Composition Space Research Articles

ProCC: Progressive Cross-Primitive Compatibility for Open-World Compositional Zero-Shot Learning

Open-World Compositional Zero-shot Learning (OW-CZSL) aims to recognize novel compositions of state and object primitives in images with no priors on the compositional space, which induces a tremendously large output space containing all possible state-object compositions. Existing works either learn the joint compositional state-object embedding or predict simple primitives with separate classifiers. However, the former method heavily relies on external word embedding methods, and the latter ignores the interactions of interdependent primitives, respectively. In this paper, we revisit the primitive prediction approach and propose a novel method, termed Progressive Cross-primitive Compatibility (ProCC), to mimic the human learning process for OW-CZSL tasks. Specifically, the cross-primitive compatibility module explicitly learns to model the interactions of state and object features with the trainable memory units, which efficiently acquires cross-primitive visual attention to reason high-feasibility compositions, without the aid of external knowledge. Moreover, to alleviate the invalid cross-primitive interactions, especially for partial-supervision conditions (pCZSL), we design a progressive training paradigm to optimize the primitive classifiers conditioned on pre-trained features in an easy-to-hard manner. Extensive experiments on three widely used benchmark datasets demonstrate that our method outperforms other representative methods on both OW-CZSL and pCZSL settings by large margins.

Frequency Shuffling and Enhancement for Open Set Recognition

Open-Set Recognition (OSR) aims to accurately identify known classes while effectively rejecting unknown classes to guarantee reliability. Most existing OSR methods focus on learning in the spatial domain, where subtle texture and global structure are potentially intertwined. Empirical studies have shown that DNNs trained in the original spatial domain are inclined to over-perceive subtle texture. The biased semantic perception could lead to catastrophic over-confidence when predicting both known and unknown classes. To this end, we propose an innovative approach by decomposing the spatial domain to the frequency domain to separately consider global (low-frequency) and subtle (high-frequency) information, named Frequency Shuffling and Enhancement (FreSH). To alleviate the overfitting of subtle texture, we introduce the High-Frequency Shuffling (HFS) strategy that generates diverse high-frequency information and promotes the capture of low-frequency invariance. Moreover, to enhance the perception of global structure, we propose the Low-Frequency Residual (LFR) learning procedure that constructs a composite feature space, integrating low-frequency and original spatial features. Experiments on various benchmarks demonstrate that the proposed FreSH consistently trumps the state-of-the-arts by a considerable margin.

Convolutional Channel-Wise Competitive Learning for the Forward-Forward Algorithm

The Forward-Forward (FF) Algorithm has been recently proposed to alleviate the issues of backpropagation (BP) commonly used to train deep neural networks. However, its current formulation exhibits limitations such as the generation of negative data, slower convergence, and inadequate performance on complex tasks. In this paper we take the main ideas of FF and improve them by leveraging channel-wise competitive learning in the context of convolutional neural networks for image classification tasks. A layer-wise loss function is introduced that promotes competitive learning and eliminates the need for negative data construction. To enhance both the learning of compositional features and feature space partitioning, a channel-wise feature separator and extractor block is proposed that complements the competitive learning process. Our method outperforms recent FF-based models on image classification tasks, achieving testing errors of 0.58%, 7.69%, 21.89%, and 48.77% on MNIST, Fashion-MNIST, CIFAR-10 and CIFAR-100 respectively. Our approach bridges the performance gap between FF learning and BP methods, indicating the potential of our proposed approach to learn useful representations in a layer-wise modular fashion, enabling more efficient and flexible learning. Our source code and supplementary material are available at https://github.com/andreaspapac/CwComp.

Quantifying the impact of sample, instrument, and data processing on biological signatures in modern and fossil tissues detected with Raman spectroscopy

AbstractRaman spectroscopy is a popular tool for characterizing complex biological materials and their geological remains. Ordination methods, such as principal component analysis (PCA), use spectral variance to create a compositional space, the ChemoSpace, grouping samples based on spectroscopic manifestations reflecting different biological properties or geological processes. PCA allows to reduce the dimensionality of complex spectroscopic data and facilitates the extraction of informative features into formats suitable for downstream statistical analyses, thus representing a first step in the development of diagnostic biosignatures from complex modern and fossil tissues. For such samples, however, there is presently no systematic and accessible survey of the impact of sample, instrument, and spectral processing on the occupation of the ChemoSpace. Here, the influence of sample count, unwanted signals and different signal‐to‐noise ratios, spectrometer decalibration, baseline subtraction, and spectral normalization on ChemoSpace grouping is investigated and exemplified using synthetic spectra. Increase in sample size improves the dissociation of groups in the ChemoSpace, and our sample yields a representative and mostly stable pattern in occupation with less than 10 samples per group. The impact of systemic interference of different amplitude and frequency, periodical or random features that can be introduced by instrument or sample, on compositional biological signatures is reduced by PCA and allows to extract biological information even when spectra of differing signal‐to‐noise ratios are compared. Routine offsets ( 1 cm−1) in spectrometer calibration contribute in our sample to less than 0.1% of the total spectral variance captured in the ChemoSpace and do not obscure biological information. Standard adaptive baselining, together with normalization, increases spectral comparability and facilitates the extraction of informative features. The ChemoSpace approach to biosignatures represents a powerful tool for exploring, denoising, and integrating molecular information from modern and ancient organismal samples.

Design and Selection of High Entropy Alloys for Hardmetal Matrix Applications Using a Coupled Machine Learning and Calculation of Phase Diagrams Methodology

This study aims to utilize a combined machine learning (ML) and CALculation of PHAse Diagrams (CALPHAD) methodology to design hardmetal matrix phases for metal‐forming applications that can serve as the basis for carbide reinforcement. The vast compositional space that high entropy alloys (HEAs) occupy offers a promising avenue to satisfy the application design criteria of wear resistance and ductility. To efficiently explore this space, random forest ML models are constructed and trained from publicly available experimental HEA databases to make phase constitution and hardness predictions. Interrogation of the ML models constructed reveals accuracies >78.7% and a mean absolute error of 66.1 HV for phase and hardness predictions respectively. Six promising alloy compositions, extracted from the ML predictions and CALPHAD calculations, are experimentally fabricated and tested. The hardness predictions are found to be systematically under‐ and overpredicted depending on the alloy microstructure. In parallel, the phase classification models are found to lack sensitivity toward additional intermetallic phase formation. Despite the discrepancies identified between ML and experimental results, the fabricated compositions show promise for further experimental evaluation. These discrepancies are believed to be directly associated with the available databases but, importantly, have highlighted several avenues for both ML and database development.

Seasonality, Latitudinal Dependence, and Structural Evolution of Arctic Cyclones

Abstract Arctic cyclones (ACs) are an important component of the Arctic climate system. While previous studies focused on case studies or samples of intense ACs, an AC tracking algorithm is applied here to ERA5 to provide more than 9300 tracks. This large sample enables evaluations of seasonality, latitudinal dependence, and the structural evolution of ACs using storm-centered composite analysis and phase space analysis. The structures of ACs of different genesis regions, polar versus midlatitude, are also examined and compared. The results show that ACs typically have an asymmetric horizontal structure with cold air to the west and warm air to the east of the cyclone center. Cyclone asymmetry decreases, and the circulation becomes more barotropic in higher latitudes. ACs of polar origin are more symmetric than ACs of midlatitude origin and dominate the cyclone occurrences over the Arctic Ocean. Regarding seasonality, winter ACs are more intense and have a stronger horizontal asymmetry, and the cyclonic circulation extends higher into the stratosphere than summer ACs. In contrast, summer ACs have stronger warm anomalies in the lower stratosphere associated with subsidence above the cyclone center, and the cyclonic circulation typically does not extend beyond 50 hPa. The latitudinal and seasonal variations of AC structure are consistent with the latitudinal and seasonal differences in environmental baroclinicity. Additionally, our analyses show that the structural evolution of ACs is characterized by reduced vertical tilt and asymmetry, weakened temperature contrast between west and east sectors in troposphere, and reduced updraft strength in the later stage of the AC life cycle. Significance Statement Arctic cyclones play a crucial role in climate projections and risk assessments for Arctic infrastructure, transportation, and other human activities. This study aims to enhance our understanding of Arctic cyclone characteristics, including their origin, structure, seasonality, and evolution. We discovered that Arctic cyclones exhibit various structures in different environments. Their degrees of symmetry, vertical tilts, temperature contrasts, and updraft strengths vary with season, latitude, and life cycle stage.

Discovering High Entropy Alloy Electrocatalysts in Vast Composition Spaces with Multiobjective Optimization.

High entropy alloys (HEAs) are a highly promising class of materials for electrocatalysis as their unique active site distributions break the scaling relations that limit the activity of conventional transition metal catalysts. Existing Bayesian optimization (BO)-based virtual screening approaches focus on catalytic activity as the sole objective and correspondingly tend to identify promising materials that are unlikely to be entropically stabilized. Here, we overcome this limitation with a multiobjective BO framework for HEAs that simultaneously targets activity, cost-effectiveness, and entropic stabilization. With diversity-guided batch selection further boosting its data efficiency, the framework readily identifies numerous promising candidates for the oxygen reduction reaction that strike the balance between all three objectives in hitherto unchartered HEA design spaces comprising up to 10 elements.

Multi-objective optimization and machine learning assisted design and synthesis of magnesium based novel non-equiatomic medium entropy alloy

Non equiatomic compositionally complex alloys have gained significant attention due to their unique combinations of properties including high strength, good ductility, excellent thermal stability and corrosion resistance. The non equiatomic compositions provide flexibility in tailoring the mechanical, thermal, magnetic, biomedical and other functional properties to meet the specific application requirements. In this study, a novel design approach was proposed to identify quaternary non equiatomic medium entropy alloy for biomedical application. The design methodology employs multi-objective optimization of empirical phase formation parameters and ML algorithm to design solid solution containing non equiatomic medium entropy alloy. The suggested approach scans over a large composition space to find possible non-equiatomic compositionally complex alloy compositions (CCA) with desired phases. The proposed methodology is an alternative to trial-and-error alloy fabrication and CCA design approach reported in the literature. This step significantly reduced the time and resources utilized in CCA design by decreasing the number of experimental runs. Based on the design strategy, Mg60Ti24.24Zn9.66Nb6.1 alloy was designed and selected for fabrication through powder metallurgy route by employing microwave sintering. The Mg rich alloy exhibited the presence of two solid solution phases viz HCP and BCC in both prealloyed powder and final sintered specimen. The developed alloy showed a high microhardness value of 153.9 ± 0.85HV with a measured density of 2.82 ± 0.03 g/cm3. The mechanical properties such as yield strength, ultimate compressive strength, elastic modulus and fracture strain of the developed alloy were found to be superior to that of pure Mg and AZ31 fabricated through the same manufacturing route. Mechanical properties of local phases present within the alloy were assessed through nanoindentation technique. Elastic modulus (E) and hardness (H) values for the Ti rich BCC phase were 118.9±7.2 GPa and 4.65±1.1 GPa, respectively, whereas the Mg rich HCP phase showed values of 65.2±7 GPa and 1.6±0.43 GPa, respectively. Furthermore, with minor changes to the constraint equations and correct ML dataset selection, the suggested design technique can be applied to any multi-principal element alloy system to design non-equiatomic compositions.

Overcoming inter-dataset discrepancies in the grain size distributions of fine-grained sediments by partial least squares regression: A case study of the Belgian Boom Formation

The grain size distribution is an important property of all clastic sediments: it determines their mechanical properties and is directly related to their mode of transport and origin. Therefore, the accurate measurement and comparability of grain size data are important. The former has been studied in detail in literature and has been demonstrated to be significantly instrument-dependent, while the latter has not been given the same attention. The current study examined in detail a large set of grain size data measured on a single clay formation, the Oligocene Boom Formation, from which the large influence of sample preparation on the grain size distribution can be inferred. Especially the use of sonication to disintegrate silt-sized aggregates was found to be of a particularly big influence on the measured distributions. As a way to still be able to valorize non-comparable datasets, a statistical conversion procedure was introduced based on partial least squares regression in a compositional data space. The converted distributions follow the stratigraphical trends in grain size expected in the Boom Formation, while also being well correlated to hydraulic conductivity measurements performed on Boom Clay samples of similar depths. This is a strong indication that the conversion was successful. In the future, this approach can be used as a tool to harmonize any combination of compositional datasets, not just limited to grain size data, allowing a valorization of the maximal amount of data.

Machine learning accelerated discovery of corrosion-resistant high-entropy alloys

Corrosion has a wide impact on society, causing catastrophic damage to structurally engineered components. An emerging class of corrosion-resistant materials are high-entropy alloys. However, high-entropy alloys live in high-dimensional composition and configuration space, making materials designs via experimental trial-and-error or brute-force ab initio calculations almost impossible. Here we develop a physics-informed machine-learning framework to identify corrosion-resistant high-entropy alloys. Three metrics are used to evaluate the corrosion resistance, including single-phase formability, surface energy and the compactness of oxide films formed on an alloy surface evaluated by Pilling–Bedworth ratios. We used random forest models to predict the single-phase formability, trained on an experimental dataset. Machine learning inter-atomic potentials were employed to calculate surface energies and Pilling–Bedworth ratios, which are trained on first-principles data fast sampled using embedded atom models. A combination of random forest models and high-fidelity machine learning potentials represents the first of its kind to relate chemical compositions to corrosion resistance of high-entropy alloys, paving the way for automatic design of materials with superior corrosion protection. This framework was demonstrated on AlCrFeCoNi high-entropy alloys and we identified composition regions with high corrosion resistance from a wide range of compositions. Machine learning predicted lattice constants and surface energies are consistent with values by first-principles calculations. The predicted single-phase formability and corrosion-resistant compositions of AlCrFeCoNi agree well with experiments. This framework provides a computationally efficient approach to navigate high-dimensional composition space of high-entropy alloys. It is general in its application and applicable to other complex materials, enabling high-throughput screening of material candidates and potentially accelerating the iteration of integrated computational materials engineering.

Learning Graph Embeddings for Open World Compositional Zero-Shot Learning.

Compositional Zero-Shot learning (CZSL) aims to recognize unseen compositions of state and object visual primitives seen during training. A problem with standard CZSL is the assumption of knowing which unseen compositions will be available at test time. In this work, we overcome this assumption operating on the open world setting, where no limit is imposed on the compositional space at test time, and the search space contains a large number of unseen compositions. To address this problem, we propose a new approach, Compositional Cosine Graph Embeddings (Co-CGE), based on two principles. First, Co-CGE models the dependency between states, objects and their compositions through a graph convolutional neural network. The graph propagates information from seen to unseen concepts, improving their representations. Second, since not all unseen compositions are equally feasible, and less feasible ones may damage the learned representations, Co-CGE estimates a feasibility score for each unseen composition, using the scores as margins in a cosine similarity-based loss and as weights in the adjacency matrix of the graphs. Experiments show that our approach achieves state-of-the-art performances in standard CZSL while outperforming previous methods in the open world scenario.

Mitigating the efficiency-voltage trade-off in magnesium air battery via a novel active learning framework

Reconciling the inherent trade-off between anodic efficiency and discharge voltage in the design of high-energy-density magnesium-air (Mg-air) batteries has been a persistent challenge. Herein, we propose a pioneering active learning strategy that integrates physically motivated variables, machine learning, exploration of the Pareto front, experimental data feedback, and generated data feedback for the purpose of designing magnesium anodes. Within an extensive compositional space (∼350,000 possibilities), we have pinpointed a novel alloy, Mg-1Ga-1Ca-0.5In, exhibiting exceptional performance with high efficiency (64 ± 5.5 % at 1 mA cm−2, 64 ± 0.5 % at 10 mA cm−2) and high voltage (1.80 ± 0.00 V at 1 mA cm−2, 1.57 ± 0.01 V at 10 mA cm−2), surpassing conventional methods of alloying. Subsequent experiments and density functional theory (DFT) calculations have unveiled that the outstanding performance of Mg-1Ga-1Ca-0.5In stems from “grain boundary activation” induced by active second phases and “intra-grain inhibition” resulting from the orbital hybridization between solute atoms and Mg atoms. This study provides a novel research paradigm and offers valuable insights for the further development of high-performance Mg-air batteries.

Magnons in a two-dimensional Weyl magnet

The discovery of topological states of matter has led to a revolution in condensed-matter science. While a non-trivial band topology in a material is often associated with intriguing transport properties, much less attention has been given to the impact on spin dynamics and non-equilibrium magnetization states. Here, we provide evidence that a chiral asymmetric magnon dispersion in the two-dimensional Weyl magnet Fe/W(110) is related to the presence of Weyl fermions close to the Fermi energy and surface Fermi arcs. We find that the large anomalous Hall conductivity and the Dzyaloshinskii–Moriya interaction are attributed to the non-trivial band topology in the composite momentum-magnetization space. Our results show the direct impact of Weyl fermions on both the charge and spin dynamics in a two-dimensional magnet. Unveiling these principles can promote innovative technologies in magnonics by utilizing topological materials, where magnons and non-trivial topological electronic states can be manipulated through magnetization.

Achieving Compatible p/n-Type Half-Heusler Compositions in Valence Balanced/Unbalanced Mg1-xVxNiSb.

In thermoelectric and other inorganic materials research, the significance of half-Heusler (HH) compositions following the 18-electron rule has drawn interest in developing and exploiting the potential of intermetallic compounds. For the fabrication of thermoelectric modules, in addition to high-performance materials, having both p- and n-type materials with compatible thermal expansion coefficients is a prerequisite for module development. In this work, the p-type to n-type transition of valence balanced/unbalanced HH composition of Mg1-xVxNiSb was demonstrated by changing the Mg:V chemical ratio. The Seebeck coefficient and power factor of Ti-doped Mg0.57V0.33Ti0.1NiSb are -130 μV K-1 and 0.4 mW m-1 K-2 at 400 K, respectively. In addition, the reduced lattice thermal conductivity (κL < 2.5 W m-1 K-1 at 300 K) of n-type compositions was reported to be much smaller than κL of conventional HH materials. As high thermal conductivity has long been an issue for HH materials, the synthesis of p- and n-type Mg1-xVxNiSb compositions with low lattice thermal conductivity is a promising strategy for producing high-performance HH compounds. Achieving both p- and n-type materials from similar parent composition enabled us to fabricate a thermoelectric module with maximum output power Pmax ∼ 63 mW with a temperature difference of 390 K. This finding supports the benefit of exploring the huge compositional space of valence balanced/unbalanced quaternary HH compositions for further development of thermoelectric devices.

Nb-doping of cation-disordered rocksalt oxides with B2O3 surface modification for superior performance cathode

As the growing price and inadequate reserves of cobalt, the demand for cobalt-free materials with superior energy density to apply in Li-ion batteries has increased distinctly. Li-rich cation disordered rocksalt (DRX) material has aroused research hotspot because of high specific capacity and wide compositional space. However, irreversible oxygen loss as well as sluggish kinetic raise the problem in large polarization and inevitable capacity decay in cycling process, which have impeded the commercialization of these electrode materials. In this work, we have synthesized cobalt-free boron trioxide (B2O3)-modified Li1.2Ni0.3Ti0.3Nb0.2O2 compounds with cation disordered rocksalt structure through mechanochemical method. It is a fact that B2O3-modified cathodes possess outstanding cyclic stability as well as rate properties. Particularly, the Li1.2Ni0.3Ti0.3Nb0.2O2 electrode with 2.0 wt% B2O3 displayed a competitive capacity of 192.4 mAh/g at 10 mA g−1 and retained 134.7 mAh/g after 100 charge/discharge times. The redox mechanism is explored through ex-situ XPS measurement, verifying the reversible cationic redox of Ni2+/Ni4+ accompanying small Ti4+/Ti3+ reduction and anionic O2−/O− evolution, which results in unusual high capacity. Such an optimized surface modification combines with the Nb doping, can promote Li ion transport, reduce polarization, protect cathode from reacting with electrolyte. These findings provide important implications to improve performance of DRX cathodes.

One-step molten salt synthesis of high entropy oxides

High-entropy oxides (HEOs) have received widespread attention owing to their tunable properties and vast composition spaces. However, the universal synthesis of HEOs with high purity still faces challenge. In this work, a one-step molten salt method is reported for the synthesis of phase-pure rock salt-type (Mg0.2Co0.2Ni0.2Cu0.2Zn0.2)O. When used as an anode material for lithium-ion batteries, (Mg0.2Co0.2Ni0.2Cu0.2Zn0.2)O exhibits high specific capacity and cycling stability, maintaining a capacity retention of 94% after 220 cycles (0.2 A g−1). The superior performance can be attributed to accelerated electron and ion transport as well as enhanced Li adsorption/coupling induced by internal defects. To further demonstrate the wide applicability of the molten salt method, single-phase spinel and pyrochlore HEOs are also successfully synthesized. The presented molten salt method, with its ease of operation, homogeneous reaction environment and low synthesis temperatures, makes it an ideal synthesis technique for the development of novel HEO powders, especially for electrode materials.

Generalized system-bath entanglement theorem for Gaussian environments.

The entanglement between system and bath often plays a pivotal role in complex systems spanning multiple orders of magnitude. A system-bath entanglement theorem was previously established for Gaussian environments in J. Chem. Phys. 152, 034102 (2020) regarding linear response functions. This theorem connects the entangled responses to the local system and bare bath properties. In this work, we generalize it to correlation functions. Key steps in derivations involve using the generalized Langevin dynamics for hybridizing bath modes and the Bogoliubov transformation that maps the original finite-temperature reservoir to an effective zero-temperature vacuum by employing an auxiliary bath. The generalized theorem allows us to evaluate the system-bath entangled correlations and the bath mode correlations in the total composite space, as long as we know the bare-bath statistical properties and obtain the reduced system correlations. To demonstrate the cross-scale entanglements, we utilize the generalized theorem to calculate the solvation free energy of an electron transfer system with intramolecular vibrational modes.

Informatics-Driven Design of Superhard B-C-O Compounds.

Materials containing B, C, and O, due to the advantages of forming strong covalent bonds, may lead to materials that are superhard, i.e., those with a Vicker's hardness larger than 40 GPa. However, the exploration of this vast chemical, compositional, and configurational space is nontrivial. Here, we leverage a combination of machine learning (ML) and first-principles calculations to enable and accelerate such a targeted search. The ML models first screen for potentially superhard B-C-O compositions from a large hypothetical B-C-O candidate space. Atomic-level structure search using density functional theory (DFT) within those identified compositions, followed by further detailed analyses, unravels on four potentially superhard B-C-O phases exhibiting thermodynamic, mechanical, and dynamic stability.

Machine Learning Design for High-Entropy Alloys: Models and Algorithms

High-entropy alloys (HEAs) have attracted worldwide interest due to their excellent properties and vast compositional space for design. However, obtaining HEAs with low density and high properties through experimental trial-and-error methods results in low efficiency and high costs. Although high-throughput calculation (HTC) improves the design efficiency of HEAs, the accuracy of prediction is limited owing to the indirect correlation between the theoretical calculation values and performances. Recently, machine learning (ML) from real data has attracted increasing attention to assist in material design, which is closely related to performance. This review introduces common and advanced ML models and algorithms which are used in current HEA design. The advantages and limitations of these ML models and algorithms are analyzed and their potential weaknesses and corresponding optimization strategies are discussed as well. This review suggests that the acquisition, utilization, and generation of effective data are the key issues for the development of ML models and algorithms for future HEA design.

Mixed Enthalpy-Entropy Descriptor for the Rational Design of Synthesizable High-Entropy Materials Over Vast Chemical Spaces.

The practically unlimited high-dimensional composition space of high-entropy materials (HEMs) has emerged as an exciting platform for functional material design and discovery. However, the identification of stable and synthesizable HEMs and robust design rules remains a daunting challenge. Here, we propose a mixed enthalpy-entropy descriptor (MEED) that enables highly efficient, robust, high-throughput prediction of synthesizable HEMs across vast chemical spaces from first-principles. The MEED is based on two parameters: the relative formation enthalpy with respect to the most stable competing compound and the spread of the point-defect formation energy spectrum. The former measures the relative synthesizability of an HEM to its most stable competing phase, going beyond the conventional thermodynamic understanding. The latter gauges the relative entropy forming ability of an HEM, entailing no sampling over numerous alloy configurations. By applying the MEED to two structurally distinct representative material systems (i.e., 3D rocksalt carbides and 2D layered sulfides), we not only successfully identify all experimentally reported HEMs within these systems but also reveal a cutoff criterion for assessing their relative synthesizability within each system. By the MEED, tens of new high-entropy carbides and 2D high-entropy sulfides are also predicted, which have the potential for a wide variety of applications such as coating in aerospace devices, energy conversion and storage, and flexible electronics.