Formidable progress has been recently achieved in the fabrication and characterization of disordered low dimensional materials with unprecedented properties. For instance, particular forms of disordered graphene (reduced graphene oxides) and other two-dimensional materials, obtained by chemical exfoliation techniques, have been found suitable to improve the performances of composite materials, with applications in energy, textiles, solar cells, thermoplastics etc. In parallel, catalytic vapor deposition methods have been used intensively to upscale the growth of two-dimensional materials at wafer size, with resulting material morphologies ranging from polycrystalline to fully amorphous phases. Recent reports on wafer-scale synthesis of ultrathin layered amorphous carbon and boron nitride, structurally dominated by sp2 hybridization but lacking crystalline orders, have shown superior performances of those materials as coating and interfacial components, providing a new platform for properties reinforcement in the industrial context.

Here, we present our computational strategies to achieve predictive modelling in this context. For the “computationally-assisted” material synthesis, we use machine learning to generate atomic interaction potential in an arbitrary disordered local environment and obtain realistic large-scale material configurations through classical molecular dynamics, while generic tight-binding models are generated based on resulting spatial atomic distribution and optimized DFT-inspired Hamiltonian parameters. Next, we employ efficient linear scaling quantum transport [1] (see also www.lsquant.org) to simulate structural models containing up to a billion atoms, hence demonstrating unprecedented supremacy to match predictive modelling with experimental scales, and therefore constituting decision-making tools for material design and industrial performances achievement. We illustrate our strategy on various examples including the (thermal) reduction process of graphene-oxides for recovering higher crystallinity, the correlation between the degree of disorder and resulting physical properties in polycrystalline materials [2], as well as amorphous phases of sp2 carbon members [3] and boron-nitride [4]. The obtained results provide scaling laws and guiding rules for researchers and engineers to target the proper material growth and integration conditions for reaching the upper limit of a given application objective, such as flexible conductors, spin channel, ultrathin heat spreaders, energy harvesters, insulating thermal barriers, or ultralow dielectric coefficient materials.
[1] Z. Fan et al. S. Roche, Physics Reports 903, 1-69 (2021)
[2] AW Cummings et al. Advanced Materials 26 (30), 5079-5094 (2014)
[3] A Antidormi, L Colombo, S Roche, 2D Materials 8 (1), 015028
[4] S. Hong, et al. A. Antidormi, S. Roche, M. Chhowalla, H.J. Shin & H.S. Shin, Nature 582, 511–514 (2020)

Multiscale Computational Methods for Exploring Electrical and Thermal properties of highly disordered Composites Materials: From Machine-learning based Molecular Dynamics to Linear Scaling Transport Methods

Nature produces a variety of materials with many functions, often out of simple and abundant materials, and at low energy. Such systems - examples of which include silk, bone, nacre or diatoms - provide broad inspiration for engineering. Here we explore the translation of biological composites to engineering applications, using a variety of tools including molecular modeling, AI and machine learning, and experimental synthesis and characterization. We review a series of studies focused on the mechanical behavior of materials, especially fracture, and how these phenomena can be modeled using a combination of molecular dynamics and machine learning. We also present various case studies of hierarchical material optimization using genetic algorithms, applied to 3D printed composites, and protein design.

Computational material synthesis: Atomistic and molecular dynamics, composites, and bio-inspired AI

**Effect of Lung Surfactant Protein B Fragment, SP-B1-9 on Model Lipid Bilayer.**
Abinu, MHK, ISV, MRM

Lung surfactant is a mixture of protein and lipid that reduces surface tension at the air-water interface in the lungs and thus reduces the work needed to breathe. Two hydrophobic proteins, SP-B and SP-C, are thought to facilitate the re-spreading of fresh and recycled surfactant material from bilayer and multilayer reservoirs. In an earlier deuterium NMR study of bilayer model membranes containing either the SP-B fragment SP-B(1-25, 63-78) or the fragment SP-B(8-25, 63-78), it was found that lipid acyl chain orientational order was perturbed more strongly by SP-B(1-25, 63-78). Both fragments contain the first and last SP-B helices but differ in whether or not the insertion motif, SP-B1-7, is present, suggesting that the insertion motif might play a role in the capacity of SP-B to promote the bilayer reorganization implicit in lung surfactant function. To gain more insight into the interaction of the insertion motif with surfactant lipids, we have studied the effect of SP-B1-9 on lipid acyl chain order in DPPC-d62 /POPG(7:3) bilayers using deuterium NMR and GROMACS molecular dynamics simulations.
Using deuterium NMR at a peptide: lipid ratio of 0.065, we find that the peptide tends to disorder acyl chains in the liquid crystalline bilayer phase suggesting that it promotes an increase in average lipid headgroup separation. MD simulation of a bilayer model containing SP-B1-9 at a peptide: lipid ratio of 0.031 also shows a decrease in chain orientational order. The MD simulations also provide information about the average peptide orientation and conformation. These findings may provide a better understanding of the extent to which the insertion motif may contribute to the capacity of SP-B to promote reorganization of bilayer surfactant material.

Supported by NSERC Discovery Grants to MRM and ISV.

Effect of Lung Surfactant Protein B Fragment, SP-B1-9 on Model Lipid Bilayer.

In this work, we present a method based on deep learning (DL) to predict the structure of amorphous silicon (a-Si). The accuracy of our approach is validated through training networks to estimate the potential energy. Two architectures, multilayer perceptron (MLP) and convolutional neural network (CNN), have been examined for this purpose.
The models have been trained on a dataset generated by molecular dynamics (MD) simulations. The dataset consists of 216-atom a-Si configurations.
The performances of the two architectures over the test set are compared while they have the same total number of parameters and layers. 
The results of training these two neural networks yield a
root mean square error of the order of 0.1 meV/atom. This error was reported in the range of 2-6 meV/atom in previous works [1-3], which confirms the accuracy of our models. Calculation of the average error shows that CNN performs better than MLP.
After validation of our approach, we trained a generative model, variational autoencoder (VAE), to generate a-Si configurations. The generated structures and their structural properties are comparable to the real ones.
In general, the performances of our models are quite satisfactory, and this suggests that they could be used to approximate the potential energy and generate a-Si structures, and thus be a viable model for investigating disordered systems.



References: 
1. Li, R., Lee, E. and Luo, T., 2020, A unified deep neural network potential capable of predicting the thermal conductivity of silicon in different phases, Materials Today Physics 12, 100181.
2. Comin, M. and Lewis, L.J., 2019, Deep-learning approach to the structure of amorphous silicon, Physical Review B 100, 094107.
3. Behler, J. and Parrinello, M., 2007, Generalized neural-network representation of high-dimensional potential-energy surfaces, Physical Review Letters 98, 146401.

Modeling Amorphous Silicon Using Neural Networks

We use Langevin dynamics (LD) simulations to investigate single file diffusion (SFD) in a dilute solution of flexible linear polymers inside a narrow tube. The transition from single-file diffusion, where the mean-square displacement scales like $\langle x^2 \rangle \sim t^{1/2}$, to normal diffusion with $\langle x^2 \rangle \sim t$, is studied as a function of the system parameters, such as the width of the channel, the polymer concentration and the polymer size. Based on our simulation results and scaling arguments, we propose a phase diagram describing the different diffusion regimes. We also map this problem onto a one-dimensional Lattice Random Walk algorithm where the diffusing object represents the polymer's center of mass. In order to model the polymers entanglement and disentanglement processes we allow several objects to temporarily share the same lattice site and diffuse together. Extension of our work to polydisperse polymer solutions, one-dimensional electrophoresis, ring polymers and DNA mapping are discussed.

Underline digital video library

Multiscale Computational Methods for Exploring Electrical and Thermal properties of highly disordered Composites Materials: From Machine-learning based Molecular Dynamics to Linear Scaling Transport Methods

Computational material synthesis: Atomistic and molecular dynamics, composites, and bio-inspired AI

Effect of Lung Surfactant Protein B Fragment, SP-B1-9 on Model Lipid Bilayer.

Modeling Amorphous Silicon Using Neural Networks

The Limits of Polymer Single File Dynamics: A phase diagram

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES