Atomistic-Continuum Coupling by Homogenization

Technical Summary: Atomistic-Continuum Coupling by Homogenization

Problem Statement
Classical atomistic simulations based on interatomic potentials offer high-fidelity resolution of lattice instabilities, defect nucleation, and microstructure evolution. However, their accessible system sizes are restricted to the nanoscale, making them unsuitable for analyzing micrometer-scale structural components and their boundary conditions. While concurrent multiscale methods (e.g., quasicontinuum) couple atomistic and continuum domains spatially, they often suffer from algorithmic challenges such as spurious interface forces ("ghost forces") and difficulties in maintaining global energy consistency. Hierarchical methods, which rely on homogenization, avoid direct spatial interfaces but traditionally require a continuum constitutive law at the microscale. The challenge addressed in this work is to establish a hierarchical coupling where the microscale is governed strictly by discrete atomistic physics (interatomic potentials) rather than a phenomenological continuum law, while maintaining energetic consistency and robustness for nonlinear structural analysis.

Methodology
The authors propose a two-scale atomistic–continuum framework based on computational homogenization and the Hill–Mandel energy equivalence principle. The approach couples a nonlinear finite-element (FE) boundary-value problem at the microscale (structural level) with periodic molecular-statics (MS) cell problems at the quadrature points of the FE mesh.

• Scale Transition: The continuum deformation gradient ($F$) drives the atomistic representative volume element (RVE). Periodic boundary conditions (PBC) are applied to the atomistic cell. The atomistic solver performs a zero-temperature energy minimization to find the equilibrium atomic configuration for the prescribed deformation.
• Stress and Tangent Transfer: Unlike standard FE$^2$ methods that rely on a stored energy function, the atomistic cell returns volume-averaged Cauchy stresses derived from the virial theorem. These are transformed into work-conjugate first Piola–Kirchhoff stresses ($P$) and second Piola–Kirchhoff stresses ($S$) required by the Lagrangian FE formulation.
• Algorithmic Tangent: To ensure robust Newton iterations on the coarse scale, an effective tangent modulus is computed. This is achieved via numerical differentiation (forward difference) of the stress response with respect to the deformation gradient. This involves perturbing the deformation gradient, performing independent MS relaxations for each perturbation, and calculating the resulting stress increments.
• Implementation: The framework integrates LAMMPS for the MS solver within a scalable MPI-based workflow. Parallelism is exploited across quadrature points, where each Gauss point triggers an independent set of atomistic simulations (one for the base state and six for the tangent perturbations).

Key Contributions

1. Energetically Consistent Framework: The authors establish a nano-to-micro scale transition strictly based on the Hill–Mandel postulate, ensuring energy equivalence between the macroscopic virtual power and the volume-averaged microscopic virtual power.
2. Discrete-to-Continuum Interface: A specific interface is defined where the continuum deformation drives a periodic atomistic RVE, and the atomistic solver returns volume-averaged stresses in work-conjugate form along with effective tangent moduli.
3. Robustness in Non-Smooth Regimes: The work demonstrates that a static-implicit Newton solver can remain robust even when the atomistic response is non-smooth due to defect nucleation and instability. The method successfully recovers near-quadratic convergence in the final iterations of critical load steps.
4. Practical Implementation: The study outlines a strategy for integrating established simulation blocks (LAMMPS) into a scalable workflow, handling the computational cost of multiple concurrent atomistic solves per macroscopic iteration.

Numerical Results
The framework was tested on single-crystalline copper (EAM potential) in several scenarios:

• Uniaxial Tension and Compression: The model captured pronounced tension–compression asymmetry. In tension, instability led to the formation of stacking faults and twin-related lamellae. In compression, the microstructure evolved into broad, layered bundles of stacking faults and hexagonal close-packed (hcp) regions. The results showed that the onset of instability depends on the simulation box size, while the pre-instability elastic response is size-independent.
• Cyclic Loading: Under reversed cyclic loading, the system exhibited a transition from an initial high-force elastic regime to a state governed by a "trained" metastable defect architecture. This resulted in broad hysteresis loops and rapid cyclic stabilization after the first instability event.
• Cantilever Bending: A micrometer-scale beam (10 µm length) was simulated. The framework resolved a localized elastic–plastic transition where defect nucleation occurred abruptly in nanodomains near the clamped end (bottom fibers) while the rest of the beam remained elastic. The Newton solver handled the sharp stiffness changes associated with these localized instabilities effectively.

Significance and Claims
The paper claims to provide the first atomistic-to-continuum upscaling that is simultaneously: (i) based on the strict energetic principle of Hill–Mandel, (ii) compatible with a static-implicit finite element formulation, and (iii) equipped with an effective tangent suitable for Newton solvers in strongly nonlinear regimes.

The authors emphasize that their approach extends potential-based atomistic modeling to structural length scales (micrometers) that are inaccessible to direct atomistic simulation in a quasi-static, athermal setting. A critical finding is that the coarse-scale Newton solver remains robust and recovers near-quadratic convergence even when the fine-scale response undergoes abrupt stiffness changes due to defect nucleation. This suggests that energetically consistent stress and tangent information from atomistic cells remain useful tools for structural analysis even in highly nonlinear, unstable scenarios.

The work is explicitly restricted to quasi-static, athermal responses at zero temperature. The authors note that future extensions could involve finite-temperature models, alternative boundary conditions, and the integration of machine-learning interatomic potentials.