Source-linked AI summary
Convergence Rates in L^2 for Elliptic Homogenization Problems
Carlos E. Kenig, Fanghua Lin, Zhongwei Shen
TL;DR
The paper asks how rapidly solutions of elliptic systems with rapidly oscillating coefficients converge in Lipschitz domains. It combines corrector and boundary-layer arguments with uniform L2 boundary estimates, obtaining solution and eigenvalue convergence rates under ellipticity, periodicity, symmetry, and domain assumptions.
Problem
The paper studies rates of L2 and H1/2 convergence for elliptic systems with rapidly oscillating coefficients in Lipschitz domains and the resulting rates for associated eigenvalues.
Method
The proofs combine corrector expansions, enlarged-domain homogenized solutions, weighted potential estimates, and uniform L2 Dirichlet and Neumann estimates.
Results
The paper proves O(ε) solution estimates under H2 regularity and eigenvalue rates O(ε) in smoother domains or O(ε^σ) for any σ > 0 in Lipschitz domains.
Takeaways & Limitations
The results provide convergence rates for Dirichlet, Neumann, and Steklov eigenvalues, extending known results to bounded Lipschitz domains and systems.
Takeaways & Limitations
The logarithmic factor in one L2 estimate is not known to be necessary, even for smooth domains.
Abstract
from arXiv · showhide
We study rates of convergence of solutions in L^2 and H^{1/2} for a family of elliptic systems {L_ε} with rapidly oscillating oscillating coefficients in Lipschitz domains with Dirichlet or Neumann boundary conditions. As a consequence, we obtain convergence rates for Dirichlet, Neumann, and Steklov eigenvalues of {L_ε}. Most of our results, which rely on the recently established uniform estimates for the L^2 Dirichlet and Neumann problems in \cite{12,13}, are new even for smooth domains.
1 Introduction
The paper establishes convergence rates for elliptic systems with periodic coefficients in bounded Lipschitz domains, under Dirichlet and Neumann conditions, and derives corresponding eigenvalue rates.
- Scope: The study targets L2 convergence rates for solutions of elliptic systems with rapidly oscillating coefficients in bounded Lipschitz domains.It also considers H1/2 and boundary maximal-function estimates.
- Main estimates: Under Dirichlet conditions, the paper proves an O(ε) estimate involving the interior L2 error and the boundary radial maximal-function error when u0 ∈ H2(Ω).The assumptions include a bounded Lipschitz domain, A ∈ Λ(µ, λ, τ), and A = A*.
- Main estimates: The same estimate is obtained for Neumann conditions when u0 ∈ H2(Ω), using uniform L2 Neumann estimates.The Neumann data satisfy the compatibility condition ∫∂Ω g = 0.
- Boundary behavior: Boundary maximal-function estimates imply L2 convergence on surfaces parallel to the boundary, while the F = 0 case yields additional estimates with a logarithmic factor.The radial maximal operator controls traces on nearby parallel surfaces.
- Approach: The approach combines corrector expansions, boundary-layer constructions, weighted singular-integral estimates, and uniform L2 Dirichlet and Neumann estimates.For Neumann problems, the proof uses an explicit conormal-derivative computation.
- Eigenvalues: Eigenvalue errors are O(ε) in C1,1 or convex settings and O(ε^σ) for any σ > 0 in Lipschitz domains, for Dirichlet, Neumann, and Steklov eigenvalues.The results extend previously known rates and are new even for smooth domains in many cases.
2 Uniform regularity estimates
This section develops uniform regularity estimates for oscillatory elliptic operators in Lipschitz domains, including maximal-function, square-function, and fractional Sobolev bounds.
- Definitions and framework: The section recalls uniform regularity estimates for {Lε} and defines non-tangential and radial maximal operators.These tools support the later convergence proofs.
- Fundamental solutions: Interior gradient estimates imply pointwise bounds for the fundamental solution and its first and mixed derivatives.The bounds scale as |x − y|^(2−d), |x − y|^(1−d), and |x − y|^(−d), respectively.
- Dirichlet estimates: For L2 Dirichlet data, solutions satisfy non-tangential maximal-function estimates, and H1 boundary data yield L2 control of the gradient maximal function.The estimates hold uniformly for the oscillatory family under the stated assumptions.
- Dirichlet estimates: The Dirichlet theory also provides an H1/2(Ω) estimate from L2 boundary data via square-function estimates and real interpolation.The double-layer potential representation supplies the square-function estimate.
- Neumann estimates: For the L2 Neumann problem, weak solutions are unique up to constants and satisfy an L2 estimate for the gradient’s non-tangential maximal function.The boundary data obey the compatibility condition ∫∂Ω g = 0.
- Maximal functions: Radial maximal functions control values on nearby parallel surfaces and, for solutions of Lεuε = 0, are comparable with interior and boundary L2 norms.The comparison uses the interior L∞ estimate.
3 Homogenization of elliptic systems
The section constructs correctors and flux potentials for periodic elliptic systems, then uses them with boundary corrections to prove L2 homogenization estimates in Lipschitz domains.
- Correctors: Periodic correctors χ are introduced for Lε = −div(A(x/ε)∇), with the homogenized operator L0 defined through the effective matrix Â.The coefficients satisfy the ellipticity and periodicity assumptions.
- Flux potentials: Divergence-free, mean-zero periodic fluxes admit antisymmetric H1 potentials, enabling the corrector-based error decomposition.The potentials satisfy wij = −wji.
- Regularity: Under A ∈ Λ(µ, λ, τ), the flux-potential derivatives have bounded regularity, supporting quantitative control of the corrector expansion.The relevant bounds depend only on m, d, µ, λ, and τ.
- Error decomposition: When Lε(uε) = L0(vε), the difference is decomposed into a corrector term and boundary or interior error terms.This identity underlies the subsequent energy estimates.
- Dirichlet homogenization: For Dirichlet problems in Lipschitz domains with u0 ∈ H2(Ω), the constructed approximation yields the main L2 and boundary maximal-function estimate.The result is stated under symmetry A = A*.
- Boundary correction: For rougher boundary data, u0 is replaced by a homogenized solution vε on a domain at distance approximately ε from Ω, with transported boundary data.This avoids requiring H2 regularity of u0 in the original Lipschitz domain.
4 Dirichlet boundary condition
For Dirichlet problems on bounded Lipschitz domains, the section constructs first-order approximations and derives L2 and boundary convergence rates for oscillatory elliptic systems.
- Construction: The approximation uses an enlarged domain Ωε, a boundary map Λε, and lifted boundary data fε.The construction defines vε on Ωε and compares it with the homogenized solution u0.
- Theorem 4.1: Theorem 4.1 provides estimates for the approximation error wε under symmetric coefficients on bounded Lipschitz domains.The constants depend only on µ, λ, τ, d, m, a, and Ω.
- Proof strategy: The proof decomposes the error into boundary, interior, and auxiliary components, then controls them using energy, duality, and weighted L2 estimates.Weighted potential estimates are established separately for use in earlier arguments.
- Convergence rates: The resulting estimates imply L2 convergence of uε to u0 and corresponding boundary maximal-function control.The error is bounded through wε together with an O(ε) comparison term.
- Convergence rates: Theorem 1.1 records the Dirichlet estimate as a consequence of Corollaries 3.5 and 4.2.The cited theorem includes the L2 and boundary estimates for the Dirichlet problem.
5 Neumann boundary condition, part I
For Neumann problems, the section develops convergence estimates by combining first-order error decompositions with duality arguments and constant-coefficient boundary estimates.
- Theorem 5.2: Theorem 5.2 analyzes Neumann solutions under a compatibility condition and symmetric coefficients on bounded Lipschitz domains.The proof separates the error into θε, zε, and a constant correction ρ.
- Convergence rates: Theorem 5.2 yields convergence estimates for the Neumann problem, including an L2 estimate for uε−u0.The section notes that the L2 estimate had previously been proved for curvilinear convex domains in R2.
- Proof strategy: The boundary component θε is controlled by a duality argument using an auxiliary L2 Neumann problem.The argument applies boundary gradient estimates and then uses constant-coefficient Dirichlet estimates.
- Remarks: The estimates also remain valid under the additional condition stated in Remark 5.4.The constant correction ρ is specified differently under that condition.
- Applications: The Neumann estimates are later used in the error analysis for Neumann eigenvalues.This connection is stated explicitly after the convergence result.
6 Neumann boundary condition, part II
The second Neumann section extends the Dirichlet construction to Neumann data and derives convergence rates through decomposition, duality, and weighted estimates.
- Construction: The construction uses vε from the enlarged-domain problem with boundary data derived from u0 on ∂Ω.The same lifted solution framework used for the Dirichlet analysis is adapted to Neumann conditions.
- Theorem 6.1: Theorem 6.1 establishes estimates for the Neumann approximation error wε under symmetric coefficients on bounded Lipschitz domains.The constants depend only on µ, λ, τ, d, m, a, and Ω.
- Convergence rates: Corollary 6.2 converts Theorem 6.1 into convergence rates for uε to u0 in L2.The theorem’s estimates are explicitly identified as the source of the corollary.
- Proof strategy: The proof proceeds as in the Dirichlet case and combines estimates for vε−u0 with duality arguments for the remaining error terms.Lemma 4.6 is used to handle one component of the decomposition.
- Main theorem: Theorem 1.2 identifies Corollaries 5.3 and 6.2 as the sources of the Neumann convergence estimates.These estimates provide the Neumann part of the paper’s main results.
- Remarks: The estimates remain valid under the additional condition in Remark 6.3.The stated parameter restriction includes any a > 1.
7 Convergence rates for eigenvalues
The section derives convergence rates for Dirichlet, Neumann, and Steklov eigenvalues from operator convergence, with O(ε) rates under stronger domain regularity and ε^σ rates in general Lipschitz domains.
- Operator framework: The eigenvalue analysis reduces Dirichlet, Neumann, and Steklov problems to compact self-adjoint operator families satisfying a spectral convergence theorem.Dirichlet and Neumann eigenvalues correspond to reciprocals of eigenvalues of solution operators, while the Steklov analysis uses the boundary solution operator.
- Regularity boundary: An O(ε) eigenvalue estimate follows from H2 regularity of the homogenized eigenfunction, which is available in C1,1 domains or scalar convex domains but not general Lipschitz domains.The section explicitly notes that H2 regularity can fail on general Lipschitz domains.
- Dirichlet eigenvalues: For general bounded Lipschitz domains, Dirichlet eigenvalues satisfy an ε^σ convergence estimate for every σ > 0, with constants independent of ε.The constant may depend on k and σ, but not on ε.
- Neumann eigenvalues: For general bounded Lipschitz domains, nonzero Neumann eigenvalues satisfy an ε^σ convergence estimate for every σ > 0.The constant depends on k and σ, but not on ε; an O(ε) estimate holds under C1,1 regularity or convexity in the scalar case.
- Steklov eigenvalues: For general bounded Lipschitz domains, nonzero Steklov eigenvalues satisfy an ε^σ convergence estimate for every σ > 0.The associated boundary operator is the inverse of the Dirichlet-to-Neumann map, and O(ε) holds under C1,1 regularity or scalar convexity.
8 Weighted potential estimates
This section develops weighted potential estimates in Lipschitz graph domains using fundamental-solution bounds, singular-integral inequalities, maximal-function estimates, and boundary non-tangential controls.
- Potential construction: The potential Hε is defined by integrating derivatives of the oscillatory fundamental solution against h, and its gradient is bounded in L2 by C∥h∥L2(Ω).The estimate is attributed to prior work and supplies a basic energy bound for the potential.
- Interior and boundary control: Interior L2 bounds for Hε follow from pointwise fundamental-solution derivative estimates and control by weighted estimates near the boundary.Compact interior estimates are combined with a boundary-layer argument over parallel domains.
- Weighted inequalities: The weighted estimates reduce, after localization and rescaling, to singular-integral and Hardy–Littlewood maximal-operator inequalities with A∞ weights.The proof invokes Calderón–Zygmund bounds, Coifman–Fefferman theory, and Sawyer’s two-weight characterization.
- Weight verification: The two-weight maximal-function condition is verified for the relevant weights after a bi-Lipschitz flattening, dimensional reduction, and rescaling to ε = 1.The argument establishes the condition for every a ≥ 0.