PaperlayerView exact revision on arXiv

Source-linked AI summary

CutIGA with Least Squares Stabilized Nitsche Boundary Conditions: The Biharmonic Problem

Mats G. Larson, Karl Larsson

arXiv:2610.11636v1math.NAcs.NA

TL;DR

Unfitted biharmonic discretizations must impose multiple boundary traces even when the domain boundary cuts the mesh arbitrarily. The paper proposes a symmetric least-squares stabilized Nitsche method based on lifting both traces. For degree-p splines with p ≥4, it establishes energy and L2 error orders p−1 and p+1, respectively.

  • Problem

    Unfitted higher-order discretizations must impose several boundary traces, while small cut intersections make penalty selection difficult.

    Source: §1 · pp. 1–2

  • Method

    An explicit lifting of both Dirichlet traces supports a symmetric Nitsche method with bulk least-squares stabilization and tangential boundary penalties.

    Source: §1 · p. 2

  • Results

    For p ≥4, energy and L2 errors have orders h^(p−1) and h^(p+1), respectively, with constants independent of the cut configuration.

    Source: §3.3 · p. 8

  • Takeaways & Limitations

    For sufficiently thin boundary strips, fixed penalties can be independent of the cut configuration and bulk stabilization parameter.

    Source: §1 · p. 2, §5 · p. 14

    Further questions include liftings near corners and edges on piecewise smooth domains, and stability and accuracy when boundary geometry is approximated.

    Source: §5 · p. 14

Abstract

from arXiv · show

We propose a symmetric Nitsche method for the biharmonic Dirichlet problem on a smooth domain whose boundary may cut arbitrarily through the computational mesh. Bulk least-squares stabilization in a boundary strip and tangential boundary penalties control the consistency terms through an explicit lifting of the two Dirichlet traces. This gives coercivity on $H^4(\Omega)$ without inverse inequalities and permits a fixed penalty independent of the cut configuration for sufficiently thin strips. Under suitable regularity assumptions, splines of degree $p\geq4$ give energy and $L^2$ error bounds of orders $h^{p-1}$ and $h^{p+1}$. We also analyze a face-stabilized extension to $C^1$ splines of degree $p=2$ and $C^2$ splines of degree $p=3$. Exact-ellipse experiments examine convergence, cut stability, and parameter sensitivity.

1 Introduction

Unfitted biharmonic discretizations must impose two boundary traces despite cuts that can make penalty selection difficult. The paper introduces a symmetric least-squares stabilized Nitsche method, with coercivity and error estimates established through an explicit lifting.

Motivation

  • Small intersections between the physical domain and background elements make penalty selection difficult when several higher-order boundary traces must be imposed.Insufficient symmetric Nitsche penalties can harm coercivity and accuracy, while excessive penalties can cause locking on curved boundaries.Source: §1 · pp. 1–2

Background

  • The biharmonic Dirichlet problem imposes both the boundary value and normal derivative as essential conditions in the H2(Ω) energy space.Nitsche’s method imposes essential conditions weakly while preserving consistency in trimmed geometries where boundaries cross the grid independently of knot lines.Source: §1 · p. 1

Earlier work

  • Unlike the cited fourth-order approaches, the method controls consistency on the full space H^4(Ω); its main analysis for C^3 or smoother splines uses only cell and physical-boundary integrals.The comparison identifies [8] as focused on simply supported plates and [9] as using a discrete-extension framework for the clamped problem.Source: §1 · p. 2

Contribution

  • The method uses an explicit Hermite lifting of both Dirichlet traces to control consistency on H^4(Ω), yielding coercivity without inverse inequalities.Bulk and tangential least-squares terms determine sufficient relative weights for the stabilization and six boundary penalties.Source: §1 · p. 2
  • For splines of degree p ≥4, the method gives energy and L2 error orders p−1 and p+1, respectively.The fixed penalty is independent of the cut configuration and bulk stabilization parameter when the boundary strip is sufficiently thin.Source: §1 · p. 2

2 Method

The method uses spline spaces on an active mesh and a boundary strip that supports bulk least-squares stabilization. Its symmetric Nitsche form combines consistency terms and trace penalties, with a face-stabilized extension for lower-degree splines.

2.1 Domain and Approximation Space

  • The main space Vh comprises restrictions to the active mesh Th(Ω) of tensor-product B-splines of degree p ≥ 4 and maximum continuity Cp−1, whose restrictions to Ω lie in V.The stability argument on V does not require a Cartesian mesh; spline approximation and numerical experiments use uniform Cartesian grids, with h the cell diameter.Source: §2.1 · p. 3
  • The bulk least-squares term is integrated over an interior strip Dδ of width δ, chosen as ρh for approximation estimates.Its inner boundary is a physical-distance surface independent of the background mesh; the alternative mesh-based strip avoids constructing that surface but varies with the mesh.Source: §2.1 · p. 4

2.2 LS Nitsche

  • Six boundary penalty terms include tangential derivatives, and their relative weights follow from lifting estimates; for sufficiently thin strips, the analysis gives β = 30.Standard symmetric Nitsche omits the bulk least-squares term and four tangential derivative penalties, together with their load terms.Source: §2.2 · pp. 4–5
  • The symmetric form combines the volume Laplacian term, bulk least-squares stabilization, symmetric consistency terms, and penalties for the two Dirichlet traces.The load includes corresponding bulk and boundary pairings; τ trades weight between bulk stabilization and boundary penalization, while β is the common boundary penalty parameter.Source: §2.2 · p. 4
  • For p = 2 and p = 3, a consistent face form penalizes jumps of ∆v and ∂nF∆v, respectively, across active faces intersecting the strip.The penalized jump vanishes for a smooth exact solution, leaving the load unchanged.Source: §2.2 · p. 5

3 Analysis

The analysis uses an explicit lifting of both boundary traces to control the consistency terms and establish coercivity without inverse inequalities. It derives cut-independent interpolation and error estimates, including stability and error results for lower-degree splines.

3.1 Lifting and Stability

  • For sufficiently thin strips, the lifting-based method is coercive without inverse inequalities, and β = 30 suffices independently of the cut configuration and bulk stabilization parameter.For C2 cubic splines, coercivity also holds with β ≥ 30 under the stated geometric and parameter conditions.Source: §3.1 · pp. 5–6

3.2 Interpolation Estimates

  • The spline interpolation estimates for functions in H^s(Ω), 4 ≤ s ≤ p + 1, have constants independent of how the boundary cuts the mesh.The local approximation estimates use a stable quasi-interpolant and bounded-overlap patches.Source: §3.2 · p. 7

3.3 Error Estimates

  • For p ≥ 4 and u ∈ H^(p+1)(Ω), the energy and L2 error orders are p − 1 and p + 1, with constants independent of the cut configuration.The estimates assume the regularity and parameter conditions stated in Assumption 3.1.Source: §3.3 · p. 8

3.4 Error Estimates for Lower-Degree Splines

  • For C1 quadratic and C2 cubic splines, the extended method has cut-independent coercivity under the stated hypotheses and error estimates for u ∈ H4(Ω).For quadratics, θ, σ ≤ 1/10 are sufficient for the stated coercivity result, but are not necessary; within the tubular radius, actual values below 1 can yield a geometry-dependent sufficient β.Source: §3.4 · pp. 8–9

4 Numerical Results

Exact-ellipse experiments assess convergence, cut stability, parameter sensitivity, and a face-stabilized extension to quadratic splines. Across the tested cuts, LS Nitsche remains positive definite where matched standard Nitsche can become indefinite.

4.1 Setup and Diagnostics

  • The experiments use exact-ellipse geometry and default C3 quartic parameters, with diagnostics based on a shared energy metric and matched penalties in comparisons.Removal is used for conditioning; the error estimates concern the unmodified spline space, and the matched standard comparison is diagnostic rather than independently tuned.Source: §4.1 · pp. 9–10

4.2 Convergence

  • For C3 quartic splines, the Laplacian rate approaches 3 and L2 rates are consistent with 5; across 25 translations, maximum errors vary little and retain essentially the same rates.The full energy rate reaches about 3.4–3.5 on the last refinements; its strip-supported bulk residual may gain an additional h^1/2 factor for smooth error profiles.Source: §4.2 · pp. 10–11, Figure 3 · p. 11

4.3 Cut Dependence and Coercivity

  • Across 39 cut positions, LS Nitsche stays positive definite with minimum eigenvalues near 0.90, while matched standard systems are indefinite near −0.01.The fixed-method refinement study likewise finds LS Nitsche near 0.90 while coarse-grid-calibrated standard Nitsche becomes indefinite; matrix conditioning still varies with the cut.Source: §4.3 · pp. 11–12, Figure 5 · p. 12, Figure 4 · p. 12

4.4 Parameter Sensitivity

  • The tested parameter choices show trade-offs: τ = 10^-3 gives the smallest L2 error at each mesh size, but very small τ worsens error and conditioning; β = 3 is indefinite.At N = 32, τ = 10^-2, 10^-3, and 10^-6 give L2 errors 1.1 · 10^-8, 5.6 · 10^-9, and 1.0 · 10^-7, respectively. Very large β also reduces accuracy; the tested practical choices are τ ∈ [10^-3, 10^-2] and β = 30, not universal optima.Source: §4.4 · pp. 12–13, Figure 6 · p. 13

4.5 Lower-Degree Splines

  • For C1 quadratic splines, face stabilization raises minimum generalized eigenvalues across 39 cuts to about 0.98, with energy and L2 rates reaching about 1.3 and 2.5.The rates are consistent with guaranteed orders one and two; the coercivity improvement does not uniformly reduce L2 error, and the sufficient conditions certify the finest tested mesh, N = 64.Source: §4.5 · pp. 13–14, Figure 7 · p. 14

5 Conclusions

The paper presents a symmetric least-squares stabilized Nitsche method for the biharmonic problem, with coercivity and error estimates for specified spline spaces. Exact-ellipse experiments support its stability and approximation results, while interface conditions and approximate or nonsmooth boundaries remain directions for further study.

5 Conclusions

  • The explicit lifting of both Dirichlet traces yields coercivity on H^4(Ω) without inverse inequalities and allows fixed penalties independent of the cut configuration and bulk stabilization parameter for sufficiently thin strips.The lifting also determines relative weights for the value, normal-derivative, and tangential penalties.Source: §5 · p. 14
  • For splines of degree p ≥4, the method gives energy and L2 error orders p −1 and p + 1; face stabilization extends the analysis to maximum-continuity splines of degrees p = 2, 3.Source: §5 · p. 14
  • The analysis leaves interface-condition extensions, liftings near corners and edges, and stability and accuracy on boundaries with approximated geometry and curvature as further questions.The paper points to Nitsche coupling on multipatch surfaces in the second-order case [19] and error estimates on approximate domains in the second-order case [20].Source: §5 · p. 14

A The Boundary Lifting

The boundary-lifting analysis uses inward normal coordinates on a sufficiently thin tubular neighborhood of the smooth boundary. It records how the geometry enters the coordinate representation and associated bounds.

A.1 Normal Coordinates

  • The tube width δ0 is chosen below the tubular radius and small enough that δ0∥κ∥L∞(Γ) ≤ 1/2.Source: §A.1 · p. 15
  • The map Φ(y, r) = y − rn(y) parametrizes an inward neighborhood of Γ, with r the inward distance and ∂n = −∂r on Γ.Source: §A.1 · p. 15
  • In the pulled-back coordinates, the Laplacian retains a normal first-derivative term, and smooth curvature gives bounds with constants depending on the fixed geometry.Source: §A.1 · p. 15

A.2 Hermite Lifting and Its Bounds

The Hermite lifting extends the boundary value and normal-derivative traces into a strip while satisfying zero value and normal derivative at its inner edge. Its bounds use product-coordinate estimates and weighted inequalities.

A.2 Hermite Lifting and Its Bounds

  • The lifting matches z = w and ∂n z = ∂n w on Γ and has zero value and normal derivative at r = δ, giving an H2 extension.Source: §A.2 · pp. 15–16
  • The product-coordinate lifting bounds are built from direct integration moments and a geometric remainder estimate on the fixed interval 0 < δ ≤ δ0.Source: §A.2 · p. 16
  • Splitting the lifting into value and normal parts and integrating by parts yields separate factors 3/4 and 15/2, combined by weighted Cauchy–Schwarz into 33/4.Source: §A.2 · p. 16

A.3 Coercivity for Cubic Splines

The appendix establishes coercivity for cubic splines and estimates for quadratic and cubic splines using two-sided normal coordinates and grid trace estimates. The stability argument does not invoke discrete inverse inequalities.

A.3 Coercivity for Cubic Splines

  • For cubic splines, the coercivity proof uses a two-sided tube, product-coordinate lifting bounds, and lattice trace estimates under a sufficiently small-mesh condition.Source: §A.3 · p. 17

B Estimates for Quadratic and Cubic Splines

B.1 Lifting the Normal Trace

For C1 quadratics, the construction retains the value lifting and regularizes only the normal trace, producing a lifting supported in the boundary tube. Product-norm and physical-coordinate estimates control the lifting under the stated parameter bounds.

B.1 Lifting the Normal Trace

  • For C1 quadratics, g0 lies in H2(Γ) and g1 in H1(Γ); the value lifting is retained while only the normal trace is regularized.The constructed lift is extended through the tube and restricted to Ω.Source: §B.1 · p. 17
  • The broken tangential Laplacian penalty is not used in the bound B ≤ B0(w), but remains part of the method.Source: §B.1 · p. 17
  • Table 2 gives product-norm bounds for the value and normal parts, with its columns multiplying P0 and P1, respectively.Parity removes cross terms, and the moments provide the bounds used in the table.Source: Table 2 · p. 18, §B.1 · p. 18
  • The lifting converges in H2, has the prescribed boundary traces, and is supported in Dδ.The profiles are extended by zero beyond unit argument.Source: §B.1 · pp. 17–18

B.2 Physical Norms

  • For θ, σ ≤ 1/10, the physical-coordinate estimates bound the Jacobian and Hessian remainders, with no rr component in the Hessian remainder.The second-derivative Laplacian remainder acts only on the tangential Hessian.Source: §B.2 · p. 18

B.3 Face Traces and Coercivity

The face-trace argument uses matching-trace identities and cellwise Green’s formula to control the terms entering the coercivity estimate. Density and Young’s inequality complete the stated claim.

B.3 Face Traces and Coercivity

  • Matching traces yield the stated jumps of the Laplacian and its normal derivative, which are used with cellwise Green’s formula.The formula takes the face normal outward from T+.Source: §B.3 · p. 19
  • Density extends the identity to the H2 lifting, and the lattice trace estimate bounds coordinate derivatives on physical faces.Restriction to physical faces decreases the sums in the trace estimate.Source: §B.3 · p. 19
  • Young’s inequality with parameter 4/5 leaves coefficients 1/5 and β − 29 ≥ 1 for the respective terms, proving the claim.Source: §B.3 · p. 19

B.4 Error Estimates

The error analysis combines continuity and coercivity with spline approximation and a duality argument. For quadratics and cubics, the dual approximation contributes one and two powers of h, respectively.

B.4 Error Estimates

  • Consistency bounds give continuity and coercivity of Ah,j on H4(Ω) + Vh, after which consistency and Galerkin orthogonality yield the error estimate.Source: §B.4 · p. 19
  • The extension and quasi-interpolant have approximation order p + 1; for p = 2, the fourth-derivative estimate scales as h^-1 on full cells.The quadratic fourth-derivative bound uses local quadratic reproduction and scaling.Source: §B.4 · p. 19
  • The stabilization weights h2p−4 and h2p−2 control derivatives of orders p and p + 1 through bounded overlap and extension stability.Source: §B.4 · p. 19
  • The dual approximation supplies one additional power of h for quadratics and two for cubics in the L2 error argument.The argument uses a dual solution in H4(Ω) ∩ H2_0(Ω).Source: §B.4 · p. 19
Loading 2610.11636v1…