The Calcium Silicate Hydrate (C-S-H) nucleation is a crucial step during cement hydration and determines to a great extent the rheology, microstructure, and properties of the cement paste. Recent evidence indicates that the C-S-H nucleation involves at least two steps, yet the underlying atomic scale mechanism, the nature of the primary particles and their stability, or how they merge/aggregate to form larger structures is unknown. In this work, we use atomistic simulation methods, specifically DFT, evolutionary algorithms (EA), and Molecular Dynamics (MD), to investigate the structure and formation of C-S-H primary particles (PPs) from the ions in solution, and then discuss a possible formation pathway for the C-S-H nucleation. Our simulations indicate that even for small sizes the most stable clusters encode C-S-H structural motifs, and we identified a C4S4H2 cluster candidate to be the C-S-H basic building block. We suggest a formation path in which small clusters formed by silicate dimers merge into large elongated aggregates. Upon dehydration, the C-S-H basic building blocks can be formed within the aggregates, and eventually crystallize.
In this work, we present ænet-PyTorch, a PyTorch-based implementation for training artificial neural network-based machine learning interatomic potentials. Developed as an extension of the atomic energy network (ænet), ænet-PyTorch provides access to all the tools included in ænet for the application and usage of the potentials. The package has been designed as an alternative to the internal training capabilities of ænet, leveraging the power of graphic processing units to facilitate direct training on forces in addition to energies. This leads to a substantial reduction of the training time by one to two orders of magnitude compared to the central processing unit implementation, enabling direct training on forces for systems beyond small molecules. Here, we demonstrate the main features of ænet-PyTorch and show its performance on open databases. Our results show that training on all the force information within a dataset is not necessary, and including between 10% and 20% of the force information is sufficient to achieve optimally accurate interatomic potentials with the least computational resources.
Computational methods, or computer-aided material design (CAMD), used for the analysis and design of materials have a relatively long history. However, the applicability of CAMD has been limited by the scales of computational resources generally available in the past. The surge in computational power seen in recent years is enabling the applicability of CAMD to unprecedented levels. Here, we focus on the CAMD for materials critical for the continued advancement of the complementary metal oxide semiconductor (CMOS) semiconductor technology. In particular, we apply CAMD to the engineering of high-permittivity dielectric materials. We developed a Reax forcefield that includes Si, O, Zr, and H. We used this forcefield in a series of simulations to compute the static dielectric constant of silica glasses for low Zr concentration using a classical molecular dynamics approach. Our results are compared against experimental values. Not only does our work reveal numerical estimations on ZrO2-doped silica dielectrics, it also provides a foundation and demonstration of how CAMD can enable the engineering of materials of critical importance for advanced CMOS technology nodes.
22 August 2024 in Journal of Chemical Theory and Computation
Machine learning potentials (MLPs) offer efficient and accurate material simulations, but constructing the reference ab initio database remains a significant challenge, particularly for catalyst-adsorbate systems. Training an MLP with a small data set can lead to overfitting, thus limiting its practical applications. This study explores the feasibility of developing computationally cost-effective and accurate MLPs for catalyst-adsorbate systems with a limited number of ab initio references by leveraging a transfer learning strategy from subsets of a comprehensive public database. Using the Open Catalyst Project 2020 (OC20)-a data set closely related to our system of interest-we pretrained MLP models on OC20 subsets using the ænet-PyTorch framework. We compared several strategies for database subset selection. Our findings indicate that MLPs constructed via transfer learning exhibit better generalizability than those constructed from scratch, as demonstrated by the consistency in the dynamics simulations. Remarkably, transfer learning enhances the stability and accuracy of MLPs for the CuAu/H2O system with approximately 600 reference data points. This approach achieved excellent extrapolation performance in molecular dynamics simulations for the larger CuAu/6H2O system, maintaining stable and accurate predictions for up to 250 ps, whereas MLPs without transfer learning become unstable before reaching 50 ps. We also examine the potential limitations of this strategy. This work proposes an alternative, cost-effective approach for constructing MLPs for the challenging simulation of catalytic systems. Finally, we anticipate that this methodology will pave the way for broader applications in materials science and catalysis research, facilitating more efficient and accurate simulations across various systems.