[Paper Review] Iterative Methods for the Force-based Quasicontinuum Approximation

Matthew Dobson, Mitchell Luskin+1 more|arXiv (Cornell University)|Oct 12, 2009

Nonlocal and gradient elasticity in micro/nano structures28 references22 citations

TL;DR

This paper proposes a preconditioned GMRES method for solving the linearized force-based quasicontinuum (QCF) equations, which are non-symmetric and indefinite due to the non-conservative nature of the QCF approximation. The method uses the QCL preconditioner and a specialized inner product, achieving stable and reliable convergence up to the critical strain, with the residual serving as an effective error predictor.

ABSTRACT

Force-based atomistic-continuum hybrid methods are the only known pointwise consistent methods for coupling a general atomistic model to a finite element continuum model. For this reason, and due to their algorithmic simplicity, force-based coupling methods have become a popular class of atomistic-continuum hybrid models as well as other types of multiphysics models. However, the recently discovered unusual stability properties of the linearized force-based quasicontinuum (QCF) approximation, especially its indefiniteness, present a challenge to the development of efficient and reliable iterative methods. We present analytic and computational results for the generalized minimal residual (GMRES) solution of the linearized QCF equilibrium equations. We show that the GMRES method accurately reproduces the stability of the force-based approximation and conclude that an appropriately preconditioned GMRES method results in a reliable and efficient solution method.

Motivation & Objective

• Address the numerical instability of existing iterative methods—such as modified nonlinear conjugate gradient and ghost force correction (GFC)—for solving the linearized force-based quasicontinuum (QCF) equations.
• Overcome the challenge posed by the indefiniteness and non-symmetry of the linearized QCF operator, which undermines the convergence and stability of standard iterative solvers.
• Develop a reliable and efficient iterative solution method for the QCF system that accurately captures the stability properties of the underlying atomistic model up to the critical strain.
• Ensure the residual serves as a reliable error indicator, enabling effective convergence monitoring in practical simulations.

Proposed method

• Apply the generalized minimal residual (GMRES) method to the linearized QCF system, leveraging its robustness for non-symmetric and indefinite linear systems.
• Use the QCL (quasicontinuum with local corrections) method as a preconditioner to improve convergence and stability of the GMRES iterations.
• Incorporate a specialized $υ^{1,2}$-inner product (based on the QCL operator) as the underlying inner product in GMRES to enhance conditioning and convergence behavior.
• Construct an eigenbasis for the generalized eigenvalue problem $ L^{-1}L^{\text{qcf}}_F $ using block structure and deflation techniques to handle high multiplicity eigenvalues, ensuring numerical stability in spectral analysis.
• Implement and validate the method on a 1D model problem with varying system sizes $ N $, refinement levels $ K $, and force parameters $ f $, using both residual and error norms.
• Use the $ \mathcal{U}^{-1,2} $-norm of the residual and $ \mathcal{U}^{1,2} $-norm of the error to quantitatively assess convergence and reliability.

Experimental results

Research questions

• RQ1 Can standard iterative solvers like modified nonlinear conjugate gradient or ghost force correction (GFC) reliably solve the linearized QCF system without numerical instability?
• RQ2 Does the GMRES method, when properly preconditioned and equipped with a suitable inner product, overcome the indefiniteness and non-symmetry of the QCF operator to ensure stable convergence?
• RQ3 Can the residual in the GMRES iteration serve as a reliable predictor of the true error in the solution of the QCF system?
• RQ4 What is the convergence rate of the preconditioned GMRES method, and does it match theoretical predictions based on the eigenvalue distribution of the system?
• RQ5 How does the method perform up to the critical strain, where the QCF system becomes unstable, and does it preserve the correct stability characteristics of the atomistic model?

Key findings

• The modified nonlinear conjugate gradient method fails numerically due to instability caused by the indefiniteness of the linearized QCF operator, even before reaching the critical strain.
• The ghost force correction (GFC) method, which uses the QCE energy as a preconditioner, becomes unstable prior to the critical strain, leading to an incorrect prediction of reduced critical strain for defect formation.
• The proposed preconditioned GMRES method with the QCL preconditioner and $ \mathcal{U}^{1,2} $-inner product achieves stable convergence up to the critical strain, preserving the correct stability behavior of the system.
• The residual norm $ \|r^{(m)}\|_{\mathcal{U}^{-1,2}} $ decays linearly with rate $ q = \frac{1 - \sqrt{A_F / \phi_F''}}{1 + \sqrt{A_F / \phi_F''}} $, matching theoretical predictions from Proposition 5.3.
• The error norm $ \|e^{(m)}\|_{\mathcal{U}^{1,2}} $ closely tracks the residual norm $ \|L^{-1/2}r^{(m)}\|_{\ell^2_\varepsilon} $, confirming that the residual is a reliable convergence indicator.
• The method remains robust across different system sizes $ N $, refinement levels $ K $, and right-hand side vectors $ f $, demonstrating scalability and reliability in the 1D model problem.