Informal dinner for all attendees, providing a relaxed environment for networking
Attendees register before the start of the workshop. A great opportunity for initial networking.
Artificial intelligence (AI) is reshaping the future of materials science by enabling faster, smarter, and more targeted discovery and design of new materials. This open talk will introduce key concepts in AI-driven materials research, highlight successful applications across fields such as energy, semiconductors, and biomedical materials, and examine critical challenges including data scarcity, model interpretability, and experimental validation.
In addition, the talk will outline the goals and objectives of the AI-Powered Materials Discovery at Great Plains workshop—namely, to build interdisciplinary connections, accelerate AI adoption in the materials community, and explore opportunities for collaborative research and funding. Designed to engage attendees from diverse backgrounds, this session will provide a shared foundation for participants and set the stage for meaningful discussion and innovation throughout the workshop.
Nanoplastics (NP) are ubiquitous, and their interactions with agricultural and food systems are shown to be associated with health concerns for plants, animals and humans. However, fast and accurate detection and characterization of NP in biosystems remains a challenge; and what governs the interactions between NP and biosystems are still largely unknown. Data driven techniques utilizing AI/ML as tools can greatly help with finding the “missing link” in today’s knowledge of NP: what factors matter in determining the characteristics of how NP interact with biomolecules? Using such knowledge, sensor platforms can be developed that support fast detection of NPs. One such platform is presented in this talk utilizing velocity profile analysis of lateral flow in multi-channel microfluidic sensors to detect and characterize NP in samples with high throughput and fidelity, yet easy to use and field-deployable. Furthermore, machine learning (ML)-enabled data analysis is used to understand NP-molecular interactions which enables material discovery for better design and development of paper-based microfluidic chips (pMFC) for NP detection with high-throughput and fidelity. The data-driven techniques will help solving problems associated with the presence of nanoplastics in various systems.
Sulfate-reducing bacteria (SRB) play a pivotal role in the global sulfur cycle and microbial metal transformations, influencing both ecological resilience and industrial challenges such as microbially induced corrosion (MIC). This research focuses on Oleidesulfovibrio alaskensis G20 (OA G20), a genetically tractable SRB model, to elucidate mechanisms underlying stress tolerance, biofilm formation, and adaptation to metal-contaminated environments.
Employing a systems biology framework, the study integrates multi-omics datasets, network biology, and literature-mining pipelines to define the functional landscape of biofilm-driven SRB physiology. A novel computational workflow was developed to identify core gene modules involved in environmental stress responses and biofilm regulation. Experimental results reveal that OA G20 exhibits distinct morphological and transcriptomic adaptations under acidic, alkaline, and copper-induced stress, with key alterations in exopolysaccharide (EPS) synthesis, energy metabolism, and metal detoxification pathways. Surface-resolved omics analyses demonstrate enhanced biofilm architecture and upregulation of genes associated with nanowire production, signal transduction, and ion homeostasis under copper exposure. Notably, this work presents the first epigenetic landscape of SRB biofilms subjected to metal stress, uncovering differential 5-methylcytosine (5mC) DNA methylation patterns in genes governing carbon metabolism and lipopolysaccharide (LPS) transport. Proteomic profiling further supports the copper-specific modulation of chemotaxis and cytoskeletal protein networks.
Given the complexity of the OMICS data and the intricacy of the mechanisms by which SRB interact with and adapt to metal surfaces, advanced artificial intelligence (AI) tools are essential for comprehensive integration and mechanistic interpretation.
The transformative potential of artificial intelligence (AI) extends beyond the laboratory. This plenary address explores the impact of AI in education, touching on how intelligent systems are being deployed in educational contexts and discussing the ethical and pedagogical challenges inherent in integrating these powerful technologies into educational settings. By drawing parallels between AI's role in accelerating discovery and its potential impacts on human intelligence and the processes of learning, this presentation aims to spark interdisciplinary dialogue and critical examination of these emerging technologies to inspire and perhaps transform today's innovative applications of AI for the betterment of education and our society.
Liquid electrolytes play a pivotal role in governing the performance, safety, and longevity of lithium and sodium ion batteries. However, designing optimal electrolytes remains a complex challenge due to the need to simultaneously satisfy multiple criteria, including high ionic conductivity, broad electrochemical stability, low viscosity, chemical compatibility with electrodes, and thermal stability. Complicating this further is the virtually infinite chemical design space arising from the vast combinations of solvents, salts, and additives.
Given these challenges, the integration of artificial intelligence (AI) into materials discovery is becoming not only advantageous but essential. In this talk, we present a novel AI framework trained on a large experimental dataset generated by a high-throughput robotic platform, enabling accurate prediction of the ionic conductivity of Li and Na-based liquid electrolytes. We demonstrate how this model can be used to rapidly screen and identify high-performance multi-component electrolyte formulations, including those with mixed solvents and salts, achieving state-of-the-art conductivity. This work highlights the potential of AI to significantly accelerate electrolyte discovery and guide the rational design of next-generation energy storage materials.
Data centers will consume 9% of U.S. electricity generation annually by 2030 according to the Electric Power Research Institute. There are two energy-related challenges for building AI data centers with a greater capacity: a shortage of electricity and the need to reduce carbon emissions. Small modular reactors (SMRs) may solve both issues by co-location with data center campuses. Compared to the traditional full-scale nuclear energy plants, SMRs have advantages such as smaller physical footprints and reduced capital investment. Companies such as Terrestrial Energy, TerraPower, NuScale, X-energy, and Rolls-Royce are developing SMR technology for commercial launching in the late 2020s. This presentation will outline the new R&D opportunities along with our research in SMRs and additive manufacturing (AM) for AI data center infrastructures, with the aim of increasing the data center efficiency and reducing the deployment time and manufacturing cost. These AI data centered powered by nuclear energy may create new economic opportunities for Nebraska and the Great Plains.
Recent advances of machine learning interatomic potentials (MLIPs) have improved both the accuracy and scalability of energy and force predictions in chemical systems for many practical applications. Here we explore the combination of MLIPs with state-of-the-art ab initio theory of thermal transport, which requires accurate estimations of higher-order derivatives of the potential energy surface (PES) to reliably replace the expensive density functional perturbation theory (DFPT) step. We present examples of different variables in MLIP training workflow and suggest the extent to which they impact the overall model performance for phonon frequency predictions.
Moiré superlattices, formed by stacking layered 2D materials with a twist in orientation, have emerged as a new platform for exploration of new physics and exotic quantum phenomena. The twist-angle dependent moiré effects and superlattice potentials offer a new route in materials design and quantum engineering. However, a direct prediction of the superlattice potential remains challenging due to the large moiré periodicity and associated computational demand. In this talk, I will introduce our computational approaches combining Density Functional Theory (DFT) calculations, machine learning and classical physics models to understand and predict the superlattice potentials for twisted bilayer hexagonal boron nitride. Optimizing the atomic structure using machine-learned interatomic potential reveals significant structural reconstructions. We introduce an efficient classical physics model that enables predictive and interpretative insights into the formation of superlattice potentials in these twisted systems.
Discussions on education and outreach initiatives for AI in materials science
This interactive workshop equips middle and high school teachers to demystify Artificial Intelligence (AI) for their students. We'll explore core AI concepts, delve into the critical topic of AI bias, and examine the practical implications of AI's capabilities and limitations within educational settings. Participants will learn effective strategies for discussing AI and Large Language Models (LLMs) with students. You'll engage with, critique, and adapt hands-on AI activities, ensuring they're classroom-ready. Attendees will leave with ready-made AI literacy materials suitable for integration into any subject, including a student workbook aligned with the Developing AI Literacy (DAILy) curriculum from Everyday-AI.org website.
The Materials Genome Initiative (MGI) has heralded a sea change in the philosophy of materials design. In an increasing number of applications, the successful deployment of novel materials has benefited from the use of computational, experimental and informatics methodologies. Here, we describe the role played by computational and experimental data generation, capture and management, polymer fingerprinting, machine-learning based property prediction models, algorithms for designing polymers meeting target property requirements, and how prior physics knowledge may be incorporated with polymer informatics workflows. These efforts have culminated in the creation of an online Polymer Informatics platform, to guide ongoing and future polymer discovery and design. Systematic steps that may be taken to apply such informatics efforts to a wide range of technological domains will be discussed. These include strategies to deal with the data bottleneck, methods to represent polymer formulations, morphology and processing conditions, and the applicability of emerging information fusion, physics enforcement, language models and generative AI algorithms to accelerate materials development.
Reduction of defects such as cracking, porosity, lack of fusion, distortion, and surface roughness and control of grain structure and phase formation are needed to improve part quality, reduce cost, and increase the market penetration of 3D printed or additively manufactured components. Reduction in defects and control of microstructure cannot be done by time-consuming and expensive experimental trials because of the involvement of many variables with a large parameter window. Physics-based mechanistic models are often used as an alternative. However, the evolution of microstructures and defects depends on many complex physical processes, and the mechanistic understanding of many of these processes is not fully developed. The use of emerging artificial intelligence (AI) tools such as machine learning and deep learning can automate several steps, including process monitoring, defect detection, sensing, and process control, and can help in the selection of appropriate processing conditions to improve structure and properties. This would minimize the need for human intervention and significantly improve the process efficiency, productivity, and part quality and reduce materials and energy waste and cost. In this work, the effectiveness of AI tools has been evaluated in reducing defects and improving the microstructure and properties of additively manufactured metallic components. Several experimental data for additive manufacturing processes were gathered, which were then used to train the machine learning and deep learning algorithms. Artificial neural networks, decision trees, random forests, and support vector machines were tested under various conditions and materials. The results indicated that the integration of AI tools in additive manufacturing can reduce cracking, residual stresses, lack of fusion, and balling defects. In addition, several examples of the use of machine learning and deep learning for in-situ process monitoring, sensing and control, parameter optimization, and controlling microstructure and properties have been provided. It has also been shown that AI tools perform better if they are trained using the variables computed using mechanistic models of the additive manufacturing process. In addition, Several examples of using generative AI tools such as ChatGPT to write codes to use machine learning and mechanistic modeling in additive manufacturing are included.
Silver iodide is extensively studied for its ionic conductivity, in addition to its thermally-driven polymorphism into several phases including wurtzite, zincblende, rocksalt, and body-centered cubic [1,2]. The exact mechanisms and driving forces behind these transformations, however, are not well understood. Recently, molecular dynamics simulations (MD) informed by machine-learned interatomic potentials (MLIP) have been a promising tool to provide an atomistic picture of mechanically- and thermally-driven transformations [3]. To that end, we report progress in MD-MLIP simulations of the dominant phases in the pressure-temperature window of interest (1-5 kbar, 200-500K) to provide insight on the atomistic mechanisms for the polymorphism with implications for other drivers for such transformations (i.e. mechanochemistry).
References
[1] S. De Panfilis, A. Di Cicco, A. Filipponi, & M. Minicucci. Solid and liquid AgI at high pressure and high temperature: A X-ray absorption spectroscopy study. High Press. Res. 22, 349 (2010).
[2] O. Ohtaka, H. Takebe, A. Yoshiasa, H. Fukui, & Y. Katayama. Phase relations of AgI under high pressure and high temperature. Solid State Commun. 123, 213 (2002).
[3] H. Zong, G. Pilania, X. Ding, G. J. Ackland, & T. Lookman. Developing an interatomic potential for martensitic phase transformations in zirconium by machine learning. npj Comput. Mater. 4, 48 (2018).
The application of large language models (LLMs) in scientific domains, particularly materials discovery, remains underexplored compared to general natural language tasks. This study presents a systematic fine-tuning of the DeepSeek-R1 model using Unsloth’s memory-efficient framework and domain-specific supervised instruction techniques. Three publicly available datasets—GPT-Narratives-for-Materials, MDF-LLM Resources, and MatChem-LLM—were curated and harmonized to develop a fine-grained instruction-response corpus relevant to material science properties, experimental conditions, and compositional analysis.
The fine-tuning process leveraged SFTTrainer with instruction tuning templates, a sequence length of 2048 tokens, and efficient gradient accumulation to adapt DeepSeek-R1 for expert-level retrieval, summarization, and hypothesis generation tasks in material science. Preliminary evaluations using a simulated GLUE-style protocol, including BLEU-based output quality assessment, indicate significant improvement in domain-specific language generation, with BLEU scores exceeding expectations on validation samples.
This work demonstrates that strategically fine-tuned LLMs can serve as foundational models for AI-driven material discovery knowledge bases, potentially accelerating hypothesis testing, experimental planning, and scientific reasoning. Future directions include integrating regression-based materials property prediction benchmarks and open-domain QA fine-tuning for broader applicability.
Courses on how AI applied to education.
Artificial Intelligence (AI) is transforming the way we live, work, and learn. In this session, students and educators will explore the basics of AI, including how machines learn from data and mimic human thinking. Using fun examples and interactive activities, we will demystify concepts like machine learning and neural networks. Participants will discover how AI is already part of everyday life and why understanding AI is important for the future.
The search for new materials is key to innovations like better batteries, faster electronics, and sustainable technologies. This lecture reveals how AI is revolutionizing materials discovery by enabling scientists to predict and create new materials faster and more efficiently than ever before. Through real-world examples and exciting breakthroughs, students and educators will learn how AI-driven science is shaping the future and how they can be part of this cutting-edge field.
Prediction processing-microstructure-property (PMP) link is critical for material processing, characterization, and discovery. We demonstrate GAN-based machine learning models that can accurately predict PMP relationships, specifically in the prediction of (1) the microstructure of alumina under arbitrary laser power, (2) the expected microstructure from the desired hardness, (3) real-time, in-situ microstructure during laser manufacturing, and (4) phases and element distributions of multi-phase materials. In all these studies, the ceramic samples were fabricated using high-throughput, ultra-fast laser convergent manufacturing of ceramics. We demonstrate that experimentally-obtained data of processing paramters, microstructure, and properties were sufficient for training of large models that contain tens of millions of parameters. An independent neural network was developed and showed that ML-predicted microstructure had less than 10% error from real ones, in terms of projected hardness. To monitor the microstructure during laser sintering, we demonstrated an ML model that can instantaneously predict the ceramic’s microstructure at the laser spot, based on the laser spot brightness. The ML model can generate more than 10 predictions per second, and the error in average grain size was less than 5% from the experimental observations.
Harnessing microbial systems for methane bioconversion necessitates a detailed understanding of enzyme architecture and microbial-surface interactions. Integrating microbial biotechnology, materials engineering, and artificial intelligence (AI) offers a transformative strategy for advancing sustainable methane bioconversion. In this study, we computationally remodeled the particulate methane monooxygenase (pMMO) of a model methanotroph, Methylosinus trichosporium OB3b, and identified multiple distinct active sites based on docking with methane homologs and tunnel analyses. Targeted mutagenesis of key residues, including B:Leu31Ser, B:Phe92Thr, B:Phe96Gly, B:Trp106Ala, and B:Tyr110Phe, significantly improved binding energies compared to the wild-type enzyme, suggesting the potential for enhanced catalytic rates. Furthermore, a comprehensive pangenomic analysis of 75 Type II methylotroph genomes revealed 256 exact core gene families and elucidated metabolic diversity across the group. We identified conserved biofilm-associated genes, including Type IVb pili genes (pilT, pilB, pilQ), quorum sensing regulators, and adhesion systems, underscoring genetic determinants critical for microbial attachment and colonization on surfaces. Moreover, methylotrophs differentially used conserved serine and ethylmalonyl-CoA (EMC) pathways. Unique adaptations, such as the acquisition of both glyoxylate and EMC pathway genes in Methylovirgula sp. 4MZ18, highlight the metabolic flexibility of these organisms under variable environments.
These discoveries lay the groundwork for engineering pMMO variants as immobilized catalysts on material surfaces, enabling the creation of efficient biointerfaces for methane oxidation. Building on this foundation, we aim to rationally design methanotroph-material systems that integrate enzymatic catalysis with microbial colonization dynamics. Realizing this vision will require AI and ML frameworks to predict enzyme-surface conjugation, model biofilm regulatory networks, and optimize material properties aligned with microbial functionalities. By integrating AI-driven predictive tools with experimental insights, we set the stage for next-generation smart catalytic surfaces, biomanufacturing platforms, and carbon capture technologies.
Sulfate-reducing bacteria (SRB) thrive in many natural environments, deep environments, and processing facilities in industrial settings and form biofilms. SRB biofilms alter the physiochemical properties of metals, inducing fouling and later biocorrosion, which cost USD 1.1Trillion to global GDP every year. To combat biocorrosion, it is essential to understand the interaction at the bacteria (biofilm)-metal surface interface. Biofilms grow on the likeliness (metal types, surface characteristics, atomic orientation, lattice, grain boundaries, grain energies) of the bacteria, a critical factor that could affect SRB biofilms, altering the biofilm-metal interface interactions.
Identification of the biofilm-metal interface interaction schemes will lead us to design a next-generation metal that will not allow biofilm formation and ultimately save resources. We hypothesize that variation in atomic lattice orientation and physical grains and grain boundaries corresponds to different surface energies, and which may affect (attract or repel) bacterial attachment for biofilm formation. To understand the interaction between the metal surface and bacteria during initial attachment and biofilm formation. We used three different types of surfaces of Copper (Cu) viz, Bare-Cu, Annealed-Cu, and CzCu, to grow biofilm of a SRB strain, Oleidesulfovibrio alaskensis G20. An anaerobic bioreactor (CDC-Bioreactor) was used to grow the biofilm for 7 days at 30°C. Biofilm was harvested and analyzed using Scanning Electron Microscopy and Confocal Laser Scanning Microscopy. It is observed that the Bare-Cu surface has the highest biofilm, followed by AnCu and CzCu. It aligned with our hypothesis that surface characters affect bacterial interactions, as BaCu has the highest number of grains, followed by AnCu and CzCu, which corresponds to their surface energies. The CzCu has the lowest number (35) of grains. High-quality, biofilm images will be used to understand the biofilm-metal interface interactions, and to develop a Unet AI model. The AI model will integrate the metal characteristics data (grains, grain boundaries, and grain energies) and surface data (before and after the biofilm growth) to predict and design a surface with desired modifications to avoid SRB biofilms.
Sulfate-reducing bacterial (SRB) biofilms are prevalent across natural and engineered environments, mediating biogeochemical sulfur cycling while accelerating biofouling and microbiologically influenced corrosion (MIC). To uncover surface-dependent biofilm adaptations, we performed a comparative transcriptomic analysis of Oleidesulfovibrio alaskensis G20 biofilms formed on copper and polycarbonate surfaces. RNA sequencing revealed 1,255 differentially expressed genes, with copper-grown biofilms exhibiting upregulation of Dde_1570 (flagellin; log₂FC 2.31) and Dde_0831 (polysaccharide chain length determinant; log₂FC 1.15), indicative of enhanced motility and extracellular polymeric substance production. In contrast, stress-related genes, including Dde_0132 (Cu/Zn efflux transporter; log₂FC -3.37), were downregulated on copper, reflecting metabolic adaptation to heavy metal exposure. Morphological characterization via SEM and AFM revealed denser biofilm clustering on copper surfaces and significant surface roughness increases (4.6-fold and 3.8-fold on copper and polycarbonate, respectively). Protein-protein interaction analysis underscored the importance of ribosomal synthesis, folate metabolism, and quorum sensing in biofilm resilience. Functional annotation further identified novel biofilm regulators, including Dde_4025 and Dde_3288.
Building on these results, we propose an artificial intelligence (AI)-driven framework to predict and optimize biofilm responses on engineered surfaces. By integrating transcriptomic profiles, morphological descriptors, and functional annotations into machine learning models, surface-specific biofilm behaviors can be classified and forecasted. Feature importance analysis using explainable AI methods will pinpoint critical genes and morphological features that govern microbial-material interactions. This predictive pipeline will accelerate the discovery and rational design of novel antifouling materials, including corrosion-resistant alloys, engineered polymer composites, nanopatterned metals, and smart stimuli-responsive surfaces, thereby offering sustainable solutions to mitigate MIC and enhance material longevity.
Relaxor ferroelectric Lead Magnesium Niobate-Lead Titanate (PMN-PT) crystal has been used in various applications such as medical ultrasound imaging, SONAR, micro-actuation, and energy harvesting due to its high electromechanical coupling coefficient and high piezoelectric coefficient. However, the high cost of PMN-PT stems from low crystal growth yield and limits the wider its application. Vertical gradient freeze (VGF) is a crystal growth method requiring relatively simple equipment and producing high quality crystals with low dislocation densities and less impurity striations and facets. This method has been extensively used to grow semiconductor and optical crystals. Here, we report the growth of ϕ 6” PMN-PT crystals by vertical gradient freeze (VGF) technology. COMSOL was used to simulate the furnace thermal field to help increase the furnace simulation to optimize the furnace geometry and increase the efficiency of experiment. The crystal structure, dielectric and piezoelectric properties of grown crystals were studied.
This project is supported by Office of Naval Research (ONR) Contract N00014-21-C-1037, and Luxium Solutions.
MXenes, owing to their unique morphology, high surface-to-volume ratio, and metallic conductivity, have gained significant attention as promising materials for gas sensing. Conventional MXene-based sensors primarily utilize electrical conductivity for signal transduction, but alternative mechanisms, such as mass-sensitive detection, can further enhance their selectivity and stability. Micro-quartz tuning forks (MQTF) offer an excellent platform for mass-sensitive gas sensing due to their high mechanical quality factor, stable resonance frequency, low power consumption, and compact size. By leveraging MQTF’s frequency shift mechanism, MXene-based gas sensors can achieve improved sensitivity and selectivity without concerns about conductivity degradation. In this study, we developed and optimized Ti₃C₂Tₓ MXene-functionalized MQTF gas sensors by modifying the surface chemistry of MXene to selectively detect CO, SO₂, and NH₃. Functionalization with -NH₂ and -F groups enabled tunable interactions with specific gases, significantly enhancing sensing performance. The Ti₃C₂Tₓ-NH₂ sensors demonstrated high selectivity for SO₂, while Ti₃C₂Tₓ-F sensors exhibited the strongest response to CO. Furthermore, increasing the surface modification temperature from 25 to 60 °C doubled the sensitivity of Ti₃C₂Tₓ-NH₂ for SO₂ detection. These findings highlight the importance of surface chemistry engineering in MXene-based gas sensors, providing a scalable strategy for designing highly selective and sensitive sensors. This work advances MXene’s potential for air quality monitoring, wearable electronics, the Internet of Things (IoT), and robotic applications. Further advancements in AI-powered materials discovery could accelerate the identification of optimal MXene compositions and functionalization strategies, enabling the design of next-generation gas sensors with unprecedented performance.
MXenes, a rapidly growing class of two-dimensional materials, are known for their diverse electronic, mechanical, and optoelectronic properties. Among these, double-transition-metal MXenes, such as Cr₂TiC₂Tx, offer even greater tunability in their properties due to the presence of two distinct transition metals. Cr₂TiC₂Tx stands out as a unique material among the MXenes experimentally tested, exhibiting p-type conductivity and a positive photoresponse. In this work, we performed density functional theory calculations to investigate the tunability of the electronic properties of Cr₂TiC₂Tx. We systematically examined the effects of Cr vacancies, oxygen substitution at carbon sites, and surface terminations. Our results show that Cr vacancies enhance p-type conductivity while oxygen substitutions induce n-type behavior, demonstrating tunable p-/n-type conductivity. These theoretical insights, supported by experimental findings, illustrate Cr₂TiC₂Tx as a promising material for advanced electronic and optoelectronic applications.
Achieving high-purity, detector-grade germanium (HPGe) crystals is essential for rare-event physics experiments such as dark matter detection and neutrinoless double-beta decay searches. We present a predictive machine learning framework that leverages Long Short-Term Memory (LSTM) networks to forecast the detector-grade yield fraction in HPGe crystal growth, based on experimentally obtained process data. The LSTM model is trained to predict detector-grade portions and assess the relative impact of input parameters, with interpretability enabled via SHAP (SHapley Additive exPlanations) analysis. Initial results demonstrate moderate-to-high predictive accuracy, revealing nontrivial correlations between impurity-related parameters and growth outcomes. To enhance the physical consistency of the input data, a physics-based impurity segregation model is incorporated to refine feature representations. The trained model is also employed in an inverse mode to identify optimal input parameter ranges that maximize detector-grade yield. This study provides a data-driven foundation for increasing the yield of detector-grade HPGe in each growth cycle by utilizing LSTM model–predicted input parameters for process control and optimization.
Nonequilibrium phenomena in quantum materials represent an exciting research frontier, in which theoretical insights are critical for both understanding cutting-edge experiments and guiding the exploration and realization of transient quantum states. I will briefly review the main challenges for practical simulations of realistic condensed matter systems based on the propagation of many-body correlation functions, which directly relate to observables, but are hindered by the high dimensionality and temporal non-locality of many-body interactions. The formalism is, however, uniquely positioned to leverage new developments in numerical and AI-enabled techniques. In this talk, I will illustrate several approaches based on dynamic mode decomposition and operator learning methods. They drastically accelerate the nonequilibrium Green’s function dynamics, transforming the computationally expensive functional forms of the system evolution into efficient surrogate models with linear temporal scaling. These approaches, along with new theoretical advances, enable real-time prediction of observables. I will further outline new avenues for AI-driven solvers that retain physical interpretability and adaptability, and the possibility of their integration into new simulation frameworks.
Large language models (LLMs) for biology such as protein language models (pLMs) trained on a large corpus of protein sequences have shown unprecedented scalability and broad generalizability in a wide range of predictive modeling tasks, but their power has not yet been harnessed for predicting protein–nucleic acid binding sites, critical for characterizing the interactions between proteins and nucleic acids. Here, we present EquiPNAS, a new pLM-informed E(3) equivariant deep graph neural network framework for improved protein–nucleic acid binding site prediction. By combining the strengths of pLM and symmetry-aware deep graph learning, EquiPNAS consistently outperforms the state-of-the-art methods for both protein–DNA and protein–RNA binding site prediction on multiple datasets across a diverse set of predictive modeling scenarios ranging from using experimental input to AlphaFold2 predictions. Our ablation study reveals that the pLM embeddings used in EquiPNAS are sufficiently powerful to dramatically reduce the dependence on the availability of evolutionary information without compromising on accuracy, and that the symmetry-aware nature of the E(3) equivariant graph-based neural architecture offers remarkable robustness and performance resilience. EquiPNAS is freely available at https://github.com/Bhattacharya-Lab/EquiPNAS.
One of the fundamental properties of semiconductors is their ability to support electric currents in the presence of electric and magnetic fields. These properties are described by transport coefficients such as drift and Hall electron and hole mobilities. During the past decade, there has been considerable progress in first-principles atomic-scale calculations of these coefficients by combining density functional theory, many-body perturbation theory, and the Boltzmann transport equation. The reliability, accuracy, and reproducibility of these calculations keep improving at a fast pace, and we are at a point where state-of-the-art methods and software are ready for data-driven approaches and machine learning tasks. In this talk, I will describe the Boltzmann transport solver of the software package EPW, and I will report on recent work on the high-throughput search for high-mobility n-type and p-type 2D channel materials for nanoscale transistors, and our first attempts at incorporating machine learning tools such as random forest regression. In addition, I will discuss recent work in our group on the development of streamlined workflows for electron-phonon physics, namely the MATCSSI cloud integration and the EPWpy abstraction and automation library. MATCSSI is a joint effort between the University of Texas, the University of California at Berkeley, the University of Binghamton, and the Texas Advanced Computing Center (TACC) aimed at making advanced many-body electronic structure calculations more accessible and more streamlined. This cloud portal supports Jupyter Notebooks which are executed directly on TACC supercomputers through a custom-made JupyterHPC integration. The aim of this initiative is to lower the barrier to entry for new users interested in advanced electronic structure calculations, including fully first-principles calculations of transport coefficients. EPWpy is a Python library that allows users to define materials as objects, and supports a high-level syntax to specify ab initio calculation workflows as methods on such objects. EPWpy aims for a lightweight, user-friendly experience that is intuitive for both users and developers, enables automation for machine learning tasks, and can be used as a standalone platform or in combination with the MATCSSI cloud portal.
Two-dimensional (2D) materials have garnered significant attention due to their unique properties which result from their reduced dimensionality and quantum confinement. In this talk, I will present our recent research on the data-driven discovery of various architectures of atomically thin materials for photocatalytic splitting of water for generating clean hydrogen. One of the most daunting challenges in solar-driven hydrogen conversion is simultaneously achieving high efficiency as well as long-term durability in the highly corrosive conditions needed for the photocatalytic reactions. I will show how we predicted 100’s of robust 2D-bulk heterostructures and oxide nanoscroll photocatalysts using big data, machine learning, and high-throughput first-principles computations. We evaluated the intrinsic properties of these 2D systems, such as their electronic properties, optical absorbance, and solubility in aqueous solutions to ascertain their efficacy in water-splitting. Finally, I discuss future research directions and the opportunities for methodological developments to enable computational design and optimization of 2D materials systems for photocatalytic applications.
The development of new materials and manufacturing processes is increasingly driven by advances in Artificial Intelligence (AI). In materials science, evaluating the properties of candidate designs often relies on computationally intensive simulations, which can become impractical when exploring vast design spaces. Machine learning offers a powerful alternative: surrogate models can approximate material properties with significantly reduced computational cost compared to traditional simulations. By leveraging mature machine learning workflows developed in computer science, materials researchers can accelerate discovery and optimization processes.
In this talk, I will present the application of Bayesian optimization across a diverse set of materials science collaborations, including alloy design, laser-induced graphene synthesis, and additive manufacturing of both polymers and metals. While each application domain expert supervision, they can be addressed with similar machine learning frameworks. I will also introduce a user-friendly, web-based interface that enables researchers without programming expertise to deploy and experiment with advanced machine learning models, democratizing access to these powerful tools.
Bio
Dr. Patrick Johnson recently joined the Materials Science and Engineering Department at Iowa State University after 17 years in the Chemical and Biomedical Engineering Department at the University of Wyoming. He has a research program that focuses on nanoscale materials, with a focus on advanced carbon species. Research topics include tuning these materials for applications in surface enhanced Raman spectroscopy (SERS) biosensors, additive manufacturing, and next generation batteries.
Patrick received his BS in Chemical Engineering from Lehigh University in 1992 and his Master’s in Biomedical Engineering from the University of Virginia in 1994. He then worked on projects for environmental remediation in Belarus followed by work on sensors for detection of biological and chemical weapons. He then received his PhD in Chemical Engineering from Columbia University in 2005.
Lunch break for all participants. Time for informal discussions and networking.
Courses on how AI applied to education.
Interdisciplinary STEM instruction holds value as a pedagogical approach but presents several challenges leading to reduced confidence and I-STEM teacher self-efficacy. While GenAI is a source of issues regarding academic integrity and learning assessment, it could also be the technology necessary to help educators increase their I-STEM teacher self-efficacy and confidence to developing complex I-STEM lessons. This presentation will discuss a research plan to study the effects of GenAI on I-STEM teacher self-efficacy among preservice and in-service educators after using GenAI as a collaborator in developing I-STEM lessons.
In this session participants will gain foundational understanding of the newly adopted K-8 Computer Science standards. Specifically highlighting and discussing computational thinking and AI-literacy, participants will have an opportunity to learn how schools are approaching these topics with students and staff.
The Haber–Bosch (HB) process is the foundation of industrial ammonia (NH₃) production, essential for manufacturing nitrate-based fertilizers and offering potential as a hydrogen carrier. However, the HB process consumes over 2% of global energy annually to produce more than 160 million tons of NH₃, primarily due to the high temperatures and pressures required by iron-based catalysts. In this presentation, we will first present a novel artificial intelligence framework that integrates deep reinforcement learning (DRL) with density functional theory (DFT) simulations to explore and evaluate complex catalytic reaction networks. By transforming first-principles-derived free energy landscapes into a dynamic DRL environment, the model autonomously evolves to identify optimal reaction pathways. Demonstrated through the Haber-Bosch process on the Fe(111) surface, this framework discovers pathways with lower energy barriers than traditional methods. Secondly, to solve the instability of the convergence issue of DRL in the chemical reactions, we introduce a reaction-agnostic framework, HDRL-FP, which combines high-throughput deep reinforcement learning with first principles DFT to explore catalytic reactions. The framework constructs a generalizable representation of reactions from atomic positions, mapping them to potential energy landscapes. HDRL-FP uses thousands of simulations on a single GPU to rapidly and cost-effectively identify optimal reaction pathways. Applied to hydrogen and nitrogen migration in the Haber-Bosch process on the Fe(111) surface, HDRL-FP reveals shared transition states between mechanisms and discovers pathways with lower energy barriers compared to traditional methods. This study presents a pathway for significantly enhancing the understanding of HB catalysts, offering a promising direction for reducing the energy demands of the ammonia synthesis process.
Tuning Soft Magnetic Properties in Fe-Based Nanocrystalline Alloys via Ge Substitution
Paul White, Department of Physics
Tula R. Paudel, Department of Physics
Soft magnetic materials play a crucial role in modern electrical and electronic devices, with ongoing research focused on enhancing their performance through compositional and structural modifications. In this study, we investigate Fe-based nanocrystalline alloys with Ge substitution for Si to improve their magnetic properties. Using a combination of experimental techniques and density functional theory (DFT) calculations, we explore the structural evolution, phase formation, and magnetization behavior of Fe72Nb4Cu1Si16-xGexB7 (x=0-16) alloys. Melt-spin casting was used to synthesize amorphous ribbons and isothermal annealing at 525°C for 30 min was used to grow nanocrystals which are surrounded by a residual amorphous matrix. X-ray diffraction (XRD) was performed on the as-cast samples to ensure amorphousness and on the nanocomposites to identify the crystalline phase. The shift in the primary XRD peak indicates that Ge substitutes for Si in the nanocrystals forming the Fe3Si1-yGey phase. Vibrating sample magnetometry measurements determined the magnetic saturation and coercivity of amorphous and nanocrystalline samples. Density functional theory calculations were able to explain the increase in magnetization through an increase in p-d hybridization with Ge-Fe compared to Si-Fe. Our findings demonstrate that Fe72Nb4Cu1Si16-xGexB7 (x=0-16) alloys exhibit enhanced magnetic saturation and small coercivity, highlighting their potential as improved alternatives to existing soft magnetic materials.
Germanium and tin both belong to group IV of the periodic table, but exhibit different properties when it comes to electrical conductivity. Ge is a semiconductor, with a direct and indirect bandgap of 0.8 eV and 0.67 eV respectively, while 𝛼-Sn is a (semi)metal. When combined as an alloy, the bandgap of GeSn can be tuned from 0.67 eV to 0 eV by varying the Sn content making it a useful material in optoelectronics within the near infrared (IR) to mid-IR ranges. However, the realization of these alloys is difficult because of the low solubility equilibrium (<1%) of Sn in Ge and the large lattice mismatch (~15%) between Ge and Sn.
In this study, we employ the cluster expansion methodology implemented in the Alloy Theoretic Automated Toolkit to predict 17 of the ground state structures of Ge1-xSnx in the complete x =0-1 range and compute their formation energies. We computed the bandgaps of the structures using the many-body perturbation theory method, GW0 method. We examined the roles of Sn clustering, using the Warren-Cowley short-range order parameter on the bandgap of the alloys. A comprehensive validation against experimental data in the literature reveals that the methods employed in this study accurately reproduce experimental observations as well as give rigorous insights into the structure, electronic properties, and the role of Sn clustering on the stability and band structures of the alloys.
Two-dimensional (2D) perovskites are promising materials for nonlinear optics, photonics, and optoelectronics due to their strong excitonic behavior, quantum confinement, and structural tunability. We investigate second-order nonlinear optical properties of mono- and few-layered (Benzylammonium)2PbX4 (where X = Br, Cl, and I), with our current focus on (Benzylammonium)2PbBr4. Bulk crystals were exfoliated to few-layered nanosheets and characterized their structural and compositional properties using X-ray fluorescence spectroscopy (XRF), X-ray diffraction (XRD), and energy-dispersive spectroscopy/scanning electron microscopy (EDS/SEM). Electronic properties were explored through density of states, band structure analysis, and dielectric response calculations under both independent particle approximation and local field effects. To further examine their optical and nonlinear properties, atomic force microscopy (AFM) will be used to quantify nanosheet thickness, while Raman spectroscopy and transmission electron microscopy (TEM) will provide insight into phonon modes, crystallinity, and structural integrity. Second-harmonic generation (SHG) and photoluminescence (PL) measurements will be performed to probe the nonlinear optical properties and emission characteristics of these materials. Moreover, we are theoretically calculating the second-order susceptibility to complement our experimental findings. Future work will extend these investigations to chloride and iodide analogues, allowing for a broader understanding of the tunable nonlinear optical properties in 2D perovskites for advanced photonic applications.
Two-dimensional (2D) materials placed on top of bulk substrates form complex heterostructures with rich interfacial physics with potential use in various applications such as nanoelectronics, sensing, and energy conversion. While numerous prior works have studied 2D-2D heterostructures, there have been relatively fewer studies that explore interfaces of 2D-bulk material heterostructures. In this study, we employ density functional theory (DFT) and machine learning (ML) models to investigate charge redistribution at the interface of over 1000 2D Janus-bulk heterostructures. We investigate the effect of metallic versus semiconducting substrates through electronic and Bader charge analyses of DFT-computed data. We then gain fundamental insights into the interfacial physics of 2D-bulk heterostructures by training predictive ML models to determine structure-property relationships of charge transfer and dipole moments across the interface. This work expands the currently available data of 2D-bulk heterostructures, thus enabling their use within a wide range of electronic, quantum computing, sensing, and energy applications.
Fouling release performance is a critical factor in marine coatings, influencing their effectiveness in preventing biofouling. In this study, Gradient Boosting Regressor (GBM) models were developed to predict fouling release properties based on experimental data for assays performed for C. lytica at 10 psi and 20 psi and N. incerta at 20 psi. The coating systems analyzed consisted of SBMA (sulfobetaine methacrylate) and PDMS (polydimethylsiloxane) in different molecular weights, which are widely used for their antifouling properties. The GBR models were trained using the weights of PDMS and SBMA in each system as weighting schemes and the traditional calculated molecular descriptors to transform into mixture-based molecular descriptors and then, use it as key input features to the GBR models. This mixture-based approach demonstrated high predictive accuracy, outperforming traditional regression models in terms of R² and, RMSE values. Feature importance analysis revealed that the difference in molecular weights of PDMS and SBMA influence fouling release behavior, providing valuable insights into structure-property relationships. To improve accessibility and practical implementation, a web application was developed, allowing users to input/tune different molecular weight values for SBMA and PDMS to obtain fouling release predictions in real time. This tool provides a reliable and user-friendly platform for researchers and industry professionals, facilitating the rational design and optimization of next-generation SBMA-PDMS fouling release coatings.
Phase change materials (PCMs), which can be reversibly switched between their high-resistance and low-resistance phases, are promising for non-volatile, high-density data storage and research in non-von Neumann computing architectures. Recent discovery and development of novel PCM superlattices consisting of various layers have demonstrated unprecedented low power and high-density at nanoscale. However, the properties of these layered PCMs are largely under-explored. In this work, the atomic and electronic properties of layered PCMs are explored using high-throughput Density Functional Theory (DFT) calculations. The physical understanding on the structure-property relationships will inform and train a machine learning model for the discovery of novel PCMs for energy-efficient memory devices.
Abstract
Identification of materials with two-dimensional properties is crucial for the creation of next-generation technologies, yet remains a computationally intensive task. The study presents an automated pipeline designed to predict if a material has two-dimensional characteristics from its physical and structural characteristics. Two-dimensional (2D) materials are a unique category of materials that are defined by their ultra thinness, usually just one or a few layers of atoms [1]. In contrast to traditional three-dimensional materials, 2D materials are extremely thin but extend significantly in the other two directions (length and width). They are important because they exhibit exceptional strength, electrical conductivity, flexibility, and surface sensitivity due to their atomic-scale thickness [2]. These unique properties enable breakthroughs in electronics, energy storage, medicine, and environmental technologies [1]. In biomedical applications, 2D materials are used for targeted drug delivery, biosensing, tissue engineering, and bioimaging [3] [4]. Their biocompatibility, high surface area, and tunable properties make them ideal for developing next-generation medical therapies and diagnostics [6].
We extracted key features from JSON files containing material properties and CIF files describing atomic structures, both of which were obtained from the Materials Project - a publicly available database for materials science research. These features were combined to classify each material based on its dimensionality, helping identify strong 2D material candidates. Random Forest is a machine learning algorithm that is mainly used for classification and regression tasks [5]. It works by creating many decision trees during training and then combining their results to make a more accurate and stable prediction. It learns patterns from material properties and votes to classify if the material is likely 2D or not [5] [7]. The Random Forest Classifier was trained with this data to predict the material dimensionality automatically. The model performance was evaluated with feature importance analysis, confusion matrix, and classification report with high predictive capacity. The dataset included a total of 120 samples, with 96 used for training and 24 for testing across three classes (0, 1, and 2). Testing on the test set resulted in a confusion matrix of 19 predicted correctly to be class 0, 2 as class 1, and 1 out of 3 as class 2, with class-wise accuracies of 100% (class 0), 100% (class 1), and 33% (class 2). Overall, the model achieved an accuracy of 92%, highlighting its reliability for 2D material screening.
References
[1] Butler, S. Z., et al. (2013). Progress, challenges, and opportunities in two-dimensional materials beyond graphene. ACS Nano, 7(4), 2898–2926. https://doi.org/10.1021/nn400280c
[2] Ward, L., et al. (2016). A general-purpose machine learning framework for predicting properties of inorganic materials. npj Computational Materials, 2(1), 1–7. https://doi.org/10.1038/npjcompumats.2016.28
[3] Gengxin Wu, Brea B. Yang & Ying-Wei Yang(2025). Two-dimensional (2D) materials for biomedical applications. APL Materials, 13(3), 030401. https://doi.org/10.1063/5.0261156
[4] A. Murali, G. Lokhande, K. A. Deo, A. Brokesh, and A. K. Gaharwar, “Emerging 2D nanomaterials for biomedical applications,” Materials Today, vol. 50, pp. 276–302, Nov. 2021, doi: 10.1016/j.mattod.2021.04.020. https://doi.org/10.1016/j.mattod.2021.04.020
[5] Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32. https://doi.org/10.1023/A:1010933404324
[6] D. Chauhan, M. Ashfaq, N. Talreja, and R. V. Managalraja, “2D materials for environment, energy, and biomedical applications,” J. Biomed. Res. Environ. Sci., vol. 2, no. 10, pp. 977–984, 2021, doi: 10.37871/jbres1340.
[7] Wang, Y., Xie, Y., Li, Y., & Wang, Y. (2021). Machine learning for 2D materials: From data mining to knowledge discovery. Materials Today, 47, 93–114. https://doi.org/10.1016/j.mattod.2021.05.001
The objective of our research is to investigate the electrocatalytic properties of novel metal-free quantum dots (QDs) composed of the recently discovered 2D material penta-CN2, with the aim of replacing costly and scarce catalysts such as Pt and Pd. Employing a first-principles density functional theory (DFT) based approach, the geometries of the three penta-CN2 quantum dots (QDs) of increasing sizes, 3 × 3, 3 × 4, 4 × 4 are optimized. Through comprehensive analysis, our research extensively explored the structural stability of penta-CN2 QDs, delved into their electronic properties, and assessed their catalytic performance concerning the Hydrogen Evolution Reaction (HER). Notably, the H-adsorbed penta-CN2 QDs exhibit a significant reduction in the HOMO-LUMO gap (Eg) ranging from 35% to 49% compared to the pristine QD. This observation underscores the crucial impact of H-adsorption on Penta-CN2 QD and is further supported by the appearance of mid-gap states in total and partial density of states plots. Next, we investigated their catalytic performance relevant to HER, using well-known descriptors: (i) adsorption energy, (ii) over-potential, (iii) Gibbs free energy and (iv) exchange current density along with the volcano curve. As far as size dependence of the catalytic performance is concerned, the value of the average change in Gibbs free energy,
, is minimum for 3 × 3 penta–CN2 QD, with those of 3 × 4 and 4 × 4 QDs being slightly larger. Our calculations predict a high value of exchange current density 2.24 x 10^(-3) A-cm−2 for one of the sites (N11 for 3 × 3 QD), which we believe will lead to significantly enhanced HER properties. The minimum value of delta G = 0.158 eV for a 3 × 3 penta-CN2 QD implies that its catalytic performance is at least as effective or perhaps better than most of the metal-free hybrid and non-hybrid structures. Our research outcomes hold great promise in advancing the discovery of abundant, non-toxic, and cost-effective catalysts for HER, playing a vital role in facilitating large-scale hydrogen production.
Courses on how AI applied to education.
Recent studies have identified titanium carbide (Ti2C) MXenes as promising 2D materials for detecting volatile organic compounds (VOCs) present in human breath. These VOCs reflect physiological changes and can serve as early biomarkers for diseases such as lung cancer. Diagnosing lung cancer through breath analysis offers a non-invasive and rapid alternative to conventional diagnostic techniques. In this study, we investigate the adsorption behavior of various VOCs on functionalized Ti2C monolayers using first-principles calculations. Our results reveal distinct adsorption energies and molecule monolayer distances, indicating strong chemisorption, particularly for the OH-functionalized Ti2C surface. Conductance and current-voltage (I-V) analyses further confirm the high selectivity of the OH-functionalized Ti2C monolayer toward aniline, due to its low work function and the electron-donating nature of aniline. To enhance the discovery and classification of VOCs biomarkers, we propose integrating machine learning algorithms trained in computed electronic, structural, and adsorption features. AI models will classify and predict VOCs interactions with various surface terminations, enabling high-throughput screening of candidate biomarkers. This combined approach of 2D materials design and AI-assisted analysis paves the way for advanced, non-invasive diagnostic platforms for early lung cancer sensing.
Laser-Induced Graphene (LIG) offers a cost-effective, scalable platform for electrochemical sensors, driving advancements in environmental monitoring, agriculture, and health diagnostics. Our studies focus on recent developments in LIG-based ion-selective electrodes for nutrient sensing in soil, supporting precision agriculture, and for non-invasive monitoring of metabolites and electrolytes in sweat, enhancing sports performance and health tracking. Additionally, we explore the functionalization of LIG with platinum nanoparticles and the surface tunability of electrodes to improve sensitivity for saliva analysis and nitrite detection in food safety. We also introduce Laser-Induced Graphene Microfluidic Integrated Sensors (LIGMIS), which combine microfluidics with LIG electrodes for real-time ion detection with high selectivity and low detection limits. These sensors enable multiplexed electrochemical detection of pesticides and ions in environmental water monitoring. Incorporating hydrophobic surface tuning and polyethyleneimine coatings, LIGMIS sensors demonstrate enhanced performance and long-term stability across a range of applications, from agriculture to wearable biosensing. This scalable, low-cost approach provides a promising solution for decentralized monitoring in precision agriculture, environmental, and health diagnostics. Looking ahead, we aim to further enhance the selectivity and electrochemical performance of the sensors through the integration of artificial intelligence (AI), which can potentially optimize sensor design by improving stability, sensitivity, and selectivity through data-driven insights.
Metal amorphous nanocomposites (MANC) possess properties distinct from crystalline materials due to their characteristic microstructure, which gives them excellent soft magnetic behavior. Recent research has focused on developing novel MANC compositions with superior magnetic and mechanical properties for next-generation lightweight inductor cores in electrified vehicles and rovers. In this study, we explored an alternative MANC composition of the commercially known alloy of FINEMET (Fe-Nb-Cu-Si-B), which achieves its excellent soft magnetic properties through Fe$_3$Si nanocrystals in the amorphous composite. Our alternative composition is Fe-Nb-Cu-Ga-B, where Ga has completely substituted for Si and will now form Fe-Ga nanocrystals. Fe-Nb-Cu-Ga-B ribbons were synthesized using melt-spin quenching which utilizes rapid cooling (~106°C/s) to form ribbons ~4mm wide by ~20μm thick and ~3m long. Their structural properties were characterized using x-ray diffraction (XRD) and small area electron diffraction (SAED), while magnetic properties were evaluated using vibrating sample magnetometry (VSM). VSM measurements show that Fe-Nb-Cu-Ga-B ribbons exhibit a significantly higher saturation magnetization—approximately 35% greater than traditional Fe-Nb-Cu-Si-B FINEMET alloys. Structural characterization from XRD and SAED found the presence of an ordered Fe$_3$Ga phase and a disordered Fe$_4$Ga phase. First-principles calculations were then performed to investigate the mechanism for the increased magnetization and found that the magnetic moment increased by 35% from Fe$_4$Ga to Fe$_3$Ga, correlating with the experimentally measured increase in magnetization. Based on the electronic density of states of the Fe-d orbitals, we determined that the Fe$_4$Ga phase has greater spin polarization than the Fe$_3$Ga phase. Combining theory and experiment, this integrated study provides insights into the potential of Fe-Nb-Cu-Ga-B alloys for future applications and advances the understanding of disordered nanocrystalline phases in metal amorphous nanocomposites.
Multi-component alloy systems based on copper (Cu), silver (Ag), titanium (Ti) and nickel (Ni) are employed in various industrial applications owing to their desirable mechanical, electrical, thermal and antimicrobial properties. Accurate prediction of mechanical properties such as melting points, hardness and modulus of elasticity is crucial for materials discovery, particularly alloy development for specific applications. In recent years, data-driven approaches have been adopted widely in materials research. This paper presents a machine learning model that utilizes the XGBoost algorithm for alloy melting point (°C), hardness (GPa) and modulus of elasticity (GPa) prediction. The model performance is based on XGBoost Regression utilizing training data sets obtained from ThermoCalc thermodynamic databases for melting points. Hardness and modulus of elasticity data was obtained by experimental data. A decision tree-based machine learning approach which combines the predictions of multiple decision trees iteratively is developed. The model is trained using gradient boosting which assigns a weight to decision trees at each iteration to minimize the loss function between predicted and theoretical values. We are able to estimate the melting points of various multi-component alloy systems accurately within 32°C of theoretical values and hardness and modulus of elasticity values within a mean absolute error percentage error of 11.7 and 11.6, respectively. The model presented herein outperforms the state of the art, and has a generalizable ability, and it can predict the mechanical properties of various alloy combinations of Cu, Ag, Ti and Ni. The ability to model intricate interactions among various elements enables the prediction of mechanical properties with higher precision across diverse materials systems. This generalized model can be used to predict the above mechanical properties for any combination of alloy systems.
Organic field effect transistors (OFETs) need to be simulated on a large scale for facilitating real time applications. To ensure that transistor-based circuits operate reliably, they need to be objectively modelled for all the physical events that occur during the operation. However, the existing models are yet to be comprehensive enough to address many of the diverse electrical behaviors of OFETs. For instance, some OFETs show the peculiar behavior of polarity conversion which implies the drain current increasing in the opposite direction during low values of source -drain voltage for n -type material and vice versa for the p-type material. So far, no existing models are capable of addressing this anomalous trend in such OFET characteristics. Here we report a solution to this particular problem through necessary modifications introduced to a widely used OFET model. Further, using extracted experimental data of real time OFETs showing polarity inversion characteristics, the modified model has been shown to be highly effective not only for this particular problem but as a universal OFET model for all type of characteristics.
The discovery of the Higgs boson at the Large Hadron Collider (LHC) through proton-proton collisions at CERN marked a major milestone in confirming the Standard Model (SM) of particle physics. While many SM parameters have now been measured with remarkable precision, the measurement of the triple Higgs coupling remains particularly challenging due to its extremely small production cross-section. High-luminosity runs of the LHC aim to improve the precision of this measurement. Alternatively, future colliders such as the FCC-ee, FCC-hh, and muon colliders are expected to offer enhanced sensitivity to the triple Higgs coupling.
In this talk, I will discuss how machine learning techniques can be used to improve the sensitivity of triple Higgs coupling measurements across hadron-hadron, electron-positron, and muon collider environments.
We present the quantum simulation of the frustrated quantum spin-$1/2$ antiferromagnetic Heisenberg spin chain with competing nearest-neighbor ($J_{1}$) and next-nearest-neighbor ($J_{2}$) exchange interactions in the real superconducting quantum computer with qubits ranging up to 100. In particular, we implement the Hamiltonian with the next-nearest neighbor exchange interaction in conjunction with the nearest-neighbor interaction on IBM's superconducting quantum computer and carry out the time evolution of the spin chain by employing the first-order Trotterization. Furthermore, our implementation of the second-order Trotterization for the isotropic Heisenberg spin chain, involving only nearest-neighbor exchange interaction, enables precise measurement of the expectation values of staggered magnetization observable across a range of up to 100 qubits. Notably, in both cases, our approach results in a constant circuit depth in each Trotter step, independent of the number of qubits. Our demonstration of the accurate measurement of expectation values for the large-scale quantum system using superconducting quantum computers designates the quantum utility of these devices for investigating various properties of many-body quantum systems. This will be a stepping stone to achieving the quantum advantage over classical ones in simulating quantum systems before the fault tolerance quantum era.
High purity germanium (HPGe) crystals play a critical role in the production of radiation detectors used in searches for rare events, where impurity concentrations need to be minimized to as low as 10¹⁰ atoms/cm³. In this presentation, I will discuss our latest research on HPGe crystal development, specifically designed for these highly sensitive applications, and investigate how AI tools such as ChatGPT and Grok can support this field.
Using these models, we conducted preliminary simulations of thermal profiles and dislocation densities that, although approximate, indicate a favorable correlation with experimental findings. I will address both the advantages and present constraints of chatbot-assisted modeling and illustrate how they can expedite initial design and hypothesis testing. Lastly, I will propose potential methods for incorporating more precise AI-driven simulations into material science processes.
Rare earth–benzene (RE-Bz, RE = La–Lu) complexes have attracted considerable attention due to their catalytic and optical properties. In this presentation, I will discuss our recent findings from a density functional theory(DFT)-based study that explores the geometries, ground-state spin multiplicities, stabilities, and magnetic properties of RE-Bz complexes.
We found that RE–benzene interactions can induce Jahn–Teller distortions, leading to lower symmetries such as C2, C2v, and C1. Eu and Yb show weak interaction with benzene due to half-filled or filled f-shells. Compared to isolated RE atoms, spin enhancement was observed in CeBz, PrBz, NdBz, ErBz, and TmBz, while spin quenching occurred in HoBz. Ongoing work includes vibrational frequency analysis and exploration of their optical properties.
Previous studies have demonstrated the successful integration of DFT and machine learning (ML) approaches to accelerate the prediction of material properties such as electronic, magnetic, and structural characteristics. Motivated by these advancements, we plan to implement similar ML-based strategies to predict the electronic, optical, and magnetic properties of RE–Bz complexes. These findings provide valuable insights into the structural stability and magnetic behavior of RE–Bz complexes, which are crucial for their potential applications in catalysis and spintronics.
Expert panel discussion focusing on the critical challenges in AI-powered materials discovery and strategies for fostering effective collaboration across disciplines and institutions. Open floor for audience questions and interactive discussion.
Panelists: Vladan Stevanovic, Vikram Gavini, Feliciano Giustino, Arunima Singh, Chenxu Yu, Dongming Mei
This talk will review recent applications of quantum machine learning to problems in high energy particle physics motivated by the analysis of data from the Large Hadron Collider at CERN, Geneva. Typical tasks include the classifications of jets as quarks or gluons; the classification of calorimeter clusters as electrons or photons; generative modelling of fragmentation and hadronization in jets; and representation learning. The explored hybrid quantum architectures include: quantum equivariant deep neural networks, quantum equivariant graph neural networks, quantum transformers, quantum diffusion models, quantum GANs, etc.
The IceCube Neutrino Observatory, located at the geographic South Pole, has opened a new window into the universe through the detection of high-energy astrophysical neutrinos. This contribution presents recent advances in the application of machine learning (ML) techniques to key challenges in astroparticle physics using IceCube data. We highlight the role of deep learning in improving event reconstruction accuracy, real-time alert systems for multimessenger follow-up, the enhanced classification of neutrino events against atmospheric backgrounds, and the advancements in cosmic ray measurements. Techniques such as graph neural networks, convolutional architectures adapted to irregular detector geometries, recurrent neural networks, and uncertainty-aware models have demonstrated significant improvements in performance and interpretability. We further discuss recent scientific results made possible through these approaches. This contribution underscores the growing importance of AI in advancing the scientific reach of large-scale observatories in the future.
The extrusion dynamics in additive manufacturing (AM) processes
such as electrohydrodynamic (EHD) printing are dependent on a large (>15) set of
processing parameters, material properties, and environmental conditions.
In EHD printing, these parameters affect the process stability, printing
behavior (droplet or filament), and quality of the deposited microstructures. In
this work, we have utilized high-speed photography to capture the ink flow
dynamics at the EHD printer nozzle tip. Later, machine learning tools such as Long Short-Term Memory (LSTM) networks and transformers were trained on high-speed video data to predict the ink flow dynamics for a new set of processing and material parameters. It was shown that combined high-speed imaging and machine learning can
establish the processing-property relationship during AM process while laying the
foundation for real-time control of printing behavior and process stability.
Summary of the workshop, closing speech, and discussion of future events and directions.
Lunch break for all participants. Time for informal discussions and networking.