Modification of a poroelastic model for zero porosity: finite element implementation and investigation of fluid mechanics in the perivascular space Authors: Mohammad Jannesari, Beatrice Ghitti, Bruce J Gluckman, Francesco Costanzo Published: 2026 ABSTRACT AbstractIn conventional formulations of poroelasticity, when the porosity approaches zero or vanishes in some parts of the poroelastic domain, if only temporarily, the governing equations degenerate to those for the solid phase thereby inhibiting a suitable determination of the fluid velocity field. To address this challenge, we reformulated a poroelastic model based on mixture theory to accommodate scenarios with zero porosity. We verified our model using the method of manufactured solutions and demonstrated its ability to handle extreme conditions in a sample test problem. As an application of our framework, we investigated peristaltic flow in the perivascular space of a penetrating arteriole in brain. Our analysis revealed that some literature-suggested parameters can drive the model to predict extreme non-physiological conditions. We further demonstrated that these extreme conditions can be somewhat mitigated by accounting for the deformation of the surrounding brain tissue.Keywords: Poroelasticity, Zero porosity, Perivascular space, Cerebrospinal fluid mechanics, Finite element method Introduction The brain is housed within the skull and immersed in cerebrospinal fluid (CSF), which circulates around the brain and spinal cord (Kucewicz et al. 2007; Tarasoff-Conway et al. 2015; Abbott et al. 2018; Hladky and Barrand 2018, 2022). The brain deformation is hypothesized to assist in this circulation by promoting the removal of metabolic waste products from the central nervous system (CNS) (Sloots et al. 2020). In addition to the large-scale circulation of CSF, evidence suggests that interstitial fluid (IF) also flows through the brain tissue itself and exchanges metabolites and other solutes with the broader CSF system (cf. Iliff et al. 2012, 2013a, b; Iliff and Nedergaard 2013; Nedergaard 2013; Rasmussen et al. 2022). Understanding these fluid flows is critical to gaining insights into both normal brain physiology and the development of pathological conditions such as neurodegenerative diseases (Abbott et al. 2018; Nedergaard 2013; Hladky and Barrand 2022). However, directly observing and quantifying these flows in vivo is extraordinarily challenging via currently available experimental techniques. For example, the pressure differences driving CSF through perivascular space (PVS) can be as small as 0.01mmHg—100 times smaller than the approximately 1mmHg resolution of current instruments (Oshio et al. 2004; Uldall et al. 2017)—and are further disrupted by invasive skull access (Mestre et al. 2018). As a result, physics-based mathematical modeling has become an attractive tool to evaluate competing hypotheses about the drivers and pathways of brain fluid flow (Hladky and Barrand 2022). Our group has been actively engaged in this field, especially in exploring the dynamics of fluid flow and exchange between the brain parenchyma and the subarachnoid space (SAS). In particular, CSF flow and exchange in the PVS have been investigated using computational modeling techniques, showing that, under idealized conditions, arterial pulsations alone may not suffice to drive flow within the arterial PVS (Kedarasetti et al. 2020a, b). A crucial insight from our work is the importance of accounting for the deformability of brain tissue when modeling these processes. In fact, we have shown that mechanical deformations affect the effective permeability of the brain parenchyma and influence how fluids can move through it (Kedarasetti et al. 2022; Garborg et al. 2025; Costanzo et al. 2025). In the course of our modeling effort, we encountered scenarios in which numerical methods struggled to adequately represent the physics. Specifically, under the action of periodic compressive loads, numerical experiments showed that deformation can strongly affect porosity so as to drive fluid out of some regions of the simulation’s domain, thereby leaving these regions nearly or entirely depleted of fluid. These results revealed a limitation in our poroelastic models when dealing with extreme cases of near-zero porosity, pointing out the need for a more refined model that can handle and accurately represent an extreme condition like vanishing porosity. Pursuing the creation of such a model is not motivated merely by numerical considerations. We believe that a stronger rationale stems from the role of computational modeling in testing physiological hypotheses through the lens of physics. Since some of these hypotheses may initially be incomplete or oversimplified, it is valuable to identify when and why a model prediction departs significantly from physiologically relevant conditions. This allows us to correct or discard any nonphysical assumptions. Motivated by these considerations, in this work we reformulate our poroelastic framework, previously adopted in several publications (Kedarasetti et al. 2020a, b, 2022; Garborg et al. 2025), to handle situations where deformation completely expels the fluid out from parts of a poroelastic body, resulting in regions of vanishing porosity. We implement the model numerically, verify it using the method of manufactured solutions, and identify stable finite element spaces suitable for these cases. Although this study does not focus primarily on modeling brain physiology, here we revisit a flow problem simulating fluid flow in the PVS surrounding a cerebral vessel, caused by peristaltic motion of the arterial walls, to demonstrate the relevance of our approach. This problem illustrates how assumptions and parameter values in a model can yield predictions that diverge from physiological expectations. In addition, our findings further reinforce our belief that accounting for tissue deformability and the physics of porosity dynamics are essential to reach consistent and meaningful conclusions. In Sect. 2, we revisit the traditional formulation of poroelasticity based on mixture theory. The modifications to capture the vanishing-porosity behavior are introduced in Sect. 3. The weak form of the refined model is presented in Sect. 4, where a verification test via the method of manufactured solutions and a local compression benchmark problem are also performed (Sects. 4.2 and 4.3, respectively). Finally, in Sect. 5 we apply our model to the analysis of fluid flow in the PVS by peristalsis. Traditional formulation: definitions and governing equations Here we report the governing equations of a poroelastic body starting from mixture theory (Bowen 1976). Constituent phases and mixture We consider a saturated binary mixture of a solid and a fluid phase. At the current time t, the mixture occupies the subset Bm(t) (Fig. 1; m stands for ‘mixture’) of an all-containing Euclidean point space Ed, with d=1,2 or 3. The constituents of the mixture can coexist at each spatial point x∈Bm(t). Subscripts s and f will be used to denote quantities pertaining to the solid and fluid phases, respectively. As shown in Fig. 1, we posit that there exists a motion χa:Ba×[0,T]→Bm(t)⊂Ed which maps the reference configuration Ba of component a (a=s,f) to the mixture’s current configuration Bm(t). For each fixed t∈[0,T], the map χa is assumed to be a diffeomorphism from Ba to Bm(t). Also, for a given (fixed) time t, with a slight abuse of notation, we will denote by χa-1 the inverse of χa at time t. The position of points in Ba will be denoted by Xa. For simplicity, the initial configurations of the components are assumed to coincide with their respective reference configurations and with Bm(0), and therefore with the initial configuration of the porous structure as a whole. We regard Ed as covered by a single Cartesian coordinate system. We denote by V the translation space of Ed and by E={e1,…,ed} the orthonormal basis of V that is induced by the coordinate cover. Current configuration Bm(t) of a biphasic mixture, along with the reference configurations of solid and fluid components, Bs and Bf, respectively. Positions within Bs and Bf will be labeled as Xs and Xf, respectively, while x signifies the position within Bm(t). The boundaries of Bs, Bf, and Bm(t) are ∂Bs, ∂Bf, and ∂Bm(t), respectively. These boundaries are oriented by their respective outward unit normal fields ns, nf, and nm. We note that the choice of reference configuration of each phase is arbitrary (cf. Bowen 1976). This arbitrariness allows one to choose the reference configurations of the phases to coincide with some single convenient domain. As done later in this paper, we chose Bs and Bf to coincide with Bm(0), the initial configuration of the mixture Concerning the domains of functions The functions describing the physics of the system can have domains Bs, Bf, and Bm(t). Later, we will reformulate the equations to solve them numerically via the finite element method (FEM) while adopting an arbitrary Lagrangian–Eulerian (ALE) scheme (cf. Fernández et al. 2009). Therefore, we will need to refer to the same physical quantity but expressed by different functions when said quantity is remapped over a specific domain. With this aim, and to avoid the proliferation of symbols, we adopt the short-hand notation in dell’Isola et al. (2009) to indicate the domain to which a function is being remapped. With a,b=s,f,m, let φ:Ba→U and ζ:Bb→W be smooth functions taking values in the vector spaces U and W, respectively. Let χab:Ba→Bb be a diffeomorphism between Ba and Bb. Then, we denote by and the following functions Kinematics Following Bowen (1976) and Costanzo and Miller (2017), for each constituent a=s,f, the displacement field, the deformation gradient, and its determinant are defined as, respectively, where we note that ua=x-Xa since x=χaXa,t. The material velocity for the constituent a is We note that the domain of va is Bm(t). The material acceleration of phase a is also a function with domain Bm(t) and is given by where the notation ∇□[⋄] denotes the action of the gradient of the quantity ‘□’ onto the field ‘⋄,’ provided such an action is meaningful. For future use we will denote the divergence operator by ‘∇·’. Remark 1 (Concerning ∇ and ∇·) The gradient and divergence operators presume the use of an underlying metric, the latter defined over a specific domain. As functions can have different domains, the meaning and practical computation of the action of these operators cannot be divorced from the domain of said function. Therefore, as an example, letting p:Bm(t)→R, the functions ∇p and are different from one another because the former is the gradient of p over Bm(t) whereas the latter is the gradient over Bs of the function . The volume fraction of species a will be denoted by ϕa(x,t). We assume that the porous medium is saturated, so we have ϕs+ϕf=1. For fluid flow models in which the fluid is assumed incompressible (the fluid’s mass density is a constant), the fluid’s velocity flux through a surface can be used as a direct indicator of the corresponding mass flux. This fact justifies the introduction of flow indicators like the flow rate in engineering fluid mechanics (cf. Çengel and Cimbala 2017). However, in the case of flow through a porous medium, the fluid’s velocity flux can no longer be taken as a direct indicator of a corresponding mass flux through the porous medium. In our current modeling context, the information in question is linked to (i) the velocity of the fluid relative to that of the solid when (ii) this velocity is appropriately scaled by the volume fraction of the fluid phase. Therefore, in porous flow models, especially when the phases are taken to be incompressible in their pure state, a new velocity field is introduced that can be used in a way similar to the true fluid velocity in pure fluid flow models. This velocity field is the filtration velocity, which is defined as In the Darcy flow problem, the filtration velocity is the primary unknown of the problem (cf., e.g., Whitaker 1986; Wang and Olbricht 2011) and it vanishes when ϕf→0. Furthermore, in the traditional form of the Darcy problem, it is the filtration velocity that is involved in the statement of boundary conditions (e.g., by setting the normal component of the filtration velocity to zero to characterize an impermeable boundary; see, e.g., Masud (2007)). Governing equations We assume that the mixture’s constituents are incompressible in their pure form. Enforcing this constraint along with the balance of mass in the absence of chemical reactions, we have where ϕRa:Ba→R is the referential volume fraction of component a (cf. Costanzo and Miller 2017). Summing the first of Eq. (8) for the solid and fluid components and considering the saturation constraint, we then have We will refer to this equation as the continuity equation as it plays a role practically identical to the continuity equation in the Stokes and Navier–Stokes problems. For the solid phase, let Cs denote the right Cauchy–Green strain tensor and C¯s its corresponding isochoric component, i.e., We assume that the solid phase in its pure form is hyperelastic and we denote by Ψ^s=Ψ^sC¯s its strain energy per unit volume. We then assume that the strain energy per unit volume of the solid’s phase in its reference configuration is In the current study, we select a neo-Hookean material model so that Ψ^s=12μsI¯1-3, where I¯1 represents the first principal invariant of C¯s and μs denotes the solid shear modulus. The elastic and total Cauchy stress tensors for the solid phase are (cf. Bowen 1980) where p=p(x,t) is the pore pressure. When surface tension effects are neglected in the porous medium, the interaction of the mixture’s constituents is typically limited to a drag force proportional to the velocity of the fluid relative to the solid. Then, following Bowen (1976) and Costanzo and Miller (2017), and enforcing the balance of momentum for the phases as well as the mixture, the equations of motion for the solid and fluid constituents are, respectively, where μf is the dynamic viscosity of the fluid component, κs is the permeability of the solid component, ba is the body force per unit mass of the constituent a, and ρa is the mass density of constituent a per unit volume of the mixture, given by where ρa∗ is the density of the constituent a in its pure form. Taken together, Eqs. (9), (13), and (14) constitute the governing equations of the system. Often Eq. (14) is presented in a slightly simplified form under the assumption that ϕf>0. In this case, one can divide Eq. (14) by ϕf and write which is an equation often associated with Darcy flow, especially when it is assumed that the solid phase is stationary (cf. Wang and Olbricht 2011). In other formulations, Eq. (13) is replaced by an equation of motion for the mixture as a whole, obtained by summing Eqs. (13) and (14): Remark 2 (The need for a modified model for vanishing porosity) Recalling the saturation constraint (ϕs+ϕf=1) and assuming that every field describing the physics remains bounded, we observe that the limit as ϕf→0 of Eq. (9) can be considered well defined. The same can be said for Eqs. (13) and (16). However, as ϕf→0, both Eqs. (14) and (15) present some issues. Equation (14) would become trivial altogether, whereas Eq. (15) becomes undefined since its very derivation required a division by ϕf. Even if one were to take Eq. (15) as the equation of motion of the fluid (as opposed to having been derived from a parent statement), the limit for ϕf→0 would present some difficulties as it would imply that the fluid’s inertia does not vanish with the fluid’s volume fraction and that the motion of the fluid is driven by the gradient of p, the latter having become a multiplier for the enforcement of the constraint needed to ensure that the motion of the solid is volume preserving. Remark 3 (Concerning traction boundary conditions) The application of traction boundary conditions in mixture theory is a delicate subject (cf. Rajagopal and Tao 1995; Rajagopal et al. 1986). In this paper, we will not need a model to manage the partition of a traction distribution on the boundary of the mixture in corresponding traction distributions over individual phases. Hence, for later use, we simply denote by s the boundary traction on the mixture: where T is the total Cauchy stress acting on ∂Bm(t) and nm is the outer unit normal orienting ∂Bm(t). For the model presented thus far, we have Reformulation for the case of vanishing porosity In elasticity, the numerical analysis of incompressible or nearly incompressible materials—where the Poisson’s ratio approaches 0.5—presents specific numerical challenges that must be addressed to obtain accurate solutions (see, e.g., Glowinski et al. 1982). Analogously, the common formulation of poroelasticity cannot handle the limit-case condition with ϕf→0. If the porosity ϕf approaches zero at some point x in the domain Bm(t), the equation of motion of the fluid, Eq. (14), becomes trivial at that point, leading to a loss of well-posedness and resulting in a singular problem. It is well known that numerical failure in solving the governing equations does not require ϕf=0; the system may become unstable or ill-posed when ϕf is sufficiently small in certain regions of the domain. These observations justify the need to develop a refined version of the model, one that can deliver solutions, especially numerical solutions, even when ϕf approaches zero over some subset of the solution’s domain. To formulate a modification of Eq. (14), we start with the observation that, in a continuum-level mixture theory, all phases coexist at all the spatial points occupied by the mixture. That is, there is no explicit recognition that the solid and fluid phases are otherwise separated so to occupy distinct regions within Bm(t). This said, we envision a microscopic picture in which this separation does exist. In this framework, we view the condition ϕf→0 as describing the vanishing of the physical space within which the fluid particles can move. We also assume that, at microscopic scale, a no-slip condition exists at any surface in which the fluid comes into contact with the solid phase. Therefore, we posit that, as ϕf→0, the fluid velocity vf approaches the solid velocity vs, i.e., we posit that if ϕf→0 at a point, ‖vf-vs‖→0 at that point. This assumption is inherently related to the question pertaining to the scaling of the permeability κs with the fluid porosity. This question has been addressed both empirically and theoretically, and the associated literature supports our assumption (cf. Yuan et al. 2024; O’Brien et al. 2006; Carman 1939; Gebart 1992). Remark 4 (The filtration velocity and its dependence on porosity) Equation (7) implies that ‖vflt‖→0 when ϕf→0. In light of the assumption we have made concerning ‖vf-vs‖, we can say that, at those points where ϕf→0, ‖vflt‖=o(ϕf), that is, ‖vflt‖ goes to zero faster than ϕf. Given the above premise, in order to formulate a suitable modification of Eq. (14), we define an auxiliary variable η=η(ϕf) as such that where μf is still treated as a material constant. The condition in Eq. (20) allows us to formalize a constitutive assumption concerning the permeability of the porous medium. Specifically, we assume that By dividing Eq. (14) by η(ϕf), then we have which, under the physically reasonable assumption that af-bf and ∇p are bounded fields over Bm(t), fulfills the assumed limit-case condition that vf→vs as ϕf→0. Remark 5 (On the boundedness of physical quantities) From a physical perspective, it is indeed reasonable to assume that af-bf and ∇p be bounded fields over Bm(t). In general, bf is a prescribed field and therefore its boundedness can be taken for granted. However, the boundedness of the other two quantities (∇p in particular) cannot be taken for granted from the perspective of the theory of partial differential equations depending on topological considerations and the regularity of the data of the underlying boundary value problem. The equation of motion for the mixture, that is, Eq. (16), is not affected by the limit case condition, since ϕf does not directly contribute to this equation. Indeed, as ϕf→0, ρf approaches zero, and the equation of motion for the solid component is recovered from Eq. (16), as expected. The equation arising from the balance of mass and incompressibility, namely Eq. (9), remains meaningful in the limit case. In fact, recalling the saturation condition ϕf+ϕs=1, we have that ϕsvs+ϕfvf=vs+vflt. Hence, recalling that ‖vflt‖→0 as ϕf→0, for the limit behavior of Eq. (9) we have which is the expression of the incompressibility of the pure solid phase (in rate form). Remark 6 (On the linearization of the general theory) Deriving a linearized theory from a general setting requires a series of formal steps that include the choice of fields relative to which the final form of the equations is to be linear, the choice of a configuration as the domain of the linearized equations, and the choice of a physical state at which the coefficients in the equations are evaluated. A rigorous linearization process applicable to this paper can be found in Sect. 7 of Bowen (1980). In this Remark, we limit ourselves to comment on the form of the porosity and of Eq. (9) under linearization relative to a quiescent equilibrium state, having chosen the Bs as the underlying domain of the linearized equations. Under these conditions, the linearized expressions of the porosity and the continuity equation are, respectively, The function ϕRs can be position-dependent and it can even be equal to zero over some subset of Bs. However, this function is not time-dependent. Therefore, while the first of Eqs. (24) does imply the possibility that, at a time t^ during a deformation process, ϕf might vanish at a point with ϕRs≠0, at that same point and time the linearized continuity equation would remain “ignorant” of such an event. This conclusion is applicable even in relation to Eqs. (13) and (14). In general, the linearization process turns the coefficients ϕf and ϕs in the field equations into time-independent objects thereby removing from said equations feedback effects like the one considered in Eq. (23). Hence, a theory in which the field equations are sensitive to the vanishing of the porosity is, in our view, inherently nonlinear in the primary unknown fields. There is no unique way to satisfy Eq. (21). Possible practicable approaches can be derived by considering the experimental literature (e.g., see Yuan et al. 2024). In addition to experimental considerations, which require to be adapted to specific physical systems, considerations can be included pertaining to practical implementations within a numerical framework. In order to meet the condition stated in Eq. (20), and for the purpose of illustrating some practical examples, here we propose the following definition for κs(ϕf) where κ¯s is the saturation value of permeability and ϕfc is a critical porosity value. The behavior of κs(ϕf)/κ¯s for 0≤ϕf≤1 is shown in Fig. 2, for specific choice of parameter ϕfc. It is worth mentioning that definition (25) of κs(ϕf) conforms to some of the experimental data available in the literature (cf. Yuan et al. 2024); clearly, other functional forms of solid permeability can be employed. Plot of κs/κ¯s as a function of ϕf/ϕfc to illustrate the behavior of the permeability κs of the solid component as defined in Eq. (25). The permeability remains constant when ϕf>ϕfc Weak formulations and verification In this section, we present the weak form that we choose for the problem’s governing equations. The weak form in question is then implemented in a FEM for the solution of the examples presented in this paper. Overall, the FEM we implement is an ALE scheme (cf. Fernández et al. 2009) where the computational domain is chosen as the initial configuration of the solid phase, the latter having been chosen to coincide with Bs. Formulation We begin by rewriting the equations as a single system: where η(ϕf) is given in Eq. (19), and where such that Here, P is the first Piola–Kirchhoff stress tensor for the mixture, Pse is the first Piola–Kirchhoff elastic stress tensor for the solid phase, and Sse,∗ is the second Piola–Kirchhoff elastic stress tensor for the pure solid phase. The strong form of the equations will now be reformulated in a weak form after we introduce appropriate functional spaces for the primary unknown fields. From a physics’s perspective, we choose the problem’s primary unknowns to be the solid’s displacement us, the fluid’s velocity vf, and the pore pressure p. However, we will solve the governing equations by adopting an ALE scheme according to which the equations are pulled back to Bs. Therefore, from an implementation perspective, the primary unknowns are us, , and , where and are viewed as primitive quantities, whereas vf and p are images under χs of and . To facilitate the expression of the boundary conditions, let Γs denote the boundary of Bs, i.e., ∂Bs in Fig. 1. With i=1,…,d, we then consider the following partitions: where the superscripts ‘D’ and ‘N’ stand for ‘Dirichlet’ and ‘Neumann’, respectively. Then, with a slight abuse of notation, let us∈Vus, vf∈Vvf, and p∈Vp, with where ugi is the Dirichlet data for the i-th component of us, pg is the Dirichlet data for the pore pressure field, and where denotes the image of Γsp under χs. We also define Remark 7 (On the regularity of us) The overall problem will be solved via an ALE scheme that remaps all equations over Bs. Hence, the regularity of us indicated in Eq. (33) is needed to guarantee the assumed regularity of the remaining fields. Remark 8 (On the regularity of p) The choice of Vp is motivated by the way in which we weakly impose Eq. (28). Specifically, after multiplying the latter by an appropriate test function, we will integrate by parts the resulting equation. In turn, this approach is motivated by a desire to avoid dealing with gradients of the mixture’s volume fractions. Depending on the context, by the notation (·,·)□ we will denote the scalar product over the domain ‘□’. Furthermore, to avoid proliferation of symbols, we denote the test functions for a field by placing a tilde over the symbol of the field in question. As the weak form of the equation of motion for the mixture in the reference configuration, we choose the following statement, which we impose for all u~s∈V0s: where, letting sgi be the prescribed i-th component of the traction acting on the boundary of the current configuration of the mixture, and recalling that the outward unit normal on ∂Bm(t) is , Differently from what is typically done in mixed FEM for fluids problems, the weak form in Eq. (38) does not manage the pressure gradient terms in the strong form of the equation through the lens of the divergence theorem. We choose as the weak form of the equation of motion for the fluid the following statement, that we impose for all v~f∈Vvf: Note that in Eq. (40) integration by parts is not applied to the pressure term. Finally, we consider the weak form of the continuity equation. Before doing so, it is important to remark that in poroelasticity problems in which the porous phase and the fluid interaction is described by the last term on the left-hand side of Eq. (14), there is a boundary condition that can be enforced limited to the normal component of the filtration velocity. In the most traditional form of the Darcy flow problem where the solid phase is assumed to be stationary, this condition corresponds to prescribing the normal component of the fluid’s velocity on the boundary of the domain (cf. Masud 2007). Referring to Eq. (7) and the boundary partitions in Eqs. (31) and (32), let vfltg denote the prescribed value of vflt·nm on , then we weakly enforce Eq. (28) by requiring that, for all p~∈V0p, we have In conclusion, we determine the motion of the poroelastic mixture under consideration by solving the following Problem 1 Find us, , and in Vus, Vsf, and Vsp, respectively, such that for all u~s, v~f, and p~ in V0s, Vsf, and V0,sp, respectively, Eqs. (38), (40), and (41) are satisfied, where , and . Remark 9 (Consistency with the strong form) The fact that the weak problem we have formulated is equivalent to that presented in strong form can be shown via standard methods relying on integration by parts and therefore is omitted here. Remark 10 (On the unusual presentation of the weak forms) The presentation of the weak problem we solve relies on a slight abuse of notation in that we do not present equations all of which are pulled back to the chosen solution domain, namely Bs. The reason for this choice is twofold. First, the equation of motion for the fluid along with the continuity equation are easier to read and to relate to the strong form of the equations when presented in Eulerian form (i.e., using functions with domain Bm(t). Second, from a computer implementation viewpoint, the problem was solved using COMSOL Multiphysics™(CMPH) (COMSOL 2023). The latter can be used as a finite element library supporting ALE schemes whereby operators implemented in Eulerian form are evaluated while taking into account a given ALE map. This is to say that we presented the equations in the way we actually implemented them in CMPH. Verification by method of manufactured solutions Here we use the method of manufactured solutions (Salari and Knupp 2000) to verify the correctness of our computer implementation, and to empirically verify that the proposed formulation can indeed deliver results under conditions of zero porosity. This analysis also includes the empirical assessment of the convergence rates of our numerical approach. We selected finite element spaces within a 2D axisymmetric domain, whose reference configuration, denoted by A, has a radius of R=5mm and a height of L=10mm. The deformed configuration is denoted by A(t). The manufactured solution consists of the following fields: where po=100Pa, uo=0.1mm, and to=1s. Moreover, (R, Z) and (r, z) are the coordinates in the reference and deformed configurations, respectively. We considered two cases: one with ϕf=0.5 and the other with ϕf=0 with all other parameters as in Table 1. Table 1Parameter values used in the convergence rate study and the case studyParameterDescriptionValueρs∗Solid true density1×103kg/m3ρf∗Fluid true density1×103kg/m3μfFluid viscosity1×10-3Pa·sκ¯sSolid Saturated permeability1×10-9m2ϕfcCritical porosity0.1μsSolid shear modulus5kPa Parameter values used in the convergence rate study and the case study Referring to Eq. (45), we note that our chosen manufactured solution is such that, when ϕf=0, the velocity of the fluid is the same as the velocity of the solid. Remark 11 (The method of manufactured solutions and Eq. (28)) The application of the method of manufactured solutions requires a source term be present in each of the governing equations. The source terms in the momentum balance equations, namely Eqs. (26) and (27), are the terms bf and bs. By contrast, Eq. (28) does not have a source term. Hence, in applying the method of manufactured solutions, Eq. (28) is modified to have the form ∇·ϕsvs+ϕfvf-q=0, where q is the needed source term. In all physical problems of interest one must set q=0. The source term q is included here only for the purpose of verification within the use of the method of manufactured solutions. Clearly, including q in the strong form of the continuity equation requires the addition of the term -(p~,q)Bm(t) in Eq. (41). The initial conditions correspond to the manufactured solution evaluated at t=0. For the solid phase, boundary conditions are imposed in the form of prescribed displacement along the entire external boundary, except the top edge where the elastic traction obtained form the manufactured solution is applied. For the fluid phase, the normal component of the filtration velocity, as determined from the manufactured solution, is enforced along the external boundaries. Along the axis of symmetry, the radial component of the displacement field is set to zero, while the axial component is prescribed by the manufactured solution; similarly, the normal component of the filtration velocity is imposed based on the manufactured solution. The formulation is numerically implemented using CMPH (COMSOL 2023), which serves as a FEM programming environment, and solved adopting a monolithic solver with Newton’s method whose damping factor is determined automatically. First-order Lagrange polynomial interpolation over triangles is used for all fields except for the solid displacement. For the latter, we employ a MINI element (P1 element enriched with a cubic bubble) (Arnold et al. 1984). A mesh composed of identical triangular elements is utilized, and five successive uniform refinement iterations are performed. The initial characteristic mesh size is he=2.5mm. For time integration, second-order BDF is used with a constant time step of 5×10-4s, and the simulation is carried out until a final time of 5×10-2s. Remark 12 (On finite element spaces) The numerical implementation of our weak formulation produces a mixed FEM (cf. Brezzi and Fortin 1991). As such, one must confront the problem of selecting appropriate stable finite element spaces. A detailed discussion of the issue at hand is outside the scope of this paper and we simply point to references (Masud 2007; Rodrigo et al. 2018; Barnafi et al. 2021) for context. This said, the use of the method of manufactured solutions allowed us to empirically confirm that our choice of finite element spaces appears to be stable. To some extent, this finding is not surprising in that, in the vanishing porosity limit, the solid’s response is that of an incompressible hyperelastic system and our selection of finite element spaces is known to be stable for these problems (cf. Karabelas et al. 2020, 2022). Figure 3a and c show color maps of the radial and axial velocity components of the solid and fluid for ϕf=0.5 and ϕf=0, respectively, as obtained from our calculations using the most refined mesh. As seen in Fig. 3c, the fluid velocity converges to that of the solid in the limit case, as discussed in Sect. 3. Figure 3b and d present the convergence rates for ϕf=0.5 and ϕf=0, respectively. As shown in Fig. 3d, the convergence rates for the solid displacement and pressure error norms in the limit case are similar to those observed in a pure solid problem (cf. Masud and Truster 2013). Model verification via the method of manufactured solutions, with ϕf=0.5 in the first row and ϕf=0 in the second row. a and c show radial and axial components of the velocity field for the solid (left) and fluid phase (right) for ϕf=0.5 and ϕf=0, respectively. As expected, both components of the fluid velocity converge to those of the solid phase under the limit case condition. b and d represent convergence rates for relative ▵-norm (all fields’ L2-norm, pressure’s and solid displacement’s H1-semi-norms, and fluid velocity’s Hdiv-semi-norm) of the error e▵□=‖□-▪‖▵/‖□‖▵ between the numerical solution ▪ and exact solution □ at time t=5×10-2s for two cases of ϕf=0.5 and ϕf=0, respectively, obtained using P1 interpolation for all fields and a cubic bubble augmentation for the solid displacement (i.e., a MINI element, Arnold et al. (1984)) The results in Fig. 3 show that our numerical implementation of our proposed model is correct and that it is indeed capable of successfully handling even a case with porosity equal to zero. We note that the convergence rates for the fluid velocity are as one would normally expect them for a flow problem like, say, Stokes’s, with P1 interpolation for the fluid velocity (cf. Masud 2007). Case study: local compression of a porous cylinder To further investigate the performance of the proposed model under local extreme conditions, we simulate a benchmark problem involving a cylinder that is locally compressed. As the reference configuration (cf. Fig. 4a), we consider a 2D axisymmetric geometry with a radius of R=0.5mm and a height of L=10mm. The cylindrical domain is subject to radial squeezing due to a time-harmonic traction distribution purely in the (current) radial direction applied to the region located between Z0=2mm and Z1=3mm from the bottom boundary, as defined by the following equation and illustrated in Fig. 4a, where po=100Pa and to=0.4s. Moreover, rect(s) is a rectangular function defined to be twice continuously differentiable and to attain unity within the interval 0.2,0.8 of its domain 0,1, for which we employed the implementation available in CMPH (COMSOL 2023). Squeezing test for a full cylinder modeled as a 2D axisymmetric domain. a Problem setup. b Initial and c deformed configuration. d Evolution of porosity over time at a point located at the center of the squeezing zone along the symmetry axis, i.e. at the coordinate (0,Z0+Z1/2), and the average porosity over the radial line passing through the same point, as shown in a in red color. In the initial state, the fluid is homogeneously distributed. Under the applied traction, the fluid redistributes, resulting in fluid-free regions and regions where fluid and solid coexist The bottom boundary is fixed, whereas the top and lateral edges, except the squeezing zone, are traction-free. Zero pressure is prescribed at the top and bottom boundaries, while the lateral surface is impermeable (i.e., vflt·nm=0). The radial displacement and the normal component of the filtration velocity are constrained to zero along the axis of symmetry. Furthermore, quiescent initial conditions are imposed. The material properties employed in this study are reported in Table 1. A uniform mesh made up of identical triangular elements is used, with an element size of 0.1mm. Time integration is carried out using second-order BDF with a constant time step of 1×10-3s, and the simulation runs until a final time of 2s. All other settings, including the element type and solver configurations, are consistent with those adopted for the verification test in Sect. 4.2. The aim of this study is to evaluate the model’s ability to accurately represent regions of locally vanishing porosity. When the traction force is applied, the fluid is driven out of the porous medium. As a result, some areas within the porous medium become devoid of fluid, while other areas retain both solid and fluid phases simultaneously, as shown in Fig. 4c. Moreover, as demonstrated in Fig. 4d, both the porosity at the center of the squeezing zone along the symmetry axis and the average porosity along a radial line passing through the middle of the squeezing zone (see Fig. 4a) decrease rapidly during the initial phase of the applied elastic traction. Figure 5a and b show the norm of the filtration velocity and the norm of the relative velocity of the fluid with respect to the solid. As the porosity approaches zero, both the filtration velocity and the relative velocity (though at a slower rate) tend to zero, confirming the assumption that the fluid velocity converges to the solid velocity. a Magnitude of the filtration velocity ‖vflt‖ at t=2s. It can be observed that the filtration velocity approaches zero wherever the porosity vanishes. b Magnitude of the relative velocity ‖vf-vs‖ at t=2s. In regions where the porosity vanishes, the fluid and solid velocities tend to converge. It is worth noting that in regions far from the squeezing zone, both the solid and fluid velocities are zero. c The porosity ϕf obtained using the traditional poroelastic formulation. The simulation stops computing in the middle (t=5×10-2s) because the traditional formulation is numerically unstable as the porosity becomes very small (5×10-4) Therefore, the model effectively simulates the behavior of porous media under extreme conditions, where some regions become devoid of fluid. These findings add to the ones from Sect. 4.2, where it was shown that the proposed finite element spaces demonstrate stability and acceptable convergence rate. It is worth emphasizing that the conventional formulation of poroelasticity is not capable of handling situations where porosity is zero in some regions while a significant amount of fluid exists in other parts of the poroelastic body. Specifically, the traditional formulation fails to solve the problem when the porosity becomes sufficiently small, below 5×10-4 in this case, as shown in Fig. 5c. Therefore, in the context of modeling CSF circulation and exchange in the CNS, these advancements can potentially facilitate the development of computational models capable of handling extreme conditions, such as the spinal cord squeezing caused by abdominal muscle contraction in mice (Garborg et al. 2025). Peristalsis in perivascular space Motivation and governing equations As mentioned in the introduction, the field of neuroscience focused on the identification of drivers and pathways for the clearance of toxic metabolic by-products has seen significant applications of physics-based fluid transport models to brain research (Hladky and Barrand 2018, 2022). Here we revisit a problem of some importance concerning a specific hypothesis that has been strongly debated in the area of waste clearance from the brain. Cerebral blood vessels have several features that make them different from blood vessels in other parts of the body (cf., e.g., Abbott et al. 2018; Ballabh et al. 2004). One of these features is that said vessels appear to be surrounded by a barrier, almost a sheath, called the glia limitans, consisting of astrocytic end-feet (cf., e.g., Hladky and Barrand 2022; Abbott and Friedman 2012; Iliff et al. 2012). The space between the outer wall of a cerebral blood vessel and the glia limitans is called the perivascular space (PVS) and it is fluid-rich (Lam et al. 2017). Based on a large set of experiments, Nedergaard and co-workers (Iliff and Nedergaard 2013; Iliff et al. 2012, 2013a, b) hypothesized that CSF is drawn from the subarachnoid space into the PVS of surface arterial vessels penetrating into the brain. Furthermore, they hypothesized that wall movements due to heart-gated arterial pulsations could drive significant CSF flow (cf. for a careful discussion of what represents significant flow, see Kedarasetti et al. (2020a, 2020b, 2022)) in the PVS by peristalsis (Iliff et al. 2013b). Wang and Olbricht (WO) (2011) had studied peristaltic flow in the PVS in a very different context, namely that of convection-enhanced drug delivery. From a mathematical viewpoint, after modeling the PVS as a porous medium, WO derived an elegant analytical solution to a problem involving steady Darcy flow (cf. Coussy 2010) through the PVS. The steady flow field was that perceived by an observer traveling at a constant velocity with the peristaltic wave. The steady flow assumption, paired with the assumptions that the porosity was constant and uniform in the moving observer’s frame, allowed WO to determine an analytical solution without having to determine the deformation of the underlying solid skeleton. We became interested in understanding to what extent the work by WO could lend support to the hypothesis proposed by Nedergaard and co-workers. Furthermore, since WO’s analysis bypassed the determination of the deformation of the solid phase, we became interested in solving the problem numerically, this time assuming that the porosity was uniform in the reference configuration of the solid skeleton, thereby allowing the porosity to change spatially due to deformation. We now revisit the problem studied by WO, utilizing the poroelastic formulation proposed in this paper, as the commonly used formulation fails to capture the case of zero porosity. Following WO, Fig. 6 shows a schematic of the PVS surrounding a cerebral artery, with the arterial wall shown in red. The z-axis corresponds to the axis of the artery, and axial symmetry around this axis is assumed. The PVS is depicted in blue. Schematic of the 2D axisymmetric geometry of the revisited WO’s problem. The red region depicts the arterial wall undergoing a peristaltic motion with a wavelength λ and velocity c, propagating in the positive z-direction. The blue region illustrates the fluid within the PVS. The violet area corresponds to the brain tissue, which is separated from the PVS by the glia limitans membrane Its outer boundary, with radius Ro, corresponds to the glia limitans (Abbott et al. 2018). As done by WO, we assume that Ro is a constant, that is, we treat brain parenchyma as rigid, thus causing the glia limitans to remain stationary. The inner boundary of the PVS coincides with the outer surface of the arterial wall, with radius ri(t), and its motion is characterized as where b is the wall displacement amplitude, h is the mean (outer) arterial radius, c is the peristaltic wave speed, and λ is the wavelength (Wang and Olbricht 2011). To account for both the deformation of the fluid pathway and the fluid flow within the PVS, we adopt the modified version of the poroelastic model described in Sect. 3. Since the peristaltic wave moves at a constant velocity, an observer traveling with the wave can be treated as inertial. Therefore, we reformulate the steady-state poroelastic model by representing it in a moving reference frame. As illustrated in Fig. 7a, we examine the deformation of the tissue around a blood vessel. Initially, the material fills a region Ω with a nominal internal radius h and an outer radius Ro. To simplify the model, the entire region Ω is treated as a homogeneous poroelastic medium. Depending on the specific value of Ro and the variation in porosity across the region, the domain Ω can be interpreted in different ways: as the PVS, as the tissue around the vessel without a clearly defined PVS, or as the combined PVS and surrounding tissue. The region Ωt (Fig. 7b) in the cylindrical coordinate system (r,θ,z) represents the deformed poroelastic domain after the peristaltic wave has propagated in the positive z-direction. A point x=rer(θ)+zez in Ωt satisfies the conditions ri(t)