Search for: All records

Despite growing interest in polymers under extreme conditions, most atomistic molecular dynamics simulations cannot describe the bond scission events underlying failure modes in polymer networks undergoing large strains. In this work, we propose a physics-based machine learning approach that can detect and perform bond breaking with near quantum-chemical accuracy on-the-fly in atomistic simulations. Particularly, we demonstrate that by coarse-graining highly correlated neighboring bonds, the prediction accuracy can be dramatically improved. By comparing with existing quantum mechanics/molecular mechanics methods, our approach is approximately two orders of magnitude more efficient and exhibits improved sensitivity toward rare bond breaking events at low strain. The proposed bond breaking molecular dynamics scheme enables fast and accurate modeling of strain hardening and material failure in polymer networks and can accelerate the design of polymeric materials under extreme conditions.

 

Finding the reduced-dimensional structure is critical to understanding complex networks. Existing approaches such as spectral clustering are applicable only when the full network is explicitly observed. In this paper, we focus on the online factorization and partition of implicit large lumpable networks based on observations from an associated random walk. We formulate this into a nonconvex stochastic factorization problem and propose an efficient and scalable stochastic generalized Hebbian algorithm (GHA). The algorithm is able to process random walk data in a streaming fashion and learn a low-dimensional representation for each vertex. By applying a diffusion approximation analysis, we show that the continuous-time limiting process of the stochastic algorithm converges globally to the “principal components” of the Markov chain. We also establish a finite-sample error bound that matches the nonimprovable state-of-art result for online factorization. Once learned the low-dimensional state representations, we further apply clustering techniques to recover the network partition. We show that when the associated Markov process is lumpable, one can recover the partition exactly with high probability given sufficient data. We apply the proposed approach to model the traffic flow of Manhattan as city-wide random walks. By using our algorithm to analyze the taxi trip data, we discover a latent partition of the Manhattan city that closely matches the traffic dynamics. 

Phyllosilicate minerals, due to their sheets structure and morphology, are known to cause anisotropy in bulk rock properties and make the bulk rock more compliant. Accurately characterizing the micromechanical behavior of phyllosilicate minerals from laboratory observations, which eventually translates to the bulk rock behavior, is still challenging due to their fine‐grained nature. Recent advances in atomistic simulations open the possibility of theoretically investigating such mineral mechanical behavior. We compare the elastic properties of biotites recovered by spherical nanoindentation with those predicted from density functional theory (DFT) simulations to investigate to what extent theoretical predictions reproduce actual phyllosilicate properties. Spherical nanoindentation was conducted using schist rocks from Poorman Formation, South Dakota, USA, to recover continuous indentation stress‐strain curves. Loading in the layer‐normal orientation shows an average indentation modulus () of about 35 GPa, while loading in the layer‐parallel orientation gives a higher average of about 95 GPa. To facilitate comparison, the elastic stiffness constants (c_ij) determined from DFT were converted to indentation modulus () using solutions proposed in this study. The majority of the nanoindentation modulus results are below the values inferred from the simulation results representing ideal defect‐free minerals. We suggest that crystal defects present at the nano‐scale, potentially ripplocations, are the dominant cause of the lower indentation modulus recovered from nanoindentation compared to those inferred from DFT simulations. Results highlight the importance of acknowledging the defects that exist down to the nano‐scale as it modifies the mechanical properties of phyllosilicates compared to its pure defect‐free form.