Stability, instability, and error of the force-based quasicontinuum approximation (Q993331)

The paper provides a detailed stability and error analysis of the force-based quasicontinuum (QCF) approximation, clarifying the relations between different approximations and the corresponding sources of error. The QCF approximation serves for coupling a general atomistic model with a finite element continuum model. In the present paper, it is proved that typically, the linearization of the QCF equilibrium equations is not coercive (i.e., positive-definite), and consequently, the error analysis is based on a more general inf-sup stability condition. The presentation is restricted to a one-dimensional chain of atoms which interact with nearest and next-nearest neighbors. Moreover, a linearization of force-based equilibrium equations is considered about a uniform strain, and the modeling error is discussed only due to the QCF approximation itself, but the authors do not consider the coarsening error. First, the paper presents a detailed description of the QCF approximation, and gives a first estimate of the truncation error with respect to the fully atomistic model. In order to combine the accuracy of the atomistic model with the efficiency of the local QC approximation, the QCF method decomposes the reference lattice into atomistic and continuum regions, assigning forces to atoms according to the region where they are located. Then, the QCF equilibrium equations are formulated in a ``weak form'' with studying its stability by considering discrete versions of duality pairing. This is equivalent to putting the QCF operator into a difference form, which indicates an interesting nonlocal effect of the atomistic-to-continuum interface. This nonlocal effect is the source of the coercivity lack, based on an explicit construction of unstable displacement. Due to this, the authors discuss inf-sup stability results that are then combined with negative-norm truncation error estimates to obtain optimal order error estimates for the QCF approximation. The main result of the paper is that the strain error for the QCF approximation is \(O(\epsilon^2)\), where \(\epsilon\) is the lattice spacing scaled by the material dimension. The prefactor for the \(\epsilon^2\) error term is a maximum norm of a third divided difference of the displacement restricted to the continuum region only. Therefore, the paper analysis predicts the observed high accuracy of the QCF method when defects are modeled in the atomistic region. Then, the lack of a uniform inf-sup constant is shown for many other common choices of duality pairing. In conclusion, the authors discuss some extensions of the present analysis to further establish the mathematical foundations of the quasicontinuum method.