Critical assessment of machine-learned repulsive potentials for the density functional based tight-binding method: A case study for pure silicon.

We investigate the feasibility of improving the semi-empirical density functional based tight-binding method through a general and transferable many-body repulsive potential for pure silicon using a common machine-learning framework. Atomic environments using atom centered symmetry functions fed into flexible neural-networks allow us to overcome the limited pair potentials used until now with the ability to train simultaneously on a large variety of systems. We achieve an improvement on bulk systems with good performance on energetic, vibrational, and structural properties. Contrarily, there are difficulties for clusters due to surface effects. To deepen the discussion, we also put these results into perspective with two fully machine-learned numerical potentials for silicon from the literature. This allows us to identify both the transferability of such approaches together with the impact of narrowing the role of machine-learning models to reproduce only a part of the total energy.

Shear transformation distribution and activation in glasses at the atomic scale.

We characterize shear transformations (STs) at the atomic scale in a model of amorphous silicon using a mapping on Eshelby inclusions. We investigate the effect of pressure, glass relaxation, as well as damage on the ST characteristics. We show that the characteristic ST effective volume, γ_{0}V_{0}, product of the ST plastic shear strain γ_{0} and volume V_{0}, does not depend significantly on an applied pressure but increases with accumulated plastic deformation from about 10Å^{3} in the pseudoelastic regime to about 60Å^{3} once plastic flow sets in. Furthermore, by using nudged elastic band calculations, we measure the energy barrier against ST activation. Analyzing different paths leading to either an isolated ST or an avalanche, we show that the barrier is systematically controlled by the first ST with an activation volume equal to the effective volume of the ST at the activated state, which represents only a fraction of the complete ST volume. The activation volume is also found smaller for avalanches, presumably because of accumulated local damage. This work provides essential information to build reliable mesoscale models of plasticity.

Mapping between atomistic simulations and Eshelby inclusions in the shear deformation of an amorphous silicon model.

In this paper we perform quasistatic shear simulations of model amorphous silicon bulk samples with Stillinger-Weber-type potentials. Local plastic rearrangements identified based on local energy variations are fitted through their displacement fields on collections of Eshelby spherical inclusions, allowing determination of their transformation strain tensors. The latter are then used to quantitatively reproduce atomistic stress-strain curves, in terms of both shear and pressure components. We demonstrate that our methodology is able to capture the plastic behavior predicted by different Stillinger-Weber potentials, in particular, their different shear tension coupling. These calculations justify the decomposition of plasticity into shear transformations used so far in mesoscale models and provide atomic-scale parameters that can be used to limit the empiricism needed in such models up to now.

Rheological properties vs. local dynamics in model disordered materials at low temperature.

We study the rheological response at low temperature of a sheared model disordered material as a function of the bond rigidity. We find that the flow curves follow a Herschel-Bulkley law, whatever is the bond rigidity, with an exponent close to 0.5. Interestingly, the apparent viscosity can be related to a single relevant time scale t rel, suggesting a strong connection between the local dynamics and the global mechanical behaviour. We propose a model based on the competition between the nucleation and the avalanche-like propagation of spatial strain heterogeneities. This model can explain the Herschel-Bulkley exponent on the basis of the size dependence of the heterogeneities on the shear rate.

A new parametrization of the Stillinger-Weber potential for an improved description of defects and plasticity of silicon.

A new parametrization of the widely used Stillinger-Weber potential is proposed for silicon, allowing for an improved modelling of defects and plasticity-related properties. The performance of the new potential is compared to the original version, as well as to another parametrization (Vink et al 2001 J. Non-Cryst. Solids, 282 248), in the case of several situations: point defects and dislocation core stability, threshold displacement energies, bulk shear, generalized stacking fault energy surfaces, fracture, melting temperature, amorphous structure, and crystalline phase stability. A significant improvement is obtained in the case of dislocation cores, bulk behaviour under high shear stress, the amorphous structure, and computation of threshold displacement energies, while most of the features of the original version (elastic constants, point defects) are retained. However, despite a slight improvement, a complex process like fracture remains difficult to model.

Role of local order in the small-scale plasticity of model amorphous materials.

The mechanism of plastic flow in amorphous solids involves nucleation-controlled shear transformations, triggered under stress from fertile sites. However, the origin of these sites is still a matter of debate. In this paper, we show that the connection between local plastic activity and coordination defects in amorphous systems depends on the nature of the interatomic interactions. In particular, the directionality of the bonds, as quantified by the three-body term in Stillinger-Weber-like interactions, affects not only the role of local defects, but also the size of the plastic rearrangements, and the global stress-strain behavior. We study the effect of structure changes due to different quenching rates as well. We conclude the paper by a comparison between amorphous plasticity and the Peierls-Nabarro theory of plasticity in crystals.

"Learn on the fly": a hybrid classical and quantum-mechanical molecular dynamics simulation.

We describe and test a novel molecular dynamics method which combines quantum-mechanical embedding and classical force model optimization into a unified scheme free of the boundary region, and the transferability problems which these techniques, taken separately, involve. The scheme is based on the idea of augmenting a unique, simple parametrized force model by incorporating in it, at run time, the quantum-mechanical information necessary to ensure accurate trajectories. The scheme is tested on a number of silicon systems composed of up to approximately 200 000 atoms.