betaFeedback

WorksPhysics and Astronomy

Inertial flows over the Koch leaf

Abstract

The main hypothesis of this paper is that inertial effects known from porous-media flow can be observed even in a simple system consisting of a single channel. The strength of inertia is controlled by the Reynolds number, ReRe. We investigated fluid flow through a channel with a cavity in the shape of a Koch leaf. The Koch leaf of level-6 was chosen because it combines regularity, repeatability, and representativeness with a complex multiscale structure and non-trivial hydrodynamic interactions. We analyzed nonlinear effects through the so-called friction factor and show that its deviation from Darcy's law is very subtle. We further examined the influence of inertia on the flow structure, including the participation number ππ and the backward-flow fraction ρ−ρ^{-}. In contrast to disordered porous media, the participation number decreases with ReRe, indicating that inertia concentrates kinetic energy in the main through-flow, while recirculation (backward flow) increasingly occupies the Koch leaf cavity. The tortuosity varies non-monotonically with ReRe, the apparent permeability decreases, and in the inertial regime the two are related by a power law with an exponent comparable to that found for natural rock. At the lowest and highest Reynolds numbers, the local kinetic energy followed a power-law distribution, P(ε)∝ε−mP(\varepsilon)\propto\varepsilon^{-m}, over several orders of magnitude with the same exponent, m≈0.83m\approx0.83; inertia modified only the tails of the distribution. Our research can be regarded as an investigation of transport in an extremely porous system with porosity φ≈1φ\approx 1.

Introduction

Fluids flowing through porous media self-organize into preferential flow channels, whose structure is set by the interaction between the moving fluid and the geometry of the fluid–solid interface [1, 2]. When the driving force is large, the flow along these channels may gain enough momentum for inertial effects to appear, even though the flow remains laminar. Understanding how such channels form and how they are modified by inertia is important for subsurface energy storage and CO2_2 sequestration [3], groundwater abstraction [4], and engineered porous systems such as microfluidic devices [5], porous-medium combustors [6], filters [7], and catalytic and thermochemical reactors [8].

At low flow velocities, fluid transport through a porous medium is described by Darcy’s law, which establishes a linear relationship between the applied pressure gradient and the flow velocity [9, 10],

−∇p=μkU,(1)-\nabla p = \frac{\mu}{k}\mathbf{U}, \tag*{(1)}

where kk is the permeability and U\mathbf{U} is the superficial velocity. As the velocity increases, inertial effects become significant and deviations from linearity emerge. These are commonly captured by the Forchheimer equation, which introduces a second-order correction [11, 12],

−∇p=μkU+βρ∣U∣U,(2)-\nabla p = \frac{\mu}{k}\mathbf{U} + \beta\rho\lvert\mathbf{U}\rvert\mathbf{U}, \tag*{(2)}

where β\beta is the non-Darcy (inertial) coefficient. The transition between the two regimes is governed by the Reynolds number and involves a number of pore-scale phenomena, including local flow separation, vortex formation, and enhanced drag [13, 14, 15]. Its onset and strength depend sensitively on the properties of the medium, in particular on porosity and heterogeneity [16, 17, 18]. Pore-scale simulations of bead packs and natural rocks have shown that heterogeneous samples can depart from Darcy flow at Reynolds numbers two to three orders of magnitude lower than homogeneous ones of similar porosity [19], and heterogeneity also modifies how inertia affects solute transport [20]. Deviations from Darcy’s law have even been reported at very low Reynolds numbers, in the so-called pre-Darcy regime [21, 22].

A further difficulty is that the onset of inertial effects does not necessarily coincide with a pronounced change in the global pressure–flow relation. Changes in the internal organization of the flow can appear considerably earlier and may remain hidden if only integral quantities, such as the friction factor or the apparent permeability, are considered [18, 23]. Pore-scale measures are therefore needed to resolve the transition. The participation number π\pi quantifies how kinetic energy is distributed over the pore space [2]; the hydraulic tortuosity τ\tau measures how strongly the flow paths deviate from the mean flow direction [24, 25]; and the backward-flow fraction and the formation of vortices characterize recirculation [18, 23]. In disordered media, the transition toward inertia-influenced flow is accompanied by an increase in π\pi, as kinetic energy is redistributed from preferential channels into the surrounding pore space [2, 18], and by a non-monotonic change in tortuosity [18, 23].

The statistics of the local kinetic energy provide complementary information. Andrade et al. [26] solved the Navier–Stokes equations in a two-dimensional percolation cluster and found that the distribution of the local kinetic energy, n(E)∝E−αn(E) \propto E^{-\alpha}, follows a power law over roughly five orders of magnitude, with α=0.90±0.03\alpha= 0.90 \pm0.03 for the full cluster and α=0.64±0.05\alpha= 0.64 \pm0.05 for the flow-carrying backbone alone. This behavior contrasts with the log-normal distribution of local currents in the analogous random resistor network and shows that slowly moving fluid in the “stagnant” zones plays a significant role in transport through porous media. The exponent remained practically constant over two decades of low Reynolds numbers, but decreased to α=0.79±0.03\alpha= 0.79 \pm0.03 at high ReRe, where several vortices and zones of flow separation appeared [26]. The energy distribution therefore links the geometry of the slow, near-wall fluid to the onset of inertia, but it has so far been studied mainly in disordered structures.

Most of these observations come from idealized granular packings or from disordered and natural porous media, in which several geometrical properties—porosity, pore-size distribution, connectivity, tortuosity, and surface roughness—act simultaneously. Consequently, it is difficult to distinguish changes produced directly by fluid inertia from those imposed by structural disorder. In a recent study of low-porosity natural rock samples, we showed that the transition to inertial flow occurs through a distinct mechanism, visible as a change in tortuosity and the early formation of vortical structures, while the friction factor shows no clear departure from Darcy’s law [23]. Rough and self-affine boundaries are likewise known to influence hydraulic transport and permeability [27, 28]. These results raise the question of which features of inertial porous-media flow originate from disorder and which can emerge solely from the nonlinear interaction between fluid momentum and a complex solid boundary.

Deterministic fractal geometries provide a way to address this question in a controlled setting, because they contain geometrical features over several length scales while remaining exactly reproducible. The Koch leaf of any level is a particularly simple example of such a multiscale boundary: at every construction level, smaller geometrical features are introduced according to the same deterministic rule. Higher-order Koch leaf geometries (order n≥3n \ge3) therefore represent an intermediate level of geometric complexity between simple obstructions and the highly irregular boundaries encountered in natural porous materials, and have already been investigated in transport-related studies. For instance, Adrover showed that convection-dominated dispersion in channels bounded by Koch leaf and Koch snowflakes exhibits a scaling different from that of smooth boundaries [29]. However, comparatively little attention has been devoted to how an increasing contribution of fluid inertia progressively alters the flow structure around an isolated Koch-type obstruction.

In this work, we investigate steady incompressible flow through a two-dimensional channel containing Koch-leaf

Geometry of the Koch leaf of order 6

Figure 1. Geometry of the Koch leaf of order 6, where the porosity of the channel together with Koch leaf obstruction is φ≈1\varphi\approx1.

of order 6 geometry. In contrast to the low-porosity rocks studied in Ref. [23], the present system represents the opposite limit: it contains a single deterministic obstruction and has a very high fluid-volume fraction, φ≈1\varphi\approx1. The Reynolds number is systematically increased while the geometry remains unchanged. We examine whether signatures characteristic of inertial porous-media flows emerge in this simplified system by analyzing the macroscopic friction response together with the tortuosity τ\tau, the participation number π\pi, the backward-flow fraction ρ−\rho^{-}, the apparent permeability, and the probability distribution of the local kinetic energy. The simulations are complemented by a semi-analytical solution of the Stokes problem, built from Moffatt corner modes of the self-similar Koch geometry, which provides an independent reference for the viscous limit. Our central hypothesis is that measurable inertia-driven changes in the flow structure can develop even in a highly open and geometrically deterministic system, while their signature in conventional macroscopic measures remains comparatively subtle.

Materials and Methods

To investigate the emergence of inertial effects in a geometrically well-defined system, we considered a two-dimensional flow domain incorporating a Koch leaf-shaped boundary connected to the channel walls. The height of the Koch leaf was fixed at L=0.3 mL = 0.3\,\mathrm{m} and was used as the characteristic length scale throughout the analysis.

The fluid was assumed to be Newtonian and incompressible, and the steady-state flow was described by the incompressible Navier–Stokes equations,

∇⋅u=0,(3)\nabla\cdot\mathbf{u} = 0, \tag*{(3)}

and

ρ(u⋅∇)u=−∇p+μ∇2u,(4)\rho(\mathbf{u} \cdot\nabla)\mathbf{u} = -\nabla p + \mu\nabla^{2}\mathbf{u}, \tag*{(4)}

where u\mathbf{u} is the velocity field, pp is the pressure, ρ\rho is the fluid density, and μ\mu is the dynamic viscosity. The flow was imposed at the inlet by prescribing a fully developed parabolic velocity profile. The inlet extended from y=−0.30 my=-0.30\ \mathrm{m} to y=0y=0, with its centerline located at y=−0.15 my=-0.15\ \mathrm{m}. The streamwise component of the inlet velocity was prescribed as

ux(y)=Umax⁡[1−(y+0.150.15)2],(5)u_x(y)=U_{\max}\left[1-\left(\frac{y+0.15}{0.15}\right)^2\right], \tag*{(5)}

with

uy=0.(6)u_y=0. \tag*{(6)}

Here, Umax⁡U_{\max} denotes the maximum inlet velocity, attained at the centerline of the inlet. The velocity decreases parabolically toward the inlet edges and becomes zero at y=−0.30 my=-0.30\ \mathrm{m} and y=0y=0. The value of Umax⁡U_{\max} was varied between simulations to obtain different flow conditions and thereby systematically increase the relative contribution of inertial effects.

No-slip boundary conditions,

u=0,(7)\mathbf{u}=0, \tag*{(7)}

were imposed on the Koch surface and all other solid boundaries of the computational domain. At these walls and at the inlet, a zero normal gradient was imposed for the pressure, ∂p/∂n=0\partial p/\partial n=0. At the outlet, a zero normal gradient was imposed for the velocity, and the pressure was fixed to a reference value,

∂u∂n=0,p=0,(8)\frac{\partial\mathbf{u}}{\partial n}=0,\qquad p=0, \tag*{(8)}

where nn denotes the outward normal direction. The governing equations were solved using the finite-volume method implemented in OpenFOAM [30]. Steady-state solutions were obtained using the simpleFoam solver, which employs the Semi-Implicit Method for Pressure-Linked Equations (SIMPLE) algorithm for pressure–velocity coupling [31]. Numerical convergence was assessed by monitoring the residuals of the pressure and velocity equations. The calculations were continued until all residuals decreased below 10−610^{-6}.

The strength of inertial effects was characterized by the Reynolds number,

Re=ULν,(9)Re=\frac{UL}{\nu}, \tag*{(9)}

where L=0.3 mL=0.3\ \mathrm{m} is the height of the Koch leaf, UU is the mean inlet velocity, and ν=μ/ρ\nu=\mu/\rho is the kinematic viscosity. For the prescribed parabolic inlet profile, the mean inlet velocity is related to the maximum velocity by

U=23Umax⁡.(10)U=\frac{2}{3}U_{\max}. \tag*{(10)}

Consequently, the Reynolds number can be written directly in terms of the imposed maximum inlet velocity as

Re=2Umax⁡L3ν.(11)Re=\frac{2U_{\max}L}{3\nu}. \tag*{(11)}

The Koch geometry, characteristic length, and fluid properties were kept unchanged throughout the simulations, while Umax⁡U_{\max} was systematically varied to cover the investigated Reynolds-number range.

The computational geometry was discretized using the cartesian2DMesh utility of the cfMesh library [32], and the resulting mesh was imported into OpenFOAM v. 2006. Local mesh refinement was applied in the vicinity of the Koch boundary in order to accurately resolve the velocity gradients close to the solid surface.

To obtain a sufficient number of sampling points to generate the probability density functions of the kinetic energy in the low-velocity areas, we oversampled the OpenFOAM solutions using RBF-FD interpolation [33]. In particular, the mesh elements with more than 3 sides were first divided into triangles. Subdivision of the resulting triangles into smaller ones was decided based on the magnitude of the kinetic energy in the triangle’s centroid – regions of lower kinetic energy were subdivided into more triangles. The sampling points were then placed at the centroids of the subdivided triangles. We used cubic radial basis functions augmented with second order monomials as the interpolation basis.

To provide an independent reference for the viscous limit, we also constructed a semi-analytical solution of the flow for Re→0Re\to0. In this limit, the inertial term in Eq. (4) vanishes and the problem reduces to the Stokes equations,

∇⋅u=0,μ∇2u=∇p,(12)\nabla\cdot\mathbf{u}=0,\qquad\mu\nabla^2\mathbf{u}=\nabla p, \tag*{(12)}

with the boundary conditions of Eqs. (5)–(8). Solutions of the Stokes problem have a stream function of the Goursat form ψ=Re⁡{zf(z)‾+g(z)}\psi=\operatorname{Re}\{z\overline{f(z)}+g(z)\}, with z=x+iyz=x+iy and ff, gg analytic [34], from which the velocity, vorticity, and pressure follow. Near a corner of opening angle 2α2\alpha, the admissible functions are powers zλz^\lambda, and the no-slip condition fixes λ\lambda through

sin⁡[2(λ−1)α]±(λ−1)sin⁡2α=0(13)\sin[2(\lambda-1)\alpha]\pm(\lambda-1)\sin2\alpha=0 \tag*{(13)}

[35]. For the 60∘60^\circ apices of the Koch triangles the exponents are complex (λ1≈5.06+1.95i\lambda_1\approx5.06+1.95i), which gives the infinite sequence of Moffatt eddies, whereas the 240∘240^\circ corners at the ends of the openings give real exponents (λ=1.616,2.149,…\lambda=1.616,2.149,\ldots).

In each Koch triangle Tk,lT_{k,l}, the velocity is written as

u=∑nak,l,nuk,l,n60∘+∑nbk,l,nuk,l,n240∘+∑nck,l,nuk,l,nreg,(14)\mathbf{u}=\sum_n a_{k,l,n}\mathbf{u}_{k,l,n}^{60^\circ}+\sum_n b_{k,l,n}\mathbf{u}_{k,l,n}^{240^\circ}+\sum_n c_{k,l,n}\mathbf{u}_{k,l,n}^{\mathrm{reg}}, \tag*{(14)}

with a separate set of coefficients in every triangle. Here, uk,l,n60∘\mathbf{u}_{k,l,n}^{60^\circ} are the Moffatt modes of the apex of the triangle and uk,l,n240∘\mathbf{u}_{k,l,n}^{240^\circ} the modes of the 240∘240^\circ corners at the ends of its opening and of the openings on its sides; each of them satisfies the no-slip condition on the two walls of its own corner. The last sum is the regular part, of Goursat form with ff and gg polynomial in z−z0z-z_0, where z0z_0 is the centroid of the triangle. Polynomial Goursat functions are complete: every biharmonic function on a simply connected domain can be approximated by them [36]. Polynomials alone, however, converge slowly near the corners, where ff and gg are singular; the corner modes remove these singularities, in the same way as in the lightning Stokes solver [37] and in the method of particular solutions [38]. Unlike in Ref. [37], the corner terms used here are the exact Moffatt modes of each triangle of the self-similar geometry.

In the channel, the solution is decomposed as

u=uP+uS,(15)\mathbf{u}=\mathbf{u}_{P}+\mathbf{u}_{S}, \tag*{(15)}

where uP\mathbf{u}_{P} is the Poiseuille profile of Eq. (5) extended to the whole channel. It satisfies the Stokes equations, the no-slip condition on both channel walls, and the inlet and outlet conditions, and it vanishes on the openings to the cavity. The correction uS\mathbf{u}_{S} is the Stokes flow that equals the cavity velocity of Eq. (14) on the openings, satisfies the outlet condition, and vanishes on the remaining boundaries. Adding this correction preserves all boundary conditions of uP\mathbf{u}_{P} and makes the velocity continuous across the openings. Since every term in Eqs. (14) and (15) solves the Stokes problem, the coefficients are determined by a least-squares fit to the no-slip condition on the walls and to the continuity of vorticity and pressure across the openings. The solution was computed for the same sixth-generation Koch leaf and the same 2×0.3 m2 \times0.3\ \mathrm{m} channel as the simulations. From it, we evaluated τ\tau, π\pi, ρ−\rho^{-}, and Kapp\mathcal{K}_{\mathrm{app}} with the same definitions as for the simulated fields [Eqs. (20), (22), (25), and (27) in Sec. III]; these Stokes-limit values are shown as solid horizontal lines in Figure 4–Figure 7.

Tortuosity versus Reynolds number

Figure 4. Tortuosity versus ReRe. The solid horizontal line marks the Stokes-limit (Re→0Re \to0) value τS=1.00316\tau_S = 1.00316 obtained from the semi-analytical solution of Eq. (12) (§ 1.1).

Plot of $K_{\mathrm{app}}$ versus $Re$

Figure 7. KappK_{\mathrm{app}} versus ReRe. The solid horizontal line marks the Stokes-limit value KS=8.019×10−3 m2K_{\mathrm{S}} = 8.019 \times10^{-3}\ \mathrm{m^2}.

Results and Discussion

The velocity-magnitude fields at two representative Reynolds numbers are shown in Figure 2. At the lowest Reynolds number, Re=6.67×10−5Re=6.67 \times10^{-5} [Figure 2(a)], the flow is strongly dominated by viscous effects, and the velocity distribution closely follows the multiscale geometry of the Koch boundary. Broad regions of reduced velocity develop in the vicinity of the solid surface, while the velocity varies smoothly away from the obstruction. As the Reynolds number increases to Re≈85.09Re \approx85.09 [Figure 2(b)], the influence of fluid inertia becomes more pronounced. The higher fluid momentum leads to a stronger localization of the velocity gradients near the Koch surface and to a redistribution of the flow around its geometrical features. In particular, the low-velocity regions become more confined to the vicinity of the boundary, while the surrounding flow retains a larger momentum. The comparison therefore indicates a redistribution of the flow with increasing ReRe, motivating a more quantitative analysis of the inertial effects in the following results.

Velocity-magnitude fields and streamlines at two representative Reynolds numbers

Figure 2. Velocity-magnitude ∣U∣|U| [m/s\mathrm{m/s}] fields at the lowest (a) Re=6.67×10−5Re=6.67 \times10^{-5} and highest (b) Re≈85.09Re\approx85.09 Reynolds numbers considered in the simulations. The dashed boxes (c) and (d) indicate the regions enlarged in the lower panels, where the corresponding streamlines are shown. The logarithmic color scale highlights low-velocity regions and illustrates the increasing spatial redistribution of the flow with increasing inertia.

A closer view of the streamlines, shown in the lower panels of Figure 2, reveals a clear change in the internal flow structure with increasing Reynolds number. At the lowest Reynolds number, several small recirculation pock-

ets appear in the vicinity of the finer geometrical features. At Re≈85.09Re \approx85.09, the increased fluid momentum reduces the ability of the flow to follow each local variation of the boundary. Consequently, the streamlines in the main channel become straighter, reflecting the increasing importance of advective inertia relative to viscous diffusion. At the same time, the flow inside the Koch cavity changes qualitatively. The main stream separates from the boundary at the entrance of the cavity, and the small, isolated recirculation pockets observed at low ReRe merge into a single large primary vortex that occupies most of the cavity. This primary vortex is accompanied by smaller secondary vortices trapped within the finer recesses of the Koch boundary. The comparison thus reveals two simultaneous effects of increasing inertia: a straightening of the main through-flow due to momentum preservation and an enhancement of separated recirculating flow inside the Koch cavity. Thus, although the external streamlines become less sensitive to the small-scale boundary geometry, the internal flow structure becomes increasingly dominated by inertia-induced vortical motion.

To assess whether the changes observed in the internal flow structure are also reflected in the macroscopic flow resistance, we calculated the friction factor [23] as

f=ΔpL⟨∣u∣⟩Ω2,(16)f = \frac{\Delta p L}{\langle|\mathbf{u}| \rangle_{\Omega}^{2}}, \tag*{(16)}

where Δp\Delta p is the pressure drop across the computational domain, and ⟨∣u∣⟩Ω\langle|\mathbf{u}| \rangle_{\Omega} is the area-weighted mean velocity magnitude,

⟨∣u∣⟩Ω=∑i=1N∣ui∣Ai∑i=1NAi.(17)\langle|\mathbf{u}| \rangle_{\Omega} = \frac{\sum_{i=1}^{N} |\mathbf{u}_i| A_i}{\sum_{i=1}^{N} A_i}. \tag*{(17)}

Here, Ω\Omega denotes the entire two-dimensional fluid domain, comprising the main channel and the Koch-leaf cavity. The quantity NN is the total number of computational cells within Ω\Omega, and AiA_i is the area of the ii-th cell. In the viscous-dominated regime, the friction factor follows the Darcy relation

fD=bRe,(18)f_D = \frac{b}{Re}, \tag*{(18)}

where bb is obtained from a fit to the low-ReRe data. As shown in Figure 3, the simulated friction factor remains very close to the Darcy fit over most of the investigated Reynolds-number range. Therefore, the influence of inertia is not immediately evident from the friction-factor curve alone, similar to the behavior reported for the geological porous medium [23].

Friction factor versus Reynolds number, showing Darcy and inertial regimes

Figure 3. Friction factor versus ReRe. The dashed horizontal line in the inset at δf=0\delta f = 0 indicates no deviation from the Darcy prediction. The vertical dashed line in the main plot and inset serves as a guide, separating the Darcy regime from the inertial regime.

To resolve the small departure from the Darcy relation more clearly, we calculated the relative deviation of the simulated friction factor from the Darcy prediction as

δf=f−fDfD,(19)\delta f = \frac{f-f_D}{f_D}, \tag*{(19)}

where ff is the friction factor obtained from the simulation and fD=b/Ref_D = b/Re is the value predicted by the Darcy fit. At low ReRe, the relative deviation remains close to zero, confirming that the flow follows the Darcy scaling. With increasing ReRe, the deviation gradually increases and becomes more pronounced at higher Reynolds numbers, reaching δf≈0.014\delta f \approx0.014 (i.e., 1.4%) at the largest ReRe considered, thereby indicating the growing contribution of inertial effects. Nevertheless, the deviation remains small compared with the changes observed in the internal flow structure.

To quantify the changes in the flow structure observed in Figure 2, we calculated the tortuosity τ\tau, which we defined as

τ=⟨∣u∣⟩Ω⟨ux⟩Ω,(20)\tau= \frac{\langle|\mathbf{u}| \rangle_{\Omega}}{\langle u_x \rangle_{\Omega}}, \tag*{(20)}

where ⟨∣u∣⟩\langle|\mathbf{u}| \rangle is the area-weighted average velocity magnitude and ⟨ux⟩\langle u_x \rangle is the area-weighted streamwise velocity component [4, 25]. For the present two-dimensional domain, Eq. (20) can equivalently be written as

τ=∑i=1N∣ui∣Ai∑i=1Nux,iAi,(21)\tau= \frac{\sum_{i=1}^{N} |\mathbf{u}_i| A_i}{\sum_{i=1}^{N} u_{x,i} A_i}, \tag*{(21)}

where AiA_i is the area of the iith computational cell [23]. A value of τ\tau close to unity corresponds to a predominantly streamwise flow, whereas larger values indicate stronger deviations from the main flow direction and an increasing contribution of recirculating motion.

The variation of τ\tau with Reynolds number is shown in Figure 4. At very low ReRe, the tortuosity remains nearly constant at τ≈1.0032\tau\approx1.0032, in agreement with the Stokes-limit value τS=1.00316\tau_S = 1.00316 obtained from the semi-analytical solution (§ 1.1; solid horizontal line in Figure 4). The simulations follow this value up to Re≈1Re \approx1, consistent with the onset of deviations from the Darcy relation in Figure 3. This is consistent with the velocity fields and streamlines shown previously: viscous effects dominate, and the structure of the flow is mainly imposed by the Koch geometry. Although the streamlines follow the local variations of the boundary and small recirculation pockets are already present, increasing the velocity within this range does not significantly modify the overall flow paths.

With increasing ReRe, τ\tau begins to decrease and reaches a minimum of τmin⁡≈1.0025\tau_{\min} \approx1.0025 at Re≈14Re \approx14. This decrease is consistent with the streamline patterns in the lower panels of Figure 2, where the main through-flow becomes progressively straighter. As inertia increases, the fluid tends to preserve its streamwise momentum and becomes less sensitive to the smaller-scale variations of the Koch boundary. The resulting straightening of the main flow paths reduces the tortuosity. A similar decrease of τ\tau with increasing inertia has previously been observed in disordered and natural porous media [18, 23].

Above the minimum, the trend reverses, and τ\tau increases rapidly with ReRe, reaching τ≈1.0049\tau\approx1.0049 at the largest ReRe considered. This behavior is again consistent with the streamline results: at higher Reynolds numbers, flow separation becomes stronger, and a large recirculation region develops inside the Koch leaf cavity, together with smaller secondary vortices. These recirculating structures increase the contribution of transverse and backward velocities to ∣u∣|\mathbf{u}|, while contributing less to the net streamwise transport, and therefore increase τ\tau. The non-monotonic behavior of tortuosity can thus be understood as the result of two competing inertial mechanisms: straightening of the main through-flow, which decreases τ\tau, and the growth of separated and vortical regions, which increases it. The same trend was reported in Ref. [23], while inertia-induced changes in tortuosity and energy distribution have also been observed in disordered porous systems [2, 18].

The tortuosity result therefore provides a quantitative confirmation of the qualitative changes already visible in the velocity and streamline fields. At intermediate ReRe, inertia primarily straightens the main flow, whereas at larger ReRe, the development of recirculation becomes dominant and causes the rapid increase in τ\tau.

To understand the mechanism behind the changes in tortuosity, we next investigated the spatial distribution of kinetic energy. For this purpose, we calculated the participation number π\pi, which quantifies the degree of localization of kinetic energy in the flow [2, 18, 23]. For the present two-dimensional domain, the area-weighted participation number is defined as

π=(A∑i=1Ndi2Ai)−1(22)\pi= \left(A \sum_{i=1}^{N} d_i^2 A_i\right)^{-1} \tag*{(22)}

where

qi=ei∑j=1NejAj,(23)q_i = \frac{e_i}{\sum_{j=1}^{N} e_j A_j}, \tag*{(23)}

AiA_i is the area of the iith computational cell, A=∑i=1NAiA = \sum_{i=1}^{N} A_i is the total fluid area, and the local kinetic-energy density satisfies

ei∝∣ui∣2.(24)e_i \propto|\mathbf{u}_i|^2. \tag*{(24)}

Since the density is constant, the factor ρ/2\rho/2 in the kinetic-energy density does not affect the value of π\pi. A larger value of π\pi corresponds to a more spatially distributed kinetic-energy field, whereas a decrease in π\pi indicates a stronger localization of kinetic energy.

The variation of π\pi with Reynolds number is shown in Figure 5. At low ReRe, π\pi remains nearly constant at π≈0.6230\pi\approx0.6230, coinciding with the Stokes-limit value πS=0.62303\pi_S = 0.62303 (solid line in Figure 5), indicating that the relative spatial distribution of kinetic energy is essentially unchanged in the viscous-dominated regime. As ReRe increases, however, π\pi decreases progressively, reaching π≈0.6199\pi\approx0.6199 at the largest ReRe considered. This behavior is opposite to that observed in disordered and natural porous media, where increasing ReRe was found to increase π\pi as kinetic energy became distributed over a larger fraction of the pore space [2, 18, 23]. For example, Andrade et al. reported a constant π≈0.37\pi\approx0.37 up to Re≈0.3Re \approx0.3, followed by a gradual increase of π\pi above this transition [2], and a similar rise of π\pi with ReRe was found by us for a disordered medium with porosity φ=0.9\varphi= 0.9 [18]. In the present Koch geometry, increasing inertia therefore produces a stronger localization of kinetic energy rather than a broader spatial distribution.

Plot of participation number versus Reynolds number

Figure 5. π\pi versus ReRe. The solid horizontal line marks the Stokes-limit value πS=0.62303\pi_S = 0.62303.

This behavior is consistent with the velocity and streamline fields shown previously. With increasing ReRe, the main through-flow becomes straighter and less sensitive to the small-scale variations of the Koch leaf boundary, while a dominant recirculation region develops inside the cavity. The flow is therefore increasingly organized into a more energetic main stream and localized recirculating structures. Consequently, a smaller fraction of the domain carries a larger fraction of the total kinetic energy, leading to the observed decrease of π\pi.

The decrease in π\pi also provides further insight into the non-monotonic behavior of τ\tau. In the intermediate Reynolds-number range, the straightening of the main flow reduces τ\tau, while the kinetic energy simultaneously becomes more localized, as indicated by the decrease in π\pi. At larger ReRe, τ\tau starts to increase because the contribution of recirculating motion becomes sufficiently strong, whereas π\pi continues to decrease. Thus, in the present geometry, tortuosity and participation number do not follow the same trend. This differs from the natural porous-medium case, where an increase in π\pi accompanied the high-ReRe increase in tortuosity [23]. To examine the contribution of recirculating motion more directly, we calculated the backward-flow fraction ρ−\rho^{-}, which measures the fraction of the fluid domain in which the streamwise velocity is negative [18, 23]. For the present two-dimensional domain, the area-weighted backward-flow fraction is defined as

ρ−=1A∑i=1Nf(ux,i)Ai,(25)\rho^{-} = \frac{1}{A}\sum_{i=1}^{N} f(u_{x,i})A_i, \tag*{(25)}

where AA is the total fluid area, AiA_i is the area of the iith cell, and ux,iu_{x,i} is the local streamwise velocity component. The indicator function f(ux)f(u_x) is defined as

f(ux)={1,ux<0,0,otherwise.(26)f(u_x) = \begin{cases} 1, & u_x < 0, \\ 0, & \text{otherwise}. \end{cases} \tag*{(26)}

Thus, ρ−=0\rho^{-} = 0 corresponds to a flow with no backward streamwise motion, whereas an increasing ρ−\rho^{-} indicates that a larger fraction of the fluid domain is occupied by reverse flow.

The variation of ρ−\rho^{-} with Reynolds number is shown in Figure 6. At very low ReRe, ρ−\rho^{-} remains nearly constant at approximately 0.048, in agreement with the Stokes-limit value ρS=0.0477\rho_S = 0.0477 (solid line in Figure 6), that is 56% of the Koch-leaf cavity area. This non-zero value is consistent with the streamline field shown in Figure 2, where small recirculation pockets are already present around the finer geometrical features of the Koch leaf boundary, even in the viscous-dominated regime [35].

Plot of $\rho^{-}$ versus $Re$

Figure 6. ρ−\rho^{-} versus ReRe. The solid horizontal line marks the Stokes-limit value ρS=0.0477\rho_S = 0.0477.

As ReRe increases, ρ−\rho^{-} begins to rise, indicating a progressive increase in the area occupied by backward flow. The increase becomes much stronger at higher Reynolds numbers, and ρ−\rho^{-} reaches approximately 0.077, that is 89.5% of the Koch-leaf cavity, at the largest ReRe considered. This trend is directly consistent with the streamline results, where the initially small and localized recirculation regions develop into a dominant large-scale vortex inside the Koch cavity. The growth of ρ−\rho^{-} therefore provides a quantitative measure of the increasing importance of inertia-induced flow separation and recirculation.

The relationship between ρ−\rho^{-} and τ\tau also helps to clarify the two competing mechanisms identified above. In the intermediate Reynolds-number range, ρ−\rho^{-} already increases while τ\tau decreases. This indicates that the growth of recirculation occurs simultaneously with a stronger straightening of the main through-flow. At this stage, the straightening effect dominates the global tortuosity. At larger ReRe, the growth of the backward-flow region becomes sufficiently rapid that τ\tau reverses its trend and begins to increase. The simultaneous increase of ρ−\rho^{-} and τ\tau at high ReRe therefore shows that recirculating motion becomes increasingly important in determining the overall flow structure.

Together with the decrease of π\pi, the growth of ρ−\rho^{-} shows that the recirculating region inside the Koch leaf cavity expands while carrying only a small share of the kinetic energy, which becomes increasingly concentrated in the main through-flow.

Finally, we calculated the apparent permeability κapp\kappa_{\mathrm{app}} from the Darcy relation as

κapp=νULpΔp,(27)\kappa_{\mathrm{app}} = \frac{\nu U L_p}{\Delta p}, \tag*{(27)}

where ν\nu is the kinematic viscosity, UU is the mean inlet velocity, and LpL_p is the streamwise length over which the pressure drop Δp\Delta p is measured. The variation of κapp\kappa_{\mathrm{app}} with Reynolds number is shown in Figure 7. At low ReRe, κapp\kappa_{\mathrm{app}} remains nearly constant at κapp≈8.02×10−3 m2\kappa_{\mathrm{app}} \approx8.02 \times10^{-3}\,\mathrm{m}^2, matching the Stokes-limit value κS=8.019×10−3 m2\kappa_S = 8.019 \times10^{-3}\,\mathrm{m}^2 (solid line in Figure 7), indicating that the pressure drop scales linearly with the flow velocity and that the system is dominated by viscous effects. With increasing ReRe, κapp\kappa_{\mathrm{app}} gradually decreases, reaching ≈7.89×10−3 m2\approx7.89 \times10^{-3}\ \mathrm{m^2} at the largest ReRe (a reduction of about 1.6%), showing the departure from the Darcy regime and the increasing contribution of inertial resistance.

The decrease becomes more pronounced at higher Reynolds numbers and is consistent with the simultaneous growth of recirculation and tortuosity observed previously. Thus, although the geometrical permeability is unchanged, the apparent permeability decreases as inertia modifies the internal flow structure. A similar decrease of apparent permeability with increasing ReRe was reported for the low-porosity geological porous medium [23].

To relate the changes in flow structure to the apparent permeability response, we examined the correlation between the tortuosity τ\tau and the apparent permeability KappK_{\mathrm{app}} (Figure 8). Both quantities vary only weakly over the investigated range of Reynolds numbers, consistent with the single, deterministic, multiscale Koch leaf boundary

Plot of $\tau$ versus $K_{\mathrm{app}}$

Figure 8. τ\tau versus KappK_{\mathrm{app}}. The five highest-ReRe points in the inertial regime are used for fitting, which represents the power law τ∝Kapp−b\tau\propto K_{\mathrm{app}}^{-b}, with b≈0.24b \approx0.24.

probed here; nevertheless, their joint variation follows the same qualitative trend reported for a natural, low-porosity rock sample [23]: as inertial effects strengthen, the apparent permeability decreases while the tortuosity increases, so that τ\tau and KappK_{\mathrm{app}} remain inversely correlated throughout the inertial regime.

In the inertial range (i.e., beyond the minimum of τ(Re)\tau(Re) discussed above), this correlation can be described by a power law,

τ∝Kapp−b.(28)\tau\propto K_{\mathrm{app}}^{-b}. \tag*{(28)}

A fit with a free exponent to the five highest-ReRe points in Figure 8 gives b≈0.24b \approx0.24 (R2=0.98R^2 = 0.98). Over this narrow range, the same points are equally well reproduced by a linear form, τ=1.247−30.8Kapp\tau= 1.247 - 30.8 K_{\mathrm{app}} (with KappK_{\mathrm{app}} in m2\mathrm{m^2}, R2=0.98R^2 = 0.98); the power-law and linear descriptions are locally indistinguishable because τ\tau and KappK_{\mathrm{app}} vary by only 0.15% and 0.6%, respectively, across this inertial-regime data range. Because the fit is based on only five points spanning such a narrow range, the power-law form and the value of bb should be regarded as indicative rather than conclusive; establishing the functional form of the τ\tau–KappK_{\mathrm{app}} relation unambiguously would require additional simulations at higher ReRe. The exponent obtained for the Koch-leaf geometry is nevertheless comparable to that of the natural rock sample [23], even though the relative variation of τ\tau and KappK_{\mathrm{app}} in the present system is roughly two orders of magnitude smaller (below 2% here versus tens of percent in Ref. [23]). We attribute this small variation to the single, isolated Koch leaf boundary probed here: inertia-driven flow redistribution only mildly affects the streamwise resistance, unlike in a low-porosity medium. That a comparable exponent appears despite this difference in scale suggests that the local form of the τ\tau–KappK_{\mathrm{app}} coupling is set by the inertial flow itself rather than by the degree of pore-space confinement.

Taken together with the non-monotonic behavior of τ(Re)\tau(Re) (Figure 4) and the monotonic decrease of Kapp(Re)K_{\mathrm{app}}(Re) (Figure 7), this result indicates that the coupling between tortuosity and apparent permeability identified for disordered, natural porous media persists even in a fully deterministic, highly open geometry such as the Koch leaf. It is therefore not a consequence of pore-scale disorder but of the more general inertia-induced change in the flow structure—the localization of kinetic energy and the growth of recirculating, backward-flow regions—that accompanies the onset of inertial effects.

To complement the participation number, which measures how kinetic energy is spread in space, we examined the probability distribution function of the local kinetic energy, ϵ=u2+v2\epsilon= u^2 + v^2, at a representative low, Re=6.67×10−5Re = 6.67 \times10^{-5}, and high, Re≈85.09Re \approx85.09, Reynolds number (Figure 9). At both Reynolds numbers, P(ϵ)P(\epsilon) follows a power law over several decades,

PDF of the local kinetic energy distributions for low- and high-Reynolds-number flow

Figure 9. PDF of the local kinetic energy for low- (Re=6.67×10−5Re = 6.67 \times10^{-5} (top)) and high- (Re≈85.09Re \approx85.09 (bottom)) Reynolds numbers. The solid line ∝ε−m\propto\varepsilon^{-m}, is shifted downward by a factor of 10 as a guide to the eye.

P(ϵ)∝ϵ−m,(29)P(\epsilon) \propto\epsilon^{-m}, \tag*{(29)}

with m=0.84m = 0.84 at low ReRe and m=0.83m = 0.83 at high ReRe. Because ϵ∝U2\epsilon\propto U^2, the energies at the two Reynolds numbers differ by about twelve orders of magnitude: the mean energy ⟨ε⟩\langle\varepsilon\rangle increases by a factor of 1.6×10121.6 \times10^{12}, close to (Rehigh/Relow)2(Re_{\mathrm{high}}/Re_{\mathrm{low}})^2. The fits were therefore performed over the same range of normalized energy, ε/⟨ε⟩≈10−16\varepsilon/\langle\varepsilon\rangle\approx10^{-16}–10−1010^{-10}, that is, over the same part of each distribution. Within this range, the exponent is unaffected by inertia. When rescaled by ⟨ε⟩\langle\varepsilon\rangle, the two distributions collapse onto each other over about twenty decades and differ only at their ends: at the lowest energies, ε/⟨ε⟩≲10−21\varepsilon/\langle\varepsilon\rangle\lesssim10^{-21}, the high-ReRe distribution lies higher, indicating a larger fraction of nearly stagnant fluid, while at the highest energies, ε/⟨ε⟩≈10−6\varepsilon/\langle\varepsilon\rangle\approx10^{-6}–11, it is slightly shallower (local slope ≈0.70\approx0.70 compared with ≈0.76\approx0.76 at low ReRe).

A power-law energy distribution of this type was first reported by Andrade et al. [26] for two-dimensional flow through a disordered percolation cluster, where n(E)∝E−αn(E) \propto E^{-\alpha} with α=0.90±0.03\alpha= 0.90 \pm0.03 for the full cluster and α=0.64±0.05\alpha= 0.64 \pm0.05 for the low-carrying backbone alone. The exponent obtained for the Koch leaf, m≈0.85m \approx0.85, is close to the full-cluster value, which includes the stagnant regions, suggesting that in both systems the power

law is set by slowly moving fluid near the solid boundary. In contrast to the percolation cluster, where the exponent decreased to α=0.79±0.03\alpha= 0.79 \pm0.03 at high ReRe as vortices and zones of flow separation appeared [26], mm does not change in the Koch-leaf geometry over the investigated range. We attribute this to the origin of the power-law range: it is formed by the slow fluid in the small-scale recesses of the Koch leaf boundary, where the local flow remains viscous and of Moffatt type [35] even at the highest ReRe considered, whereas inertia acts mainly on the main through-flow and on the large primary vortex in the cavity.

This also clarifies the relation between the energy distribution and the participation number. In Ref. [2] and in the natural porous medium of Ref. [23], π\pi increases with ReRe, reflecting the spatial delocalization of kinetic energy as new flow channels open up. In the Koch-leaf geometry, by contrast, π\pi decreases with ReRe (Figure 5) while mm remains unchanged. The decrease in π\pi therefore does not originate from the power-law part of P(ε)P(\varepsilon), but from its ends. Because π\pi depends on εi2\varepsilon_i^2 [Eq. (22)], it is dominated by the high-energy cells, and the change in the high-energy tail, together with the growth of nearly stagnant fluid at the low-energy end, reflects the increasing contrast between the faster, straighter main through-flow and the expanding recirculation region inside the cavity. Inertia thus concentrates kinetic energy into the main stream without altering the scale-free statistics of the slow, near-wall flow set by the Koch leaf geometry.

Conclusions

We studied inertial effects in steady, laminar flow through a channel with a sixth-order Koch-leaf cavity (φ≈1\varphi\approx1) by means of OpenFOAM simulations covering 6.67×10−5≤Re≤85.096.67 \times10^{-5} \le Re \le85.09. The simulations were validated against a semi-analytical solution for Stokes regime built from Moffatt corner modes, which reproduces the low-ReRe values of π\pi, ρ\rho, ρ−\rho^-, and κapp\kappa_{\mathrm{app}}. The macroscopic response hides the transition: the friction factor deviates from Darcy’s law by only 1.4%, and the apparent permeability decreases by 1.6%. The internal flow, by contrast, changes qualitatively. Recirculation pockets merge into a primary vortex, the backward-flow region grows from 56% to 89.5% of the cavity, and the tortuosity varies non-monotonically, with a minimum at Re≈14Re \approx14, as flow straightening first dominates and recirculation later prevails. In contrast to the disordered porous media, we found that the participation number decreases with ReRe, as inertia concentrates kinetic energy in the main stream, whereas the local kinetic energy follows a power-law distribution with the same exponent, m≈0.83m \approx0.83, at the lowest and highest ReRe, with inertia altering only its tails. In the inertial regime, tortuosity and apparent permeability are related by a power law with an exponent comparable to that of natural rock [23]. These results show that key inertial effects known from disordered porous media also arise in a single deterministic geometry, making fractal geometries a controlled testbed for separating the effects of inertia from those of the structure.

Acknowledgment

Funded by National Science Centre, Poland under the OPUS call in the Weave programme 2021/43/I/ST3/00228. This research was funded in whole or in part by National Science Centre (2021/43/I/ST3/00228). For the purpose of Open Access, the author has applied a CC-BY public copyright licence to any Author Accepted Manuscript (AAM) version arising from this submission. M. Matyka acknowledges the financial support from the Slovenian Research And Innovation Agency (ARIS) research core funding No. P2-0095.

References

  1. [1]J. D. Hyman, Flow channeling in fracture networks: Characterizing the effect of density on preferential flow path formation, Water Resources Research 56, e2020WR027986 (2020).DOI
  2. [2]J. S. J. Andrade, U. M. S. Costa, M. P. Almeida, H. A. Makse, and H. E. Stanley, Inertial effects on fluid flow through disordered porous media, Physical Review Letters 82, 5249 (1999).DOI
  3. [3]S. C. Hovorka, R. H. Treviño, M. J. Nicot, et al., Dynamic processes of CO₂ storage in the field: 1. multiscale and multiphase flow behavior, Water Resources Research 55, 1 (2019).
  4. [4]M. Matyka and M. Madrala, Granica laminarnego przepływu w obsypce żwirowej studni eksploatacyjnej, Przegląd Geologiczny 65, 1334 (2017).
  5. [5]S. C. Cao, J. C. Jung, and M. Radonjic, Application of microfluidic pore models for flow, transport, and reaction in geological porous media: From a single test bed to multifunction real-time analysis tool, Microsystem Technologies 25, 4035 (2019).DOI
  6. [6]A. Banerjee and D. Paul, Developments and applications of porous medium combustion: A recent review, Energy 221, 119868 (2021).
  7. [7]Y. Zhang, Y. Zhou, Y. Zhang, Z. Zhang, J. Liu, J. Zhu, Z. Sun, L. Li, and H. Gu, Optimal design of silver-loaded zeolite filter structure based on porous media model, Separation and Purification Technology 334, 125905 (2024).DOI
  8. [8]F. Wang, Y. Shuai, Z. Wang, Y. Leng, and H. Tan, Thermal and chemical reaction performance analyses of steam methane reforming in porous media solar thermochemical reactor, International Journal of Hydrogen Energy 39, 718 (2014).DOI
  9. [9]S. P. Neuman, Theoretical derivation of darcy’s law, Acta Mechanica 25, 153 (1977).DOI
  10. [10]S. Whitaker, Flow in porous media i: A theoretical derivation of darcy’s law, Transport in Porous Media 1, 3 (1986).DOI
  11. [11]S. Whitaker, The forchheimer equation: A theoretical development, Transport in Porous Media 25, 27 (1996).DOI
  12. [12]R. D. Barree and M. W. Conway, Beyond beta factors: A complete model for darcy, forchheimer, and trans-forchheimer flow in porous media, in Proceedings of the 2004 SPE Annual Technical Conference and Exhibition (Society of Petroleum Engineers, 2004).
  13. [13]M. Fourar, R. Lenormand, M. Karimi-Fard, and R. N. Horne, Inertia effects in high-rate flow through heterogeneous porous media, Transport in Porous Media 60, 353 (2005).DOI
  14. [14]J. Arthur, PIV study of flow through and over porous media at the onset of inertia, Advances in Water Resources 146, 103793 (2020).DOI
  15. [15]M. T. Balhoff and M. F. Wheeler, A predictive pore-scale model for non-darcy flow in porous media, SPE Journal 14, 579 (2009).
  16. [16]Z. Zeng and R. Grigg, A criterion for non-darcy flow in porous media, Transport in Porous Media 63, 57 (2006).DOI
  17. [17]S. Arabi and M. Sahimi, The transition from darcy to nonlinear flow in heterogeneous porous media: I—single-phase flow, Transport in Porous Media 151, 795 (2024).
  18. [18]D. Śnieżek, S. B. Naqvi, and M. Matyka, Inertia onset in disordered porous media flow, Physical Review E 110, 045103 (2024).arxiv.org/abs/2403.00456
  19. [19]B. P. Muljadi, M. J. Blunt, A. Q. Raeini, and B. Bijeljic, The impact of porous media heterogeneity on non-darcy flow behaviour from pore-scale simulation, Advances in Water Resources 95, 329 (2016).DOI
  20. [20]A. Nissan and B. Berkowitz, Inertial effects on flow and transport in heterogeneous porous media, Physical Review Letters 120, 054504 (2018).
  21. [21]P. Basak, Non-darcy flow and its implications to seepage problems, Journal of the Irrigation and Drainage Division 103, 459 (1977).DOI
  22. [22]L. Wang, Y. Li, G. Zhao, N. Chen, and Y. Xu, Experimental investigation of flow characteristics in porous media at low reynolds numbers (re → 0) under different constant hydraulic heads, Water 11, 2317 (2019).
  23. [23]S. B. Naqvi, D. Śnieżek, D. Strzelczyk, M. Mądrala, and M. Matyka, Inertial effects on fluid flow through natural porous media, Water Resources Research 62, e2025WR041274 (2026).DOI
  24. [24]A. Koponen, M. Kataja, and J. Timonen, Permeability and effective porosity of porous media, Physical Review E 56, 3319 (1997).DOI
  25. [25]A. Duda, Z. Koza, and M. Matyka, Hydraulic tortuosity in arbitrary porous media flow, Physical Review E 84, 036319 (2011).DOI
  26. [26]J. S. J. Andrade, M. P. Almeida, J. Mendes Filho, S. Havlin, B. Suki, and H. E. Stanley, Fluid flow through porous media: The role of stagnant zones, Physical Review Letters 79, 3901 (1997).DOI
  27. [27]M. Madadi and M. Sahimi, Lattice boltzmann simulation of fluid flow in fracture networks with rough, self-affine surfaces, Physical Review E 67, 026309 (2003).pubmed.ncbi.nlm.nih.gov/12636802
  28. [28]M. Madadi, C. D. Van Siclen, and M. Sahimi, Fluid flow and conduction in two-dimensional fractures with rough, self-affine surfaces: A comparative study, Journal of Geophysical Research: Solid Earth 108, 2396 (2003).DOI
  29. [29]A. Adrover, Convection-dominated dispersion in channels with fractal cross-section, Physics of Fluids 23, 013603 (2011).DOI
  30. [30]J. H. Jasak, A. Jemcov, Z. Tukovic, et al., OpenFOAM: A c++ library for complex physics simulations, in International Workshop on Coupled Methods in Numerical Dynamics, Vol. 1000 (2007) pp. 1–20.
  31. [31]S. V. Patankar, Numerical Heat Transfer and Fluid Flow (Hemisphere Publishing Corporation, Washington, DC, 1980).
  32. [32]F. Juretic, cfMesh User Guide, Creative Fields, Zagreb (2015).
  33. [33]D. Strzelczyk, M. Rot, G. Kosec, and M. Matyka, Cross-validation of meshless navier–stokes solvers in porous media flows, Scientific Reports 15, 38387 (2025).DOI
  34. [34]E. Goursat, Sur l’équation δδu = 0, Bulletin de la Société Mathématique de France 26, 236 (1898).
  35. [35]H. K. Moffatt, Viscous and resistive eddies near a sharp corner, Journal of Fluid Mechanics 18, 1 (1964).DOI
  36. [36]I. N. Vekua, New Methods for Solving Elliptic Equations (North-Holland, Amsterdam, 1967).
  37. [37]P. D. Brubeck and L. N. Trefethen, Lightning Stokes solver, SIAM Journal on Scientific Computing 44, A1205 (2022).arxiv.org/abs/2107.01572
  38. [38]L. Fox, P. Henrici, and C. Moler, Approximations and bounds for eigenvalues of elliptic operators, SIAM Journal on Numerical Analysis 4, 89 (1967).DOI

Paper details

vlnc:W_m98k9p9g
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents