PaperlayerView exact revision on arXiv

Source-linked AI summary

Uniform Inf-Sup Norm Equivalence and Robust Operator Preconditioning for Stokes Flow in tight domains with Periodic Pillars

Qi Xin, Yan Xie, Chen-song Zhang, Shihua Gong, Jinchao Xu

arXiv:2609.06999v1math.NAcs.NA

TL;DR

Dense periodic pillar arrays make Stokes saddle-point systems difficult because pressure stability deteriorates as pillar density rises. The paper identifies the induced pressure norm through a restriction–extension construction and proves a scale-aware norm equivalence. In two-dimensional experiments, iteration counts remain flat across a 16× range in pillar density, while the restriction construction degenerates near close packing.

  • Problem

    In proportional-hole pillar arrays, the LBB constant decays as Θ(m^-1) and the Schur-complement condition number grows as Θ(m^2), making dense-domain Stokes systems difficult to solve.

    Source: §1 · p. 1

  • Method

    A mesh-free restriction–extension argument characterizes the divergence-induced pressure norm on periodic pillar networks and connects it to a K-functional Riesz map.

    Source: §1 · p. 2, §6 · p. 19

  • Results

    FGMRES iteration counts are flat in pillar density m over a 16× range in two-dimensional experiments.

    Source: §5.7 · p. 18, §6 · p. 19

  • Takeaways & Limitations

    The norm equivalence supplies the pressure norm needed to apply operator preconditioning to perforated geometries.

    Source: §6 · p. 19

    The restriction-operator construction is tailored to vanishing or proportional holes with κ bounded away from 1/2 and degenerates near close packing.

    Source: §3.3 · p. 7

Abstract

from arXiv · show

Many microfluidic and porous-media computations reduce to the same core task: solving a Stokes saddle-point system on a domain perforated by a dense periodic array of pillars, as in deterministic lateral displacement (DLD) particle sorters. After rescaling the device to unit size, the geometry is controlled by a single dimensionless parameter $m$---the number of pillars across the device, equal to the inverse period. In realistic devices $m$ reaches the hundreds or thousands, and as it grows the Stokes inf-sup constant decays like $m^{-1}$, the pressure Schur complement becomes severely ill-conditioned, and standard block solvers slow down in proportion to the pillar density. We remove this bottleneck by identifying the pressure norm that the divergence operator induces on such geometries. For periodic pillar arrays in the proportional-hole regime, we prove that this inf-sup norm is uniformly equivalent to the $L^2+\sigma_\epsilon H^1$ $K$-functional norm at the pore scale $\sigma_\epsilon\asymp\epsilon$, with constants independent of the period, the pillar count $m$, and the mesh size $h$. The equivalence identifies the perforated Stokes problem with a Brinkman problem at a homogenized permeability, and its Riesz map reduces to a pressure-mass inverse plus a scaled stiffness inverse. Combined with operator preconditioning, this yields a block preconditioner built from standard algebraic-multigrid solves whose iteration count is essentially independent of both mesh size and pillar density. Two-dimensional Taylor--Hood experiments confirm the predicted robustness in mesh refinement, pillar density, geometric scale, close packing, and time step.

1. Introduction

Dense periodic pillar arrays create a geometry-driven pressure-stability problem whose severity grows with pillar density. The paper characterizes the induced pressure norm and uses that equivalence to construct a density-robust block preconditioner.

1. Introduction

  • In proportional-hole arrays, the LBB constant decays as Θ(m^-1), the Schur-complement condition number grows as Θ(m^2), and standard block-preconditioner iterations grow with pillar count.The passage attributes this deterioration to the multiply connected pressure network and narrow throats, not mesh resolution.Source: §1 · p. 1
  • The paper proves that the divergence-induced pressure norm is uniformly equivalent to an L2+σ_ϵH1 K-functional norm, with σ_ϵ=ϵ|log(aϵ/ϵ)|^1/2 and σ_ϵ≈ϵ for proportional holes.The equivalence constants are independent of ϵ and the pillar density m.Source: §1 · p. 2
  • A periodic-network restriction operator, its dual pressure extension, and a weighted Nečas estimate establish the lower bound; integration by parts and Poincaré’s inequality give the upper bound.The argument isolates close-packing deterioration in the restriction-operator constant.Source: §1 · p. 2
  • The K-functional Riesz map yields a pressure block combining a mass inverse and a scaled stiffness inverse, producing a block preconditioner independent of perforation count under the stated discrete stability assumptions.The pressure matrices are the mass and Neumann stiffness matrices.Source: §1 · p. 2
  • Two-dimensional Taylor–Hood experiments test the predicted behavior under mesh refinement, changing pillar density and scale, close packing, and time-dependent Stokes flow.The paper relates this pressure-norm construction to channel-specific work [9] and pressure mass–stiffness combinations for channel-dominated flows.Source: §1 · p. 2

2. Mathematical Description of the problem

The paper models a fixed macroscopic domain perforated by periodically repeated pillars, then formulates incompressible Stokes flow on the remaining fluid domain. Its stability is measured by the LBB condition for the velocity–pressure coupling.

2.1. Geometric Configuration of the Periodic Pillars

  • The geometry uses a fixed macroscopic container, a reference cell and obstacle, and two scales: period ϵ and obstacle size aϵ; the proportional-hole regime has aϵ≈ϵ.The period determines the physical cell size, while the scaled obstacle is replicated on a lattice; the stated estimates are uniform as ϵ→0 under separation assumptions.Source: §2.1 · p. 3
  • Interior pillars are removed from the container to form the connected fluid domain, with the pillar density recorded by m proportional to ϵ^-1.Pillars are retained only when they lie strictly inside the container, preserving the no-slip boundary at the lateral walls.Source: §2.1 · p. 3

2.2. The Stokes Problem and the inf-sup Condition

  • The steady model is incompressible Stokes flow with no-slip conditions on the container and pillar surfaces, written in a mixed weak formulation coupling velocity and pressure.The velocity bilinear form is coercive, while well-posedness depends on stability of the velocity–pressure coupling.Source: §2.2 · p. 4
  • The classical LBB condition measures the stability of that coupling and governs well-posedness of the Stokes saddle-point problem on the perforated domain.The pressure space is zero-mean L2, and the velocity space is H1_0 with gradient norm.Source: §2.2 · p. 4

3. Norm Equivalence in the Perforated Domain

The section establishes a period-independent equivalence between the perforated-domain inf-sup norm and the pore-scale K-functional pressure norm, deriving a lower bound via pressure extension and a weighted Nečas estimate and stating a uniform Fortin condition for discrete transport. It extends the equivalence to conforming discrete spaces and backward-Euler Stokes flow, uniformly in the relevant discretization and geometry parameters.

3.1. Pressure-stability norms

  • The inf-sup norm is uniformly equivalent to the L2 + σ_ϵH1 K-functional norm, with constants independent of ϵ and pillar density m.The paper identifies σ_ϵ as the perforation-scale weight and connects the equivalence to the homogenized permeability.Source: §3.1 · p. 5

3.2. Upper bound

  • The upper bound follows from the triangle inequality and the σ_ϵ-weighted Poincaré inequality on the perforated domain.The weighted Poincaré constant is independent of ϵ.Source: §3.1 · p. 5, §3.2 · p. 6

3.3. Restriction operator and the pressure extension

  • A divergence-preserving restriction operator is dualized to construct a pressure extension, with σ_ϵ ≍ ϵ for proportional holes.The cellwise construction applies when holes are small or proportional and the packing fraction stays bounded away from 1/2; it degenerates near close packing, where thin throats and corners carry fluid the inscribed-ball construction misses.Source: §3.3 · pp. 6–7

3.4. Proof of the lower bound

  • The lower bound uses a fixed-domain weighted Nečas inequality to control the pressure extension at the K-functional scale.The estimate is interpreted as the dual inf-sup condition for a Brinkman problem on the filled domain, with homogenized permeability σ_ϵ^2.Source: §3.4 · p. 9

3.5. From the continuous to the discrete norm equivalence

  • A uniform Fortin condition transports the continuous norm equivalence to conforming finite-element subspaces, with constants independent of ϵ and h.For classical inf-sup stable pairs, the condition is obtained by local quasi-interpolation on the perforated pillar-array geometry.Source: §3.5 · p. 10
  • The discrete inf-sup norm is uniformly equivalent to the L2+σ_ϵH1 norm, with constants independent of ϵ, m, and h.The result applies to conforming finite-element pairs satisfying the uniform Fortin-operator assumption.Source: §3.5 · p. 10

3.6. Discretized time-dependent Stokes flow

  • For backward-Euler Stokes flow, the pressure-norm equivalence and preconditioner iteration count are uniform in τ, alongside the established ϵ- and h-uniformity.The combined short scale accounts for both perforation and time-step effects; the equivalence assumes the geometric conditions and an L2-stable restriction operator.Source: §3.6 · pp. 11–12

4. Norm-induced Preconditioners

The discrete inf-sup norm identifies the pressure Schur complement with a K-functional Riesz map, yielding an upper block-triangular preconditioner whose GMRES convergence rate is independent of pillar density and mesh size.

4. Norm-induced Preconditioners

  • The discrete inf-sup Riesz map is equivalent to the K-functional Riesz map, giving an upper block-triangular preconditioner with quality independent of pillar density and mesh size.The construction inherits the equivalence constants used to relate the discrete inf-sup norm to the K-functional norm.Source: §4 · p. 12

4.1. Discrete saddle-point system

  • A conforming, inf-sup stable finite element discretization produces a saddle-point system with a velocity Laplacian block and a discrete divergence operator.The pressure Schur complement is formed from the discrete divergence and velocity block.Source: §4.1 · p. 12

4.2. Schur complement as the inf-sup Riesz map

  • The Schur complement represents the discrete inf-sup norm and is spectrally equivalent to the K-functional Riesz map under the stated uniform bounds.Any tractable spectral equivalent of that Riesz map can therefore serve as a preconditioner for the Schur complement.Source: §4.2 · p. 12

4.3. Closed form of the K-functional Riesz map

  • The closed-form K-functional Riesz map is obtained by inverting a parallel-sum expression, yielding two summands that represent its distinct frequency regimes.The pressure mass term dominates at low frequencies, while the scaled stiffness term captures the high-frequency regime.Source: §4.3 · p. 13

4.4. Upper block-triangular preconditioner

  • The upper block-triangular preconditioner uses an AMG V-cycle for the velocity block and two pressure-space AMG V-cycles for the pressure block.The pressure action combines an L2 inverse and a scaled Laplacian inverse.Source: §4.4 · p. 13

4.5. Spectral bound for the preconditioned operator

  • GMRES convergence is independent of ε, pillar density m, and mesh size h when the velocity and pressure blocks satisfy the stated uniform spectral-equivalence bounds.The preconditioned operator's eigenvalues lie in a bounded set away from zero, with endpoints determined by those bounds.Source: §4.5 · p. 14

5. Numerical Results

Two-dimensional Taylor–Hood experiments test the preconditioner across pillar density, spectral behavior, time steps, tilted DLD arrays, and close packing. The parallel inverse-sum pressure block remains robust across these tests, with close packing the main residual sensitivity.

5.1. Experimental setup

  • The experiments use a unit-square domain with pitch 1/m, disk pillars of radius κ pitch, and throat width (1−2κ)/m, with Taylor–Hood P2–P1 discretization.The inlet is parabolic, the outlet traction-free, and exterior walls and pillar surfaces are no-slip.Source: §5.1 · p. 14

5.2. Pillar density and the necessity of both summands

  • At κ = 0.25, inverse-sum iterations rise only 28→29 as m goes from 2→32, while the operator sum rises 62→520 and mass-only reaches >1000.At κ = 0.40, inverse-sum grows 35→40; the stiffness-only block fails to converge, showing that both summands are needed and their parallel combination matters.Source: §5.2 · p. 15, Figure 2 · p. 15, Table 1 · p. 15

5.3. Spectral mechanism

  • The unpreconditioned Schur condition number grows as Θ(m^2), while the preconditioned spectrum stays in [0.12, 1.88] at κ = 0.25 and [0.09, 2.06] at κ = 0.40.The ratio λmax/λmin varies by under 4% across the tested range.Source: §5.3 · pp. 15–16, Figure 3 · p. 16, Table 2 · p. 16

5.4. Discretized time-dependent Stokes: τ-uniformity

  • Across τ = 10^-4 to 10^3, the combined scale varies only 1.13–1.75× and stays below 29 iterations; one-sided scales fail at opposite ends of the sweep.At τ = 10^-4, the combined scale gives iterations for m = [4, 8, 16, 32], while the time-only control reaches 938 and 1332 at τ = 10^-2 for m = 16 and 32.Source: §5.4 · p. 16, Figure 4 · p. 17, Table 3 · p. 17

5.5. Application: tilted DLD arrays

  • On tilted DLD arrays, iteration counts remain comparable to regular arrays: 28–34 versus 29 at κ = 0.25, and 48–61 versus 39–40 at κ = 0.40.The tested row-shifts traverse the lattice flow field without exceeding the iteration budget.Source: §5.5 · p. 17, Figure 5 · p. 18

5.6. The close-packing failure boundary

  • Near close packing, iterations grow as N ∼ (1−2κ)^−0.81, with a threshold κ* ≈ 0.47 independent of m, yet the solver still converges at κ = 0.49.At m = 16 and κ = 0.49, it converges in 236 iterations with 3.1 × 10^6 degrees of freedom; the sensitivity is tied to the restriction-operator constant.Source: §5.6 · pp. 17–18

5.7. Summary

  • The experiments support density robustness, the need for both Riesz-map summands, spectral compression, time-step uniformity, and performance on tilted arrays, while close packing retains radius-dependent cost.The remaining close-packing growth is attributed to the restriction-operator constant CR.Source: §5.7 · p. 18

6. Conclusion

The paper establishes a pressure-norm equivalence for periodic pillar arrays and uses it to construct a robust Stokes preconditioner. Two-dimensional experiments support robustness across pillar density and time step, while close packing remains the main source of iteration growth.

6. Conclusion

  • The inf-sup norm is equivalent to the K-functional norm on periodic pillar arrays in the proportional-hole regime, with constants independent of periodicity.The proof uses a dual pressure extension and a fixed-domain weighted Nečas estimate.Source: §6 · pp. 18–19
  • The equivalence extends to inf-sup stable finite-element pairs admitting an ε-uniform Fortin operator, with constants independent of ε, m, and h.The argument is mesh-free and applies to arbitrary pressure functions in the stated space.Source: §6 · p. 19
  • The K-functional Riesz map yields a Brinkman-type pressure block combining a pressure-mass inverse and a scaled Laplacian inverse as a Schur-complement surrogate.The associated upper block-triangular preconditioner uses three AMG V-cycles per iteration, with cost linear in degrees of freedom.Source: §6 · p. 19
  • Iterations remain flat across a 16× range of pillar density, while close-packing growth follows N ∼ (1−2κ)^−0.81 past κ* ≈ 0.47 and all tested cases converge.The same framework yields a τ-uniform preconditioner across the time-discrete transition.Source: §6 · p. 19, Table 5 · p. 19
  • Both Riesz-map summands are individually necessary; their parallel inverse-sum, unlike the operator-sum at the same weight, delivers m-robustness.Spectral diagnostics attribute this to compression of an Θ(m^2) Schur-complement interval into an m-independent one.Source: §6 · p. 19
  • The backward-Euler extension uses the combined scale σ_{ε,τ} and yields a τ-uniform preconditioner across the time-discrete transition.The paper confirms this extension numerically.Source: §6 · p. 19

CRediT authorship contribution statement

The contribution statement assigns methodology, analysis, and writing roles across the authors, with software, validation, supervision, and funding roles also specified.

CRediT authorship contribution statement

Loading 2609.06999v1…