PaperlayerView exact revision on arXiv

Source-linked AI summary

A counterexample to Kenig's conjecture for the Laplace double-layer operator

Matthew J. Colbrook, Siavash Sadeghi

arXiv:2609.07947v1math.APcs.NAmath-phmath.MPmath.NAmath.SP

TL;DR

Kenig’s conjecture asks whether connected boundaries satisfy a strict spectral-radius bound that ensures operator-norm convergence of the associated Neumann series. The authors refine smooth periodic graph cells to grow adjoint resolvent solutions, then normalize and transfer the resulting small-residual densities to a single boundary. For every t ∈ [1/2, 1/2+ε), the domain in Theorem 1.1 satisfies r(B) ≥ t, so neither associated Neumann series converges in operator norm.

  • Problem

    Kenig’s conjecture asks whether connected boundaries satisfy a strict spectral-radius bound that ensures operator-norm convergence of the associated Neumann series.

    Source: §1 · p. 2

  • Method

    The authors refine smooth periodic graph cells to grow adjoint resolvent solutions, then normalize and transfer the resulting small-residual densities to a single boundary.

    Source: §1 · pp. 2, 4

  • Results

    For every t ∈ [1/2, 1/2+ε), the domain in Theorem 1.1 satisfies r(B) ≥ t, so neither associated Neumann series converges in operator norm.

    Source: §1 · pp. 2–3

  • Takeaways & Limitations

    The conjectured strict bound fails for the Laplace double-layer operator on a bounded simply connected planar Lipschitz domain.

    Source: §1 · p. 2

    For general Lipschitz graphs, the operators use a bounded principal-value realization; agreement of horizontal and Euclidean principal values is established for smooth graphs.

    Source: §2.1 · p. 6

Abstract

from arXiv · show

Layer potentials provide a classical approach to boundary value problems for Laplace's equation on Lipschitz domains. Kenig's 1994 spectral-radius conjecture for the double-layer operator would ensure operator-norm convergence of the associated Neumann series on mean-zero $L^2$ densities when the boundary is connected. We disprove this conjecture by constructing a bounded simply connected planar Lipschitz domain whose double-layer operator on arclength $L^2$ has essential spectral radius strictly greater than $1/2$. More precisely, for every $t>1/2$ sufficiently close to $1/2$, we obtain such a domain with $\pm i t$ in its Fredholm essential spectrum. The construction starts from smooth graphs whose shapes repeat under translation. In the limit of separated scales, refinement makes solutions of adjoint resolvent equations grow with fixed forcing. The graph slopes remain uniformly bounded. A computer-assisted certificate proves this growth through an inequality for Hermitian $2\times2$ matrices. Its strict margin at $- i/2$ persists at nearby spectral parameters. Normalisation and a Floquet transform then give compactly supported densities with small residuals on the full graphs. We insert rescaled segments of successive graphs into one bounded boundary, where these densities form a weakly null sequence of approximate eigenvectors. The same spectral conclusion holds on a single periodic Lipschitz graph. The certificate combines continuous estimates, exact rational arithmetic and rigorous interval enclosures.

1. Introduction

The paper disproves Kenig’s strict spectral-radius conjecture for the Laplace double-layer operator by constructing a bounded simply connected planar Lipschitz domain. Its construction turns refinement-driven resolvent growth on periodic graphs into small-residual densities on a single boundary.

1. Introduction

  • The conjecture concerns a strict spectral bound proposed to ensure operator-norm convergence of Neumann series for connected boundaries.For connected Γ, the strict spectral-radius and essential-spectrum formulations are equivalent.Source: §1 · pp. 2–3
  • For every t in [1/2, 1/2+ε), a bounded simply connected Lipschitz domain has spectral radius at least t, so the conjectured strict bound fails.The resulting Neumann series do not converge in operator norm, although both boundary integral operators are invertible.Source: §1 · pp. 2–3
  • The construction refines two smooth periodic cells with tilts ±1/2; restricting refinement to matching straight segments keeps slopes uniformly bounded.More refinements are used for smaller residuals, and the profiles are fixed across t while scales and the resulting domain may vary.Source: §1 · pp. 2, 4
  • At z = −i/2, a positive-map certificate proves A(X) > (21/20)X, forcing growth under refinement that persists at nearby z = −it.The uniform bound for the auxiliary singular integral then forces the density norm to grow; strong convergence transfers normalized densities and small residuals to smooth periodic graphs.Source: §1 · p. 4
  • A Floquet transform and truncation produce compactly supported unit densities with residuals tending to zero on the full graphs.Long rescaled graph segments are inserted on disjoint intervals accumulating at the origin to form a single bounded boundary.Source: §1 · pp. 4–5

2. Graph operators and periodic cells

The section sets up graph-based double-layer operators in horizontal coordinates, relates them to boundary arclength, and develops periodic-cell operators and their Floquet decomposition. It also establishes boundedness and continuity results needed to compare operators as graph profiles vary.

2.1. Horizontal coordinates and normalisation.

  • Horizontal coordinates provide a common unweighted space for comparing graph operators, with arclength norms controlled by the slope bound.The direct operator is represented by P_F in horizontal coordinates, while the physical adjoint transfers residuals and norms to the boundary with comparisons depending only on the Lipschitz bound.Source: §2 · p. 5, §2.1 · p. 6

2.2. Periodisation and Floquet decomposition.

  • For smooth graph profiles, the auxiliary singular integral R_F couples to P_F, while T_F is recovered from their operator identity.The periodized kernels yield bounded operators; smoothness gives compactness of P_F and R_F.Source: §2.1 · p. 5, §2.2 · p. 8
  • Symmetric periodisation defines operators on one cell, and the Floquet transform decomposes full-line graph operators into phase fibres.The periodic operators are bounded with norms controlled by the graph slope; P_F is compact and the stated mean identities hold.Source: §2.1 · p. 6, §2.2 · pp. 6–8

2.3. Continuity of a fixed family.

  • C² convergence of profiles and convergence of tilts imply operator-norm convergence of P_F and R_F on the period.The local operators are uniformly bounded on compact sets of smooth profile parameters and bounded tilts.Source: §2.3 · p. 8

3. A joint two-scale limit

The section separates slow graph position from fast cell position to derive joint two-scale limits for the double-layer operator and an auxiliary singular integral. These limits hold strongly for fixed inputs and identify the corresponding adjoint limit.

3. A joint two-scale limit

  • The two-scale setup tracks the double-layer operator and auxiliary singular integral together because the former couples through the latter.Unfolding places the slow position and cell coordinate on a product space, where the local periodic operators act in the cell variable.Source: §3 · p. 9
  • Near the diagonal, kernel comparison and periodic tail estimates recover the full local periodic operators as the scale separation increases.The estimates first compare with a frozen cell, then control the omitted tail before taking the near-region limits.Source: §3 · pp. 9–10
  • Away from the diagonal, periodic averaging gives the slow-graph contribution, while symmetric image cancellation controls remote translates.The auxiliary singular integral contributes its slow-graph limit; the double-layer limit also contains a term involving the oscillating source derivative.Source: §3 · pp. 10–11
  • For fixed inputs, the unfolded operators converge strongly to local cell operators plus averaged slow-graph terms, and the adjoints converge to the adjoint limit.The double-layer limit includes both the local operator and terms involving the slow-graph operators; uniform boundedness extends convergence from smooth inputs to all L2 inputs.Source: §3 · pp. 9–11
  • The auxiliary singular integral can act on fast oscillations even when the original graph is unchanged, while killing constant fast inputs.For an affine graph, a product Fourier input with nonzero fast frequency corresponds to a combined frequency in the original variable.Source: §3 · p. 11

4. Finite refinements with a stopping rule

The section realizes finite product-space refinements on smooth periodic graphs using successively separated scales. A stopping rule limits slopes, while a resolvent recursion establishes invertibility at each finite depth.

4.1. Realisation at separated scales.

  • At any fixed depth, a smooth periodic graph and vector can approximate a product-space vector’s norm and adjoint residual to any prescribed error, with uniform operator bounds.The integers setting the scales are chosen from coarse to fine, since each finer threshold may depend on earlier choices.Source: §4.1 · pp. 12–13
  • The two-scale limit is iterated in an ordered sequence of scales, with the finest variable unfolded first and each finer scale chosen after the coarser ones.This produces the product-space operators as strong limits of smooth finite-depth graph operators.Source: §4.1 · pp. 12–13

4.2. Stopping and the slope bound.

  • The stopping rule inserts a new cell only when all preceding coordinates lie in J, keeping slopes bounded independently of refinement depth.The stopped recursions and their adjoints are strong limits of smooth realisations and obey the same slope bound.Source: §4.2 · pp. 13–14

4.3. The resolvent recursion.

  • Splitting each refinement into preceding variables and a new mean-zero component makes the operator triangular; induction gives invertibility at each depth.The local lower block is a fixed cell compression or zero, while the full inverse norm may depend on depth.Source: §4.3 · pp. 14–15
  • On the refined set, the recursion updates the solution through the cell functions; outside J, the update matrix is the identity.When a preceding coordinate lies outside J, the local graph is affine and the new mean-zero component vanishes.Source: §4.3 · pp. 14–15

5. Growth under refinement

The section gives a sufficient positive-map condition for resolvent solutions to grow under repeated refinement. A fixed forcing then yields normalized densities with small residuals on smooth periodic graphs.

5. Growth under refinement

  • A strict matrix inequality for a positive map yields growth of the resolvent solutions under repeated refinements.The two positive maps encode the integrated quadratic form of the solution pair on portions that continue to be refined; positivity permits a lower bound by retaining those portions.Source: §5 · pp. 15–16
  • Using one fixed forcing at every depth, the growing solution norms produce mean-zero normalized vectors with residuals tending to zero on smooth periodic graphs.The construction removes any small mean introduced during realisation; the resulting graph slope is at most 9/2.Source: §5 · pp. 16–17

6. Two fixed cells

The section constructs two fixed smooth cells from polynomial profiles, verifies a strict matrix inequality, and shows that the inequality persists for nearby spectral parameters. The cells’ slopes and resolvent bounds are controlled.

6. Two fixed cells

  • The polynomial cell and its reflection satisfy a strict matrix inequality at z = −i/2, with a certified margin.The certificate uses exact rational integration and bounds, including a positive remaining margin of 49/4950.Source: §6 · pp. 17–18
  • Smoothing the polynomial profiles preserves the inequality and yields two fixed C∞ cells with slopes bounded by 7/2.Their mean-zero resolvents at z0 have norms less than 80, and their slopes remain −m on a neighbourhood of J.Source: §6 · p. 18
  • The strict inequality persists for spectral parameters near z0 while the smooth cells remain fixed.Continuity of the moment maps gives an ε-neighbourhood in which the cell resolvents exist and the inequality holds.Source: §6 · pp. 17–18
  • For t in [1/2, 1/2 + ε), the fixed cells satisfy the growth criterion at z = −it, with geometric cell slopes ±1/2.The forcing is chosen once, while later approximation and truncation choices depend on the depth and preceding choices.Source: §6 · p. 19

7. From one period to the full graph

Approximate eigenvectors on one period are transferred to compactly supported densities on the full smooth graphs. This yields a periodic Lipschitz graph with ±it in its Fredholm essential spectrum for t just above 1/2.

7. From one period to the full graph

  • For every t in [1/2, 1/2+ε), a real 1-periodic Lipschitz graph with slope at most 27/2 has ±it in its Fredholm essential spectrum.The densities are supported on arclength L2 of the graph.Source: §7 · p. 19
  • Mean-zero approximate eigenvectors on one period transfer to full graphs because the adjoint fibre operator’s rank-one jump at zero phase vanishes on mean-zero densities.Phases sufficiently close to zero preserve small residuals.Source: §7 · p. 19
  • The transferred vectors yield unit-norm densities with small adjoint residuals, which may be chosen with compact horizontal support.The construction uses a unitary representation on unweighted horizontal space, and truncation preserves the norm and residual conclusion.Source: §7 · p. 19

8. The periodic graph and the bounded domain

Compactly supported approximate eigenvectors from successively refined graphs are inserted at separated scales to produce weakly null sequences on a bounded domain and a periodic graph. The construction gives the stated essential-spectrum result, while a Calkin-algebra argument establishes spectral symmetry on a closed boundary.

8. The periodic graph and the bounded domain

  • The construction places compactly supported densities on disjoint, rescaled graph segments, making a weakly null sequence whose residual tends to zero on the entire boundary.Large graph segments control the tail, while sufficiently small scales fit them into disjoint intervals accumulating at the origin.Source: §8 · pp. 20–22
  • The inserted top graph remains Lipschitz and differentiable at its sole accumulation point, and the resulting bounded domain is simply connected.The top graph has Lipschitz constant at most 3L0.Source: §8 · pp. 20–21
  • For each t in [1/2, 1/2+ε), the constructed bounded simply connected domain has ±it in the Fredholm essential spectrum of its double-layer operator.The proof uses a weakly null normalized sequence with residual tending to zero; the top graph has slope at most 27/2.Source: §8 · p. 21

8.1. Spectral symmetry on a closed boundary.

  • On a bounded simply connected planar Lipschitz domain, conjugation and Calkin-algebra similarity give essential-spectrum symmetry about both the real axis and the origin.The proof uses the real kernel for conjugation symmetry and the normalized Cauchy singular integral for symmetry about the origin.Source: §8.1 · p. 22

Appendix A. Certification of the local cell

The appendix certifies a local matrix inequality by approximating the continuous local operator and a candidate function pair. Exact rational calculations and interval enclosures bound the errors and transfer positivity to the true pair.

Appendix A. Certification of the local cell

  • An approximation of the continuous local operator in Hilbert–Schmidt norm provides an inverse bound on the full space.The inverse bound uses interval enclosures, while subsequent residual and matrix calculations use integers and rationals.Source: Appendix A · pp. 22–23
  • Exact rational polynomial integration bounds the residuals of the proposed functions, and the inverse estimate controls their distance from the true pair.The certified functions provide the candidate pair for the local inequality.Source: Appendix A · p. 22
  • A final perturbation estimate transfers the positive matrix inequality from the candidate functions to the true pair.The proposition states the certified local inequality for the specified graph.Source: Appendix A · pp. 22–23

A.1. The exact graph and exact input functions.

The certified cell is a smooth periodic profile built from a rounded indicator and explicit polynomial transitions. Exact piecewise-polynomial inputs satisfy continuity and mean-zero constraints, with derivative bounds controlling the graph.

A.1. The exact graph and exact input functions.

  • The profile uses m=1/2, p=3/4, and δ=1/16, with F′=7/2−4w0 and h=F−x/2, where w0 is a rounded indicator of [0,p].The transition function rises near zero and falls near p.Source: §A.1 · p. 23
  • Explicit polynomial pieces define the graph, whose periodic extension satisfies F(x+1)=F(x)+1/2 and matches through order four.Only the four transition joins require separate kernel estimates.Source: §A.1 · pp. 23–24
  • The graph satisfies derivative bounds ∥F′∥∞≤7/2, ∥F′′∥∞≤70, and ∥F′′′∥∞≤6720, with oscillation of h at most 3/2.A fifth-derivative bound is also supplied for polynomial pieces.Source: §A.1 · p. 24
  • Four exact piecewise-polynomial functions are specified by rational coefficients; continuity at the 64 joins and the two zero-mean conditions are verified exactly.The first two functions have integral zero, and the exact operator inputs are integrated using Chebyshev polynomial identities.Source: §A.1 · p. 24

A.2. Bounds for the continuous kernels.

The continuous-kernel estimates control local and remote terms, including every remote image, and establish regularity across the periodisation seam and profile joins.

A.2. Bounds for the continuous kernels.

  • The periodised kernel is absolutely continuous along horizontal and vertical segments, with derivative bounds that exclude singular terms at the seam, diagonal, and profile joins.The rational comparisons use 3 < π < 22/7 and π^2 < 10.Source: §A.2 · p. 25
  • Schur’s test bounds the remainder after separating the target Hilbert coefficient and accounting for local and remote terms.The estimate applies to R1 and includes every remote image.Source: §A.2 · p. 26
  • The sampled step-kernel operator is represented by the midpoint kernel values divided by the grid size in the normalized interval-indicator basis.Its Frobenius norm equals the Hilbert–Schmidt norm after extension by zero off the grid subspace.Source: §A.2 · p. 26

A.3. An inverse bound from the squared operator.

The inverse estimate is obtained by bounding the squared operator, proving continuous-norm convergence of a Neumann series, and restricting the resulting inverse to mean-zero data.

A.3. An inverse bound from the squared operator.

  • The Neumann series for P*^2 + 1/4 converges in the continuous operator norm, yielding the full-space inverse by factorisation.The inverse estimate is obtained by estimating the square of the operator.Source: §A.3 · p. 26
  • The inverse bound for P* + i/2 is less than 40.The passage gives the bound as 1203000/32891 < 40.Source: §A.3 · p. 26
  • A mean-zero right-hand side has a mean-zero solution, so restriction gives the inverse stated in (A.3).The argument uses W(P* + i/2)u = (i/2)Wu.Source: §A.3 · p. 26

A.4. Interval evaluation of the sampled operator.

The sampled operator is evaluated with rigorous interval arithmetic, using exact rational inputs and outward enclosures to certify the bounds needed for the inverse estimate.

A.4. Interval evaluation of the sampled operator.

  • Arb interval arithmetic within FLINT encloses real and complex quantities, including conversion and rounding errors, at 96-bit working precision.The matrix products are enclosed in 32 blocks of 128 rows, and their outward dyadic endpoints are summed using exact rational arithmetic.Source: §A.4 · pp. 26–27
  • The profile values and derivatives used in the midpoint calculations are formed as exact rationals, with the profile assembled from matched piecewise primitives.The passage gives F = 7x/2 − 4∫w0 and F′ = 7/2 − 4w0.Source: §A.4 · p. 26
  • The sampled kernel matrix is formed from the exact periodised Cauchy kernel, using a removable diagonal value and polynomial evaluations with certified tail bounds.At every off-diagonal sample, interval arithmetic verifies |ζ| < 6; the program checks that S0 excludes zero and the local denominator is positive before division.Source: §A.4 · p. 27
  • The interval enclosures have upper endpoints strictly below (61/100)^2 and (95/1000)^2, proving (A.24) and hence (A.26).The decimal endpoints are exact rationals chosen outward from dyadic enclosures.Source: §A.4 · p. 27

A.5. Exact treatment of the infinite periodic tail.

This appendix section completes the certificate by rigorously bounding periodic-tail and logarithmic contributions, residuals, and auxiliary singular-integral actions using exact calculations and rational enclosures. It verifies the matrix inequality, transfers it to the true pair, and documents the verification procedures.

A.5. Exact treatment of the infinite periodic tail.

  • The periodic tail is handled by exact polynomial moments and rational Euler–Maclaurin bounds, yielding uniform action bounds for all four actions on f0 and f1.The calculation retains images −8 through 8, bounds the omitted odd powers and coefficient errors, and uses the stored tail polynomial in residual estimates.Source: §A.5 · pp. 28–29

A.6. Finite-image polynomial enclosures.

  • The finite-image polynomial enclosure has action error below 10^-15 on every target interval.The partition treats all seventeen retained images, with exact source integration and an interval enclosure for 1/(2π).Source: §A.6 · pp. 29–30

A.7. The two adjoint residuals.

  • Both adjoint residuals have L2 norms below 7 × 10^-8 after combining polynomial norm calculations with analytic action-error bounds.The exact squared norms satisfy s0 < 2.56 × 10^-20 and s1 < 5.52 × 10^-20; the analytic action error is below 6.812 × 10^-8.Source: §A.7 · p. 31

A.8. The principal value for the auxiliary singular integral.

  • The smooth-action error eR,sm is below 6.812 × 10^-8; the analysis allows 10^-8 in L2 for each input, and seam,1 is below 4.26 × 10^-17.Source: §A.8 · p. 32

A.9. Rational logarithms and the final residuals.

  • Rational logarithm enclosures are combined with target-interval refinement to approximate the remaining three-panel principal-value terms.The refinement gives 784 intervals; logarithm and polynomial-rounding errors are charged explicitly.Source: §A.9 · p. 33
  • Both final residual norms are below 8×10^-8, completing the four residual bounds used in A.11.The largest accepted approximation degrees are 19 for logarithms and 12 for a.Source: §A.9 · pp. 33–34

A.10. Exact moments and the positive matrix test.

  • The exact Gram test establishes (A.6), with determinant lower bounds 1/4 and 1/2000 for the middle and final matrices.The test verifies positive definiteness using leading entries and determinants.Source: §A.10 · p. 34

A.11. Perturbation of the pair and the positive map.

  • Perturbation estimates transfer the approximate-pair matrix inequality to the true pair and establish Proposition A.1.The transfer uses weakened residual bounds for the approximate pair.Source: §A.10 · p. 34, §A.11 · p. 35

A.12. Certificate files and verification.

  • The verifier recomputes exact moments, squared residual norms, analytic errors, and matrix inequalities; replay rebuilds all 784 principal-value and assembly rows.The certificate checks also combine finite checks with kernel identities and tail, corner, and perturbation bounds.Source: §A.12 · pp. 35–36
Loading 2609.07947v1…