Achieving DFT accuracy with a machine-learning interatomic potential: Thermomechanics and defects in bcc ferromagnetic iron

Dragoni, Daniele; Daff, Thomas D.; Csänyi, Gábor; Marzari, Nicola

doi:10.1103/physrevmaterials.2.013808

Cited by 246 publications

(243 citation statements)

References 98 publications

(109 reference statements)

Supporting

Mentioning

232

Contrasting

Order By: Relevance

“…Overall as seen from the table, DFT reproduces well the defects that have low formation energies but somewhat overestimates those with higher formation energies. This is consistent with other systems such as Fe 32 . The DFT diffusion energy barrier ,-.,H;I associated with the vacancy movement to an adjacent position within the lattice is in very good agreement with the experimental value.…”

Section: Point Defects In Bulk Cu and Zrsupporting

confidence: 92%

“…Assessing a force-fields' ability to accurately optimize exotic lattice structures is another useful test to demonstrate the robustness of the potential. 13,32,[72][73] As most of these structures are not readily observed experimentally, the ultimate utility of reproducing these values is merely for benchmarking purposes. These calculations can show the ability of the potential to distinguish between initially similar lattice spacing (e.g.…”

Section: Cu and Zr Additional Lattice Structuresmentioning

confidence: 99%

See 1 more Smart Citation

Optimization and validation of a deep learning CuZr atomistic potential: Robust applications for crystalline and amorphous phases with near-DFT accuracy

Andolina

Williamson

Saidi

2020

The Journal of Chemical Physics

View full text Add to dashboard Cite

We show that a deep-learning neural network potential (DP) based on density functional theory (DFT) calculations can well describe Cu-Zr materials, an example of a binary alloy system that can coexist in several ordered intermetallics and as an amorphous phase. The complex phase diagram for Cu-Zr makes it a challenging system for traditional atomistic force-fields that fail to describe well the different properties and phases. Instead, we show that a DP approach using a large database with ~300k configurations can render results generally on par with DFT. The training set includes configurations of pristine and bulk elementary metals and intermetallics in the liquid and solid phases in addition to slab and amorphous configurations. The DP model was validated by comparing bulk properties such as lattice constants, elastic constants, bulk moduli, phonon spectra, surface energies to DFT values for identical structures. Further, we contrast the DP results with values obtained using well-established two embedded atom method potentials. Overall, our DP potential provides near DFT accuracy for the different Cu-Zr phases but with a fraction of its computational cost, thus enabling accurate computations of realistic atomistic models especially for the amorphous phase. I.quenching rate. Both requirements pose a challenge for DFT simulations. Classical atomistic simulation methods, of which molecular dynamic simulations (MD) is the most common, have been employed in materials design for several decades to provide fast computations of atomic energies and forces. However, the parameters for these potentials to describe interatomic interactions are nontrivial to optimize. To date, the most accurate classical potentials for Cu-Zr systems are based on embedded atom method 12 (EAM) particularly those developed by Sheng et al. [13][14][15] (EAM-HS) and Mendeleev et al. 16 (EAM-MM) by fitting to a mixed experimental and ab-initio input data. While these two EAMs provide a general good description of many properties of Cu-Zr systems especially for properties that are included in the training such as the Cu-Zr melting curve as is the case of EAM-MM, they have limited transferability to configurations not included in the training as we demonstrate here. Further, and most importantly, EAM potentials have a limited functional form, and thus cannot be utilized to describe complex interactions such as for covalent and ionic bonds between Cu-Zr and reactive gasses.Machine learning (ML) and particularly deep neural network-based force-fields have the flexibility and non-linearity necessary to describe complex potential energy surfaces. [17][18][19][20][21][22][23] In traditional atomistic force-field methods such as the EAMs, the complex potential energy surface is approximated using privileged functional forms that are selected based on physical motivation. On the other hand, ML potentials do not employ any explicit functional form for the dependence of the energies and forces on the atomic coordinates, but rather "learn" how atoms interac...

show abstract

Section: Point Defects In Bulk Cu and Zrsupporting

confidence: 92%

Section: Cu and Zr Additional Lattice Structuresmentioning

confidence: 99%

Optimization and validation of a deep learning CuZr atomistic potential: Robust applications for crystalline and amorphous phases with near-DFT accuracy

Andolina

Williamson

Saidi

2020

The Journal of Chemical Physics

View full text Add to dashboard Cite

show abstract

“…Comparison of different atomistic ML potentials (presented in section 2.3.3.1) was studied for water interactions [421]. Gaussian approximation potentials (GAPs) have been extensively used to study different systems, such as elemental boron [422], amorphous carbon [423,424], silicon [425], thermal properties of amorphous GeTe and carbon [426], thermomechanics and defects of iron [427], prediction structures of inorganic crystals by combing ML with random search [428], λ-SOAP method for tensorial properties of atomistic systems [247], and a unified framework to predict the properties of materials and molecules such as silicon, organic molecules and proteins ligands [429]. A recent review of applications of high-dimensional neural neural network potentials [430] summarized the notable number of molecular and materials systems studied, which ranges from simple semiconductors such as silicon [233,431,432] and ZnO [433], to more complex systems such as water and metallic clusters [434], molecules [435][436][437], surfaces [438,439], and liquid/solid interfaces [414,440].…”

Section: Discovery Energies and Stabilitymentioning

confidence: 99%

From DFT to machine learning: recent approaches to materials science–a review

et al. 2019

View full text Add to dashboard Cite

Recent advances in experimental and computational methods are increasing the quantity and complexity of generated data. This massive amount of raw data needs to be stored and interpreted in order to advance the materials science field. Identifying correlations and patterns from large amounts of complex data is being performed by machine learning algorithms for decades. Recently, the materials science community started to invest in these methodologies to extract knowledge and insights from the accumulated data. This review follows a logical sequence starting from density functional theory as the representative instance of electronic structure methods, to the subsequent high-throughput approach, used to generate large amounts of data. Ultimately, data-driven strategies which include data mining, screening, and machine learning techniques, employ the data generated. We show how these approaches to modern computational materials science are being used to uncover complexities and design novel materials with enhanced properties. Finally, we point to the present research problems, challenges, and potential future perspectives of this new exciting field. characterized by its diverse and huge volume, usually ranging from terabytes to petabytes of data, being created in or near real-time. Such data is found either structured and unstructured in nature, and is exhaustive, usually aiming to capture entire populations in a scalable manner [7]. Simple tasks represent challenges in this scale: capture, curation, storage, search, sharing, analysis, and visualization of the data cannot be accomplished without the proper tools. Thus, it can be effectively summarized by the popular 'five V's': volume, velocity, variety, veracity, and value, shown in figure 3(right). A related sixth V is visualization, although not exclusive to Big Data, which requires different techniques to handle data with various characteristics.Striving to tackle the challenges imposed by Big Data, the field of Data Science has arisen. It is largely interdisciplinary being a combination of mathematics and statistics, computer science and programming, and domain knowledge for problem definition and solving, as shown in figure 3(left). Its objective is, roughly speaking, to deal with the whole process of data production, cleaning, preparation, and finally, analysis. Data science encompasses areas such as Big Data, which deals with large volumes of data, and data mining, which relates to analysis processes to discover patterns and extract knowledge from data, part of the so-called Knowledge Discovery in Databases (KDD).The analysis process within Data Science is challenging, as the techniques are very different from traditional static and rigid datasets, generated and analyzed under a predetermined hypothesis. The distinction from traditional data is based on the larger abundance, exhaustivity, and variety of Big Data. It is also much more dynamic, messy and uncertain, being highly relational [7]. Recently, the possibility of overcoming such a challenge slowly sta...

show abstract

“…Low-index bcc surface structures were taken from [54] ((1 1 1) surfaces from [58]). Additionally, to make our GAP applicable to surface irradiation and improve the transferability to arbitrary surface structures, we also included damaged and half-molten (1 1 0) and (1 0 0) surfaces.…”

Section: Trainingmentioning

confidence: 99%

Machine-learning interatomic potential for radiation damage and defects in tungsten

et al. 2019

View full text Add to dashboard Cite

We introduce a machine-learning interatomic potential for tungsten using the Gaussian Approximation Potential framework. We specifically focus on properties relevant for simulations of radiationinduced collision cascades and the damage they produce, including a realistic repulsive potential for the short-range many-body cascade dynamics and a good description of the liquid phase. Furthermore, the potential accurately reproduces surface properties and the energetics of vacancy and self-interstitial clusters, which have been long-standing deficiencies of existing potentials. The potential enables molecular dynamics simulations of radiation damage in tungsten with unprecedented accuracy.

show abstract

Achieving DFT accuracy with a machine-learning interatomic potential: Thermomechanics and defects in bcc ferromagnetic iron

Cited by 246 publications

References 98 publications

Optimization and validation of a deep learning CuZr atomistic potential: Robust applications for crystalline and amorphous phases with near-DFT accuracy

Optimization and validation of a deep learning CuZr atomistic potential: Robust applications for crystalline and amorphous phases with near-DFT accuracy

From DFT to machine learning: recent approaches to materials science–a review

Machine-learning interatomic potential for radiation damage and defects in tungsten

Contact Info

Product

Resources

About