Predicting properties from a material’s composition or structure is of great interest for materials design. Deep learning has recently garnered considerable interest in materials predictive tasks with low model errors when dealing with large materials data. However, deep learning models suffer in the small data regime that is common in materials science. Here we develop the AtomSets framework, which utilizes universal compositional and structural descriptors extracted from pre-trained graph network deep learning models with standard multi-layer perceptrons to achieve consistently high model accuracy for both small compositional data (<400) and large structural data (>130,000). The AtomSets models show lower errors than the graph network models at small data limits and other non-deep-learning models at large data limits. They also transfer better in a simulated materials discovery process where the targeted materials have property values out of the training data limits. The models require minimal domain knowledge inputs and are free from feature engineering. The presented AtomSets model framework can potentially accelerate machine learning-assisted materials design and discovery with less data restriction.
- Award ID(s):
- 1651668
- NSF-PAR ID:
- 10248527
- Date Published:
- Journal Name:
- Annual Review of Materials Research
- Volume:
- 50
- Issue:
- 1
- ISSN:
- 1531-7331
- Page Range / eLocation ID:
- 27 to 48
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
more » « less
Abstract -
Discovering new materials with desired properties has been a dominant and crucial topic of interest in the field of materials science in the past few decades. In this work, novel carbon allotropes and ternary B–C–N structures were generated using the state-of-the-art RG 2 code. All structures were fully optimized using density functional theory with first-principles calculations. Several hundred carbon allotropes and ternary B–C–N structures were identified to be superhard materials. The thermodynamic stability of some randomly selected superhard materials was confirmed by evaluating the full phonon dispersions in the Brillouin zone. The new carbon allotropes and ternary B–C–N structures possess a wide range of mechanical properties generally and Vickers hardness specifically. Through 2D Pearson's correlation map, we first reproduced the well-accepted explanation and relationship of the Vickers hardness of the generated structures with other mechanical properties such as shear modulus, bulk modulus, Pugh's ratio, universal anisotropy, and Poisson's ratio. We then propose two fundamentally new descriptors from the electronic level, namely local potential and electron localization function averaged over a unit cell, both of which exhibit a strong correlation with Vickers hardness. More importantly, these descriptors are easy to access from first-principles calculations (at least two orders of magnitude faster than the traditional calculation of elastic constants), and thus can serve as a fast and accurate approach for screening superhard materials. We also combined these new descriptors with known composition and structural descriptors in the machine learning training process. The new descriptors significantly enhance the performance of the trained machine learning model in predicting the Vickers hardness of unknown materials, which provides strong evidence for local potential and electron localization function to be considered in future high-throughput computation. This work unravels more fundamental but previously unexplored knowledge about superhard materials and the newly proposed electronic level descriptors are expected to accelerate the discovery of new superhard materials.more » « less
-
null (Ed.)Structural materials have lagged behind other classes in the use of combinatorial and high-throughput (CHT) methods for rapid screening and alloy development. The dual complexities of composition and microstructure are responsible for this, along with the need to produce bulk-like, defect-free materials libraries. This review evaluates recent progress in CHT evaluations for structural materials. High-throughput computations can augment or replace experiments and accelerate data analysis. New synthesis methods, including additive manufacturing, can rapidly produce composition gradients or arrays of discrete alloys-on-demand in bulk form, and new experimental methods have been validated for nearly all essential structural materials properties. The remaining gaps are CHT measurement of bulk tensile strength, ductility, and melting temperature and production of microstructural libraries. A search strategy designed for structural materials gains efficiency by performing two layers of evaluations before addressing microstructure, and this review closes with a future vision of the autonomous, closed-loop CHT exploration of structural materials. Expected final online publication date for the Annual Review of Materials Science, Volume 51 is August 2021. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.more » « less
-
more » « less
Abstract Decomposition of coarse detritus (e.g., dead organic matter larger than ~1 mm such as leaf litter or animal carcasses) in freshwater ecosystems is well described in terms of mass loss, particularly as rates that compress mass loss into one number (e.g., a first‐order decay coefficient, or breakdown rate, “
k ”); less described are temporal changes in the elemental composition of these materials during decomposition, with important implications for elemental cycling from microbes to ecosystems. This stands in contrast with work in the terrestrial realm, where a focus on detrital elemental cycling has provided a sharper mechanistic understanding of decomposition, especially with specific processes such as immobilization and mineralization. Notably, freshwater ecologists often measure carbon (C), nitrogen (N), and phosphorus (P), and their stoichiometric ratios in decomposing coarse materials, including carcasses, wood, leaf litter, and more, but these measurements remain piecemeal. These detrital nutrients are measurements of the entire detrital–microbial complex and are integrative of numerous processes, especially nutrient immobilization and mineralization, and associated microbial growth and death. Thus, data relevant to an elemental, mechanistically focused decomposition ecology are available in freshwaters, but have not been fully applied to that purpose. We synthesized published detrital nutrient and stoichiometry measurements at a global scale, yielding 4038 observations comprising 810 decomposition time series (i.e., measurements within a defined cohort of decomposing material through time) to build a basis for understanding the temporality of elemental content in freshwater detritus. Specifically, the dataset focuses on temporally and ontogenetically (mass loss) explicit measurements of N, P, and stoichiometry (C:N, C:P, N:P). We also collected ancillary data, including detrital characteristics (e.g., species, lignin content), water physiochemistry, geographic location, incubation system type, and methodological variables (e.g., bag mesh size). These measurements are important to unlocking mechanistic insights into detrital ontogeny (the temporal trajectory of decomposing materials) that can provide a deeper understanding of heterotroph‐driven C and nutrient cycling in freshwaters. Moreover, these data can help to bridge aquatic and terrestrial decomposition ecology, across plant or animal origin. By focusing on temporal trajectories of elements, this dataset facilitates cross‐ecosystem comparisons of fundamental decomposition controls on elemental fluxes. It provides a strong starting point (e.g., via modeling efforts) for comparing processes such as immobilization and mineralization that are understudied in freshwaters. Time series from decomposing leaf litter, particularly in streams, are common in the dataset, but we also synthesized ontogenies of animal‐based detritus, which tend to decompose rapidly compared with plant‐based detritus that contains high concentrations of structural compounds such as lignin and cellulose. Although animal‐based data were rare, comprising only three time series, their inclusion in this dataset underscores the opportunities to develop an understanding of decomposition that encompasses all detrital types, from carrion to leaf litter. There are no copyright or proprietary restrictions on the dataset; please cite this data paper when reusing these materials. -
Poisson’s ratio (ν) defines a material’s propensity to laterally expand upon compression, or laterally shrink upon tension for non-auxetic materials. This fundamental metric has traditionally, in some fields, been assumed to be a material-independent constant, but it is clear that it varies with composition across glasses, ceramics, metals, and polymers. The intrinsically elastic metric has also been suggested to control a range of properties, even beyond the linear-elastic regime. Notably, metallic glasses show a striking brittle-to-ductile (BTD) transition for ν-values above ~0.32. The BTD transition has also been suggested to be valid for oxide glasses, but, unfortunately, direct prediction of Poisson’s ratio from chemical composition remains challenging. With the long-term goal to discover such high-ν oxide glasses, we here revisit whether previously proposed relationships between Poisson’s ratio and liquid fragility (m) and atomic packing density (Cg) hold for oxide glasses, since this would enable m and Cg to be used as surrogates for ν. To do so, we have performed an extensive literature review and synthesized new oxide glasses within the zinc borate and aluminoborate families that are found to exhibit high Poisson’s ratio values up to ~0.34. We are not able to unequivocally confirm the universality of the Novikov-Sokolov correlation between ν and m and that between ν and Cg for oxide glass-formers, nor for the organic, ionic, chalcogenide, halogenide, or metallic glasses. Despite significant scatter, we do, however, observe an overall increase in ν with increasing m and Cg, but it is clear that additional structural details besides m or Cg are needed to predict and understand the composition dependence of Poisson’s ratio. Finally, we also infer from literature data that, in addition to high ν, high Young’s modulus is also needed to obtain glasses with high fracture toughness.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1146/annurev-matsci-110519-094700
