e − v s e ) ⋅ ( F s F s f − 1 δ χ s f ) e ) ⊙ ] − ∫ S s [ 2 1 μ S ∥ ϕ f ⊙ ( v f ⊙ − v s )∥ ⋅ ϕ f ⊙ F s F s f − 1 δ χ s f ] J s ∥ F s − T m s ∥ . (106)
The claim follows from substituting the contributions from Eqs. (B.16), (B.18)–(B.20), and Eq. (B.24) into Eq. (47) , applying a localization argument in time, and recalling the definitions for T s T_s T s , T f T_f T f , and p f \mathbf{p}_f p f .
Acknowledgments
The authors gratefully acknowledge partial support from the Pennsylvania Department of Health using Tobacco CURE Funds (the Department specifically disclaims responsibility for any analyses, interpretations, or conclusions).
The authors also wish to express their gratitude to Profs. Bruce J. Gluckman and Patrick J. Drew, both of the Penn State Center for Neural Engineering, and to Prof. Anna Mazzucato of the Penn State Department of Mathematics for their feedback and suggestions.
CID: PMC9107702.
PMC8939891.