A DPG method for the circular arch problem

This paper presents and analyzes a Discontinuous Petrov-Galerkin (DPG) method with optimal test functions for an elastic circular arch model, demonstrating through theoretical analysis and numerical experiments that the approach achieves optimal convergence rates and can be further improved by employing a suitably scaled test space norm.

The Gist

Imagine you are an engineer tasked with designing a beautiful, curved stone bridge (an arch). You need to know exactly how it will bend, stretch, and twist under the weight of people walking across it. If your math is slightly off, the bridge could be unsafe or over-engineered (wasting money).

This paper is about a new, super-precise way of doing the math to predict how these curved bridges behave. The authors, Norbert Heuer and Antti Niemi, are introducing a method called DPG (Discontinuous Petrov–Galerkin).

Here is the breakdown of their work using simple analogies:

1. The Problem: The "Curved" Puzzle

Most standard math tools for building bridges work great for straight beams (like a ruler). But when you curve that ruler into an arch, the math gets messy.

• The Locking Effect: Imagine trying to slide a rigid ruler through a curved tube. If the math isn't perfect, the computer thinks the ruler is "stuck" or "locked" and can't move, even though it should. This leads to wrong answers.
• The Curvature Trap: The deeper the arch (the more curved it is), the harder it is for standard math to handle the "twist" caused by the curve. Sometimes, the error gets amplified, making the deep arch look like it's about to collapse when it's actually fine.

2. The Solution: The "Ultra-Weak" DPG Method

The authors propose a new strategy called DPG. Think of it like this:

• Standard Method (The Rigid Team): Imagine a team of workers trying to measure a curved wall. They all stand in a straight line and try to force their measurements to fit the curve. If the curve is too sharp, they get confused and make mistakes.
• The DPG Method (The Flexible Team): This method breaks the wall into small, independent puzzle pieces (elements).
  • Discontinuous: The pieces don't have to perfectly glue together at the seams immediately. They are allowed to have tiny gaps or jumps.
  • Ultra-Weak: Instead of forcing the math to be perfect everywhere, it focuses on the "balance of forces" at the edges of these pieces.
  • Optimal Test Functions: This is the secret sauce. Usually, when you check your math, you use a standard ruler. Here, the authors create a custom-made, shape-shifting ruler for every single puzzle piece. This ruler is perfectly shaped to catch the specific errors of that piece. It's like having a detective who knows exactly where the criminal (the error) is hiding.

3. The "Deep Arch" Problem

The paper discovered something interesting:

• If the arch is shallow (like a gentle hill), the math works fine.
• If the arch is deep (like a rainbow), the standard way of measuring error can get "scared" by the curve. The math starts to overreact, predicting huge errors where there are none.

The Fix: The authors found that by scaling their custom rulers (the test space norm), they could calm down the math. It's like turning down the volume on a microphone that is too sensitive to wind noise. By adjusting the "sensitivity" of their math tool based on how curved the arch is, they get a stable, accurate result every time.

4. The Results: Proof in the Pudding

They tested this on two scenarios:

1. A Cantilever Arch: Like a diving board that is curved. They compared their new method against the current "best" method.
   • Result: Their method was just as good at predicting movement, but it was much better at predicting the internal "stress" (the pressure inside the material), which is crucial for safety.
2. A Fully Clamped Arch: A bridge fixed tightly at both ends.
   • Result: When they used the "scaled" rulers, the errors dropped significantly, especially for very thin, deep arches. The standard method struggled, but the DPG method remained steady.

The Big Takeaway

This paper is like giving engineers a super-powered, adaptive toolkit for building curved structures.

• It admits that curved structures are tricky.
• It breaks the problem into small, manageable chunks.
• It uses a "smart" checking system that adapts to the shape of the curve.
• Most importantly, it prevents the math from panicking when the arch gets very deep or very thin, ensuring that the final design is both safe and efficient.

In short: They figured out how to stop the math from getting confused by curves, ensuring that our bridges and arches are designed with maximum precision.

Technical Summary

1. Problem Statement

The paper addresses the numerical analysis of thin circular arches within the framework of computational mechanics. The physical model incorporates three primary mechanical effects:

• Membrane (axial) effects: Represented by axial force $n$.
• Transverse shear effects: Represented by shear force $q$.
• Bending effects: Represented by bending moment $m$.

The governing equations are a system of first-order equilibrium and constitutive relations involving axial displacement ($u$), transverse displacement ($w$), and rotation ($\theta$). The system is characterized by three non-dimensional parameters:

• $\epsilon$ (Slenderness): $\epsilon^2 = I/(AL^2)$. Small $\epsilon$ indicates a thin arch, leading to potential "locking" phenomena (shear and membrane locking) in standard finite element methods.
• $\lambda$ (Curvature/Depth): $\lambda = L/R$. This parameter dictates the structural regime (deep arch, shallow arch, or straight beam limit).
• $\mu$ (Material Anisotropy): $\mu = E/(\kappa G)$. Relates axial stiffness to shear stiffness. The limit $\mu \to 0$ corresponds to the Euler–Bernoulli beam theory (infinite shear stiffness).

The challenge lies in developing a numerical method that is robust (stable) with respect to small $\epsilon$ (thin limit) and handles the complex dependence on curvature $\lambda$ without suffering from numerical locking or error amplification.

2. Methodology: Discontinuous Petrov–Galerkin (DPG)

The authors propose an ultra-weak variational formulation based on the Discontinuous Petrov–Galerkin (DPG) method.

• Ultra-Weak Formulation: The first-order system is tested with discontinuous test functions. Integration by parts is performed element-wise, moving all derivatives onto the test functions. This results in a formulation where the trial space contains discontinuous stress and displacement interpolations, and continuity is enforced weakly via interface variables (traces) defined at the nodes.
• Optimal Test Functions: The core of the DPG method is the use of "optimal test functions." A trial-to-test operator $T: U \to V$ is defined such that the test functions are chosen to maximize the stability of the discrete system. This ensures that the discrete problem inherits the stability properties of the continuous problem.
• Trace Norms: The formulation introduces specific trace norms ($\|\cdot\|_\gamma$) for the interface variables. These norms are crucial for establishing the well-posedness of the problem and are shown to be uniformly equivalent to the dual norm of the trace operator.
• Stability Analysis: The authors utilize the Babuška–Brezzi theory for ultra-weak formulations. They prove well-posedness by:
  1. Establishing a lower bound for the trace operator.
  2. Proving the well-posedness of the continuous adjoint problem (a mixed formulation).
  3. Deriving a stability constant ($C_{stab}$) that depends on the curvature $\lambda$ and the parameters $\epsilon, \mu$.

3. Key Contributions

• Extension of DPG to Circular Arches: This work extends previous DPG methodologies (previously applied to straight beams, plates, and shells) to the more complex geometry of circular arches, which introduces curvature-dependent coupling between membrane and bending modes.
• Rigorous Stability Analysis: The paper provides a detailed theoretical analysis revealing that the stability constant depends intricately on the curvature parameter $\lambda$. Specifically, for certain boundary conditions (e.g., sliding supports), the stability constant can grow as $|\cos \lambda|^{-1}$, potentially leading to error amplification for deep arches.
• Scaled Test Space Norm: A significant contribution is the proposal of a parameter-dependent scaled graph norm for the test space. The standard $H^1$ norm can lead to error growth in deep arch regimes. By scaling the test norm with parameters related to $\epsilon$ and $\lambda$, the authors demonstrate that this error amplification can be effectively mitigated, ensuring robustness across different structural regimes.
• Handling of Locking: The ultra-weak formulation naturally avoids the numerical locking effects (shear and membrane locking) that plague standard variational formulations for thin structures, without requiring reduced integration or mixed formulations with independent stress variables in the traditional sense.

4. Results

• Theoretical Convergence: The analysis proves that the DPG scheme is well-posed and converges quasi-optimally. The error bound is proportional to the best approximation error in the trial space, scaled by the stability constant $C_{stab}$.
• Numerical Experiments:
  • Cantilever Arch: Tested with a point load. The DPG method showed quadratic convergence rates consistent with the polynomial degree of the trial space. It performed comparably to a state-of-the-art reduced-strain Finite Element Method (FEM) but with the advantage of providing stress approximations of the same accuracy as displacements.
  • Fully Clamped Arch: Tested under distributed loads with varying curvature ($\lambda=3$) and slenderness ($\epsilon = 10^{-1}, 10^{-3}$).
    • Standard Norm: Showed elevated error levels for deep arches, confirming the theoretical prediction of error amplification due to curvature.
    • Scaled Graph Norm: When the proposed scaled test space norm was employed, the error levels were significantly reduced, and the method maintained robust quadratic convergence across all tested values of $\epsilon$ and $\lambda$.
• Enrichment: The study noted that while an enrichment degree of $\Delta p = 2$ (test functions degree $p+2$) is sufficient for stability in the standard norm, a higher enrichment ($\Delta p = 4$) is required when using the scaled graph norm to avoid parametric locking in local problems.

5. Significance

This paper makes a substantial contribution to the field of computational mechanics by:

1. Solving the Locking Problem: It demonstrates that the DPG framework is a powerful alternative to traditional mixed FEM for thin arches, naturally circumventing locking issues without ad-hoc stabilization techniques like reduced integration.
2. Robustness in Deep Arches: It identifies and solves a specific stability issue related to high curvature. The introduction of the scaled test space norm provides a practical recipe for ensuring numerical stability in deep arch problems where standard methods might fail or degrade.
3. Unified Framework: It offers a unified approach for analyzing arches ranging from shallow to deep, and from thick to extremely thin (Euler–Bernoulli limit), with rigorous error estimates that account for all physical parameters.

In summary, the authors successfully adapt the DPG method to circular arches, proving its theoretical robustness and demonstrating through numerical experiments that a carefully scaled test norm is essential for maintaining accuracy in deep arch configurations.