Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “edge network”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 289 records · Page 16
All sourcesNASA NTRSDOE OSTI
DOE OSTI2021

Characterizing Topological Order with Matrix Product Operators

One of the most striking features of gapped quantum phases that exhibit topological order is the presence of long-range entanglement that cannot be detected by any local order parameter. The formalism of projected entangled-pair states is a natural framework for the parameterization of gapped ground state wavefunctions which allows one to characterize topological order in terms of the virtual symmetries of the local tensors that encode the wavefunction. In their most general form, these symmetries are represented by matrix product operators acting on the virtual level, which leads to a set of algebraic rules characterizing states with topological quantum order. This construction generalizes the concepts of $\mathsf G$- and twisted injectivity; the corresponding matrix product operators encode all topological features of the theory and provide a complete picture of the ground state manifold on the torus. We show how the string-net models of Levin and Wen fit within this formalism and in doing so provide a particularly intuitive interpretation of the pentagon equation for F-symbols as the pulling of matrix product operators through the string-net tensor network. Our approach paves the way to finding novel topological phases beyond string nets and elucidates the description of topological phases in terms of entanglement Hamiltonians and edge theories.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗
DOE OSTI2025

Knowledge Oriented Graph Unified Transformer (KOGUT) v0.1

KOGUT — Knowledge Oriented Graph Unified Transformer KOGUT implements the Relational Graph Transformer (RelGT) architecture for knowledge graph link prediction in biological domains, with a primary focus on microbial growth media prediction. While the original RelGT (arXiv:2505.10960) targets relational tables, time series, and multi-table databases, KOGUT adapts this architecture for heterogeneous biological knowledge graphs, providing first-in-class AI predictive models for microbial cultivation. Key Adaptations Beyond Original RelGT: - Knowledge Graph Focus: Applied to biological KGs with semantic node types (taxa, chemicals, media, phenotypes, environments) versus generic relational database tables, trained on the KG-Microbe knowledge graph (1.3M entities, 2.9M edges, 24 relation types). - Multimodal Node Encoding: Integrates node labels, categories, descriptions, and synonyms from KG metadata through learned embedding layers—adapting relational column features to graph node attributes with textual semantics. - Extended K-Hop Subgraph Strategy: Optimized neighborhood sampling (3-hop default, configurable up to 200 nodes) tuned for sparse biological networks, building on the original local-global attention framework with biological relation preservation. - Biolink Predicate Preservation: Type-specific transformations for 24 biological edge semantics (occurs_in, consumes, produces, has_phenotype, subclass_of) beyond standard relational foreign keys, enabling multi-relation link prediction. - Inductive Learning Support: Enables zero-shot predictions for novel taxa through feature-based embeddings (temperature, oxygen requirements, gram stain, cell shape), extending the original transductive relational benchmark scope to uncultured microorganisms. CheapSOTA Performance Optimizations (This Distribution): - VQ-EMA Centroid Attention: Vector quantization with exponential moving average for improved global context modeling (+5-10% MRR improvement). - HDF5 Precomputed Data Loading: One-time preprocessing of k-hop subgraphs to eliminate redundant graph traversals (2-5× training speedup). - Distributed Data Parallel Training: Multi-GPU support for scaling to larger knowledge graphs (tested on 4× NVIDIA A100 GPUs at NERSC Perlmutter). - Mixed Precision Training: Automatic mixed precision (AMP) for memory efficiency and faster training. Advantages Over Standard Knowledge Graph Embedding Models: Combines RelGT's proven multi-element tokenization (features, type, hop, structure) with graph-native biological representations, enabling interpretable link prediction across heterogeneous entities that standard embedding models (TransE, RotatE, ComplEx) and table-based transformers cannot directly model. Achieves near-perfect performance on microbial growth media prediction (MRR: 0.9966, Precision@1: 0.9932, Hit@10: 1.0000) while maintaining explainability through attention-based reasoning over biological pathways. Training Data: - KG-Microbe merged knowledge graph: 1,379,337 nodes, 2,960,472 edges - 24 biological relation types including taxonomic hierarchies, metabolic interactions, phenotype associations, and environmental relationships - Primary prediction task: Growth media suitability for microbial taxa (biolink:occurs_in, 50K edges) - Multi-relation capability: Predicts links for any of the 24 relation types, including chemical consumption/production, phenotype associations, and taxonomic classification Citation: Original RelGT Architecture: Dwivedi et al., "Relational Graph Transformer", arXiv:2505.10960, 2025 KOGUT Implementation: Knowledge Oriented Graph Unified Transformer for Microbial Growth Media Prediction Developed at Lawrence Berkeley National Laboratory (LBNL) Trained on NERSC Perlmutter supercomputer

Joachimiak, Marcin [Lawrence Berkeley National Lab↗
DOE OSTI2021

Cybersecurity Certification Recommendations for Interconnected Grid Edge Devices and Inverter Based Resources

Escalating deployment of PV and grid-edge devices on the distribution grid has increased the sustainability and efficiency of the electric grid. However, the increasing number of distributed energy resources (DERs) deployed creates a heightened cyber-physical interdependency on the distribution grid and thus creates more vectors for cyber-attacks to exploit through information and communication technology (ICT) systems and networks. For example, control signal packets can be modified, intercepted, or corrupted due to vulnerabilities in communication protocols used by microgrid controllers and grid edge devices for power control. Therefore, to mitigate and prevent cyber-attacks on grid edge devices and the inverter-based resources connected to the distribution grid, the U.S. Department of Solar Energy Technologies Office (SETO) awarded funding to the National Renewable Energy Laboratory and Sandia National Laboratory (SNL) to research, develop, and harmonize cybersecurity standards for Photovoltaic (PV) systems and for other kinds of DERs. To help develop a standard for DER cybersecurity, NREL established certification recommendations and test cases, in consensus with the solar industry and UL, for ensuring intrinsic design security for DERs. These recommendations were developed to bolster the cybersecure functionalities such as TLS, MAC, CRL, session resumption/renegotiation, and password, system, and service security management within the DER devices. The proposed test cases verify authentication, authorization, confidentiality, and data integrity for data and communications of DERs that use Transmission Control Protocol/Internet Protocol (TCP/IP). They were also developed to protect DER communications from eavesdropping, replay, man-in-the-middle, denial of service (DoS), spoofing through security certificates, least-privilege violation, and brute-force credentials. This report, which has been validated and reviewed by UL, expands upon those test cases to provide DER cybersecurity certification recommendations which increase DER resiliency and help to mitigate cyber-attacks. UL's collaboration with NREL and approval of this document will accelerate the adoption of a UL standard for DER cybersecurity.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2025

Even Higher-Level Synthesis: An Exploration of AI Hardware Accelerators using HLS4ML

With the rise of artificial intelligence, the popularization of deep learning, and a constantly evolving industry, the demand for flexible and efficient tools has never been greater. As algorithms grow more complex, their runtime and energy consumption increase exponentially. Customized hardware accelerators, long used for specific mathematical operations, remain essential for managing modern applications' computational and power demands. Hardware accelerators can speed up complex computations by orders of magnitude, but their manual design and verification processes are often challenging and time-consuming. High-Level Synthesis (HLS) provides a solution by transforming high-level algorithm descriptions, typically written in C++ or SystemC, into synthesizable RTL suitable for hardware implementation. This approach reduces development time for RTL engineers while offering flexibility beyond what traditional handwritten RTL can provide. We extended this capability to the machine-learning domain with the open-source framework hls4ml, which allows neural networks trained in Python frameworks like Tensorflow or PyTorch to be synthesized into efficient hardware representations for the traditional FPGA and ASIC flows. This breakthrough addresses the growing need for reduced design turnaround and easy verification of ML hardware accelerators with low latency and power efficiency constraints. During this tutorial, we will demonstrate how Python complements HLS by simplifying the ML design process, bridging the gap between software and hardware development. Attendees will explore how we translate neural networks modeled in Python into fixed-point C++ models suitable for HLS workflows. We will dive into strategies like Value-Range Analysis and Quantization-Aware Training, which optimize these designs for deployment and evaluate their accuracy, power consumption, and energy efficiency. To exemplify these concepts, experts from Fermilab will share their experiences applying this technology to high-energy physics experiments, where real-time, low-latency processing is critical. Over the years, Fermilab engineers have demonstrated how deep neural networks, optimized for hardware using hls4ml, can meet the stringent requirements of trigger systems at the CERN Large Hadron Collider. These systems rely on rapid decision-making to process immense data volumes while retaining only the most relevant events for further analysis. The application of hls4ml has also been extended to innovative technologies like smart pixel arrays. These smart pixels integrate ML inference capabilities directly into sensor devices, enabling localized data processing at the pixel level. This approach drastically reduces the need to transmit raw data to external processing units, significantly decreasing power consumption and latency. By embedding neural networks within the pixel architecture, the smart pixels can identify and prioritize relevant data in real time, providing a highly efficient solution for edge computing in scenarios such as particle detectors and imaging systems. Fermilab's work highlights the potential of hardware-accelerated ML in scenarios where both speed and power efficiency are mission-critical. Through this tutorial, attendees will gain valuable insights into the challenges and solutions of deploying ML in hardware. Understanding how HLS and hls4ml streamline the development of neural network-based hardware accelerators is fundamental for the industry's future. Participants will learn how these technologies are shaping the future of AI and scientific computing.

Di Guglielmo, Giuseppe [Fermilab]↗
DOE OSTI2022

Tunable Magnetic Labyrinth for Abrikosov Vortices

We study Abrikosov vortex dynamics in a superconducting layer covered with an array of T- and I-shaped magnetically soft permalloy elements. Application of the in-plane field H // polarizes thin permalloy bars forming the array and creates a periodic network of mutually perpendicular lines of the attractive or repulsive magnetic potential for vortices along the bar edges, which can be easily tuned by the rotation of H // . Magneto-optical imaging of the vortex trajectories confirms that such a reconfigurable magnetic potential dominates the vortex motion in a wide temperature range and can be a useful tool for manipulating vortices in fluxonic devices for low-loss microelectronics.

36 MATERIALS SCIENCE↗
DOE OSTI2020

Observational Study of the Thermodynamics and Morphological Characteristics of a Midlatitude Continental Cold Pool Event

Abstract Convectively generated cold pools are important to the Earth system as they exert strong controls on deep convective-storm initiation, intensity, and life cycle. Despite their importance, efforts to introduce such cold pool controls into weather and climate models lack guidance and/or physical constraints from cold pool observations. This work presents a detailed, purely observational analysis of a cold pool event that took place on 23–24 May 2011 in north-central Oklahoma. The characteristics of the cold pool, and the spatiotemporal evolution of the hydrometeors and dynamics in the proximity of the cold pool, are studied with high-resolution observations. The unprecedented dataset used in this work to study cold pool characteristics includes an enhanced network of surface weather stations, a high-temporal-frequency sounding array, and the NEXRAD and Atmospheric Radiation Measurement (ARM) Southern Great Plains radar networks. The potential use of NEXRAD surveillance scans to estimate height and propagation speed of the leading edge of the cold pool (LECP) is presented in this work. Manual identification and tracking of the LECP from NEXRAD imagery shows a spatial and temporal heterogeneity of the LECP properties. Surprisingly, over its detected life cycle, the LECP speed remains almost constant, even though the strength of the cold pool diminishes in time and its height varies. Radar analysis shows that pulses of graupel and hail within downdrafts in the convective system generating the cold pool appeared to be related to temporary increases in the LECP height.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2025

Diffusion Codes: Self-Correction from Small(er)-Set Expansion with Tunable Non-locality

Optimal constructions of classical LDPC codes can be obtained by choosing the Tanner graph uniformly at random among biregular graphs. We introduce a class of codes that we call ``diffusion codes'', defined by placing each edge connecting bits and checks on some graph, and acting on that graph with a random SWAP network. By tuning the depth of the SWAP network, we can tune a tradeoff between the amount of randomness -- and hence the optimality of code parameters -- and locality with respect to the underlying graph. For diffusion codes defined on the cycle graph, if the SWAP network has depth $\sim Tn$ with $T> n^{2β}$ for arbitrary $β>0$, then we prove that almost surely the Tanner graph is a lossless ``smaller set'' vertex expander for small sets up size $δ\sim \sqrt T \sim n^β$, with bounded bit and check degree. At the same time, the geometric size of the largest stabilizer is bounded by $\sqrt T$ in graph distance. We argue, based on physical intuition, that this result should hold more generally on arbitrary graphs. By taking hypergraph products of these classical codes we obtain quantum LDPC codes defined on the torus with smaller-set boundary and co-boundary expansion and the same expansion/locality tradeoffs as for the classical codes. These codes are self-correcting and admit single-shot decoding, while having the geometric size of the stabilizer growing as an arbitrarily small power law. Our proof technique establishes mixing of a random SWAP network on small subsystems at times scaling with only the subsystem size, which may be of independent interest.

Combinatorics (math.CO)↗
DOE OSTI2022

Solid State Power Substations (SSPS): A Multi-Hierarchical Architecture from Substation to Grid Edge

With the growing deployment of distributed generation, or power electronic interfaced renewable energy and storage technologies, the nature and behavior of the grid is changing. Synchronous machine-driven asset contributions to the generation mix are shrinking, leading to concerns regarding grid stability. Furthermore, the scale of smaller distributed PE resources needed for managing the electrical network could dwarf the existing system leading to more complex optimization problems and communication interconnections. This paper introduces the concept of a hierarchal system of controllers that spans the grid edge or the customer end to distribution scale substations or solid-state power substation (SSPS). This concept focuses on minimizing the number of interfaces and optimization considerations in the grid by clustering resources into nodes and hubs. The work validates the concept in a controller hardware-in-the-loop (cHIL) platform.

Chinthavali, Madhu Sudhan↗
DOE OSTI2023

A Medium-/Low-Voltage Joint State Estimator Through Linear Uncertainty Propagation

Traditionally, distribution system state estimations (DSSE) are challenged by the lack of measurements at both primary and secondary sides of the system. The widely available cable television (CATV) voltage sensors installed in low-voltage (LV) networks bring opportunities to achieve higher quality DSSE covering a broader area of the distribution network. This study proposes a medium-/low-voltage (MV/LV) joint distribution system state estimation approach using the untapped CATV measurements. It aims at addressing the need for system situational awareness at the grid edge while improving the estimation accuracy at both the primary and secondary sides compared to its disjointed counterpart. Linearized measurement functions and boundary condition uncertainty propagation rules are derived to ensure the computational efficiency and accuracy of the joint state estimator. Numerical experiments are conducted on an IEEE test feeder to demonstrate the efficacy of the proposed method and the value of CATV measurements.

joint state estimation↗
DOE OSTI2025

Probing Alkylpolyglycoside Hydrogen Bonding and Its Destabilization by Sulfonate Hydrotropes with X‑ray and Vibrational Spectroscopy

It has been proposed that intersurfactant H-bonding networks can produce highly stable foams, specifically those generated from solutions containing alkylpolyglycoside surfactants. In this work, we aim to characterize the presence and destabilization of these networks by introducing a hydrotrope (sodium p-toluene sulfate) at various concentrations into an alkylpolyglycoside (Glucopon 225 DK) surfactant containing solution. Solution surface properties are probed by using aerosol velocity map imaging X-ray photoelectron spectroscopy (A-VMI-XPS) and tensiometer measurements. Bulk properties below the surface are probed using Fourier transform infrared spectroscopy (FTIR) and C-edge near-edge X-ray fine structure spectroscopy (NEXAFS). The surface measurements provide the framework to describe the destabilization of the solution H-bonding network, while the bulk solution measurements provide hints about the disruption of the hydrogen bonding network upon hydrotrope addition. The collected data support the hypothesis that the destabilization of the intersurfactant H-bonding network frees surfactant molecules from the bulk, increasing surfactant population at the air/water interface. This was quantified through an increase in peak area and width in XPS measurements as well as a decrease in surfactant critical micelle concentration. Three regimes with increasing amounts of hydrotrope addition are suggested, described as (1) a hydrotrope affecting only surfactant surface properties, (2) a hydrotrope affecting surfactant surface and bulk properties, and (3) hydrotrope-dominated surface and bulk properties. Future studies will characterize foam stability across the hydrotrope concentration regimes to better define correlations between intersurfactant H-bonding networks and foam stability.

Molecules↗
DOE OSTI2022

Fast and Accurate Predictions of Total Energy for Solid Solution Alloys with Graph Convolutional Neural Networks

We use graph convolutional neural networks (GCNNs) to produce fast and accurate predictions of the total energy of solid solution binary alloys. GCNNs allow us to abstract the lattice structure of a solid material as a graph, whereby atoms are modeled as nodes and metallic bonds as edges. This representation naturally incorporates information about the structure of the material, thereby eliminating the need for computationally expensive data pre-processing which would be required with standard neural network (NN) approaches. We train GCNNs on ab-initio density functional theory (DFT) for copper-gold (CuAu) and iron-platinum (FePt) data that has been generated by running the LSMS-3 code, which implements a locally self-consistent multiple scattering method, on OLCF supercomputers Titan and Summit. GCNN outperforms the ab-initio DFT simulation by orders of magnitude in terms of computational time to produce the estimate of the total energy for a given atomic configuration of the lattice structure. We compare the predictive performance of GCNN models against a standard NN such as dense feedforward multi-layer perceptron (MLP) by using the root-mean-squared errors to quantify the predictive quality of the deep learning (DL) models. We find that the attainable accuracy of GCNNs is at least an order of magnitude better than that of the MLP.

Lupo Pasini, Massimiliano↗
DOE OSTI2021

Inference of Gene Regulatory Network Uncovers the Linkage between Circadian Clock and Crassulacean Acid Metabolism in Kalanchoë fedtschenkoi

The circadian clock drives time-specific gene expression, enabling biological processes to be temporally controlled. Plants that conduct crassulacean acid metabolism (CAM) photosynthesis represent an interesting case of circadian regulation of gene expression as stomatal movement is temporally inverted relative to stomatal movement in C3 plants. The mechanisms behind how the circadian clock enabled physiological differences at the molecular level is not well understood. Recently, the rescheduling of gene expression was reported as a mechanism to explain how CAM evolved from C3. Therefore, we investigated whether core circadian clock genes in CAM plants were re-phased during evolution, or whether networks of phase-specific genes were simply re-wired to different core clock genes. We identified candidate core clock genes based on gene expression features and then applied the Local Edge Machine (LEM) algorithm to infer regulatory relationships between this new set of core candidates and known core clock genes in Kalanchoë fedtschenkoi. We further inferred stomata-related gene targets for known and candidate core clock genes and constructed a gene regulatory network for core clock and stomata-related genes. Our results provide new insight into the mechanism of circadian control of CAM-related genes in K. fedtschenkoi, facilitating the engineering of CAM machinery into non-CAM plants for sustainable crop production in water-limited environments.

59 BASIC BIOLOGICAL SCIENCES↗
DOE OSTI2024

Multi-Level Structural Damage Characterization Using Sparse Acoustic Sensor Networks and Knowledge Transferred Deep Learning

Standard structural health monitoring techniques face well-known difficulties for comprehensive defect diagnosis in real-world structures that have structural, material, or geometric complexity. This motivates the exploration of machine-learning-based structural health monitoring methods in complex structures. However, creating sufficient training data sets with various defects is an ongoing challenge for data-driven machine (deep) learning algorithms. The ability to transfer the knowledge of a trained neural network from one component to another or to other sections of the same component would drastically reduce the required training data set. Also, it would facilitate computationally inexpensive machine learning based inspection systems. In this work, a machine-learning-based multi-level damage characterization is demonstrated with the ability to transfer trained knowledge within the sparse sensor network. A novel network spatial assistance and an adaptive convolution technique are proposed for efficient knowledge transfer within the deep learning algorithm. Proposed structural health monitoring method is experimentally evaluated on an aluminum plate with artificially induced defects. It was observed that the method improves the performance of knowledge transferred damage characterization by 50% during localization and 24% during severity assessment. Further, experiments using time windows with and without multiple edge reflections are studied. Results reveal that multiply scattered waves contain rich and deterministic defect signatures that can be mined using deep learning neural networks, improving the accuracy of both identification and quantification. In the case of a fixed sensor network, using multiply scattered waves shows 100% prediction accuracy at all levels of damage characterization.

36 MATERIALS SCIENCE↗
DOE OSTI2022

Short- to Intermediate-Range Structure, Transport, and Thermophysical Properties of LiF–NaF–ZrF4 Molten Salts

LiF–NaF–ZrF 4 multicomponent molten salts are identified as promising candidates for coolant salts in molten salt reactors and advanced high-temperature reactors. This study focused on low-melting point salt compositions of interest: 38LiF–51NaF–11ZrF 4 , 42LiF–29NaF–29ZrF 4 , and 26LiF–37NaF–37ZrF 4 . Ab-initio molecular dynamics (AIMD) calculations were performed and compared with available experimental data to assess the ability of rigid ion models (RIM) to reproduce short- to intermediate-range structure, transport, and thermophysical properties of the LiF–NaF–ZrF 4 salt mixtures. It is found that as ZrF 4 mol% increases, the average cation–anion coordination number (CN) of monovalent cations (Li + , Na + ) obtained from RIM calculations decreases, while multivalent Zr 4+ CN varied from 15% to 19% in comparison to corresponding AIMD values. In addition, RIM is found to predict the existence of 7, 8, and 9 coordinated fluorozirconate complexes, while AIMD and the available experimental data showed an occurrence of 6, 7, and 8 coordinated complexes in the melt. The intermediate-range structure analysis revealed that while the RIM parameters are able to reproduce a local structure for lower ZrF 4 mol% salts such as in 38LiF–51NaF–11ZrF 4 , an extensive fluorozirconate network formation is observed in RIM simulations for higher ZrF 4 mol% compositions. The network generated by RIM parameters is found to be mainly connected by “corner-sharing” fluorozirconate complexes as opposed to both “edge-sharing” and “corner-sharing” connectively portrayed by AIMD. It is found that a close agreement between AIMD and the RIM salt structure for the 11-mol% ZrF 4 salt resulted in good agreement in the calculated Zr diffusivities and the viscosity values. However, due to the inaccurate short- to intermediate-range structure prediction by RIM for higher ZrF 4 mol% compositions, thermophysical properties such as densities and heat capacity differ by up to 26% and 27%, respectively, upon comparison with AIMD and experimental values. Also, the network-dominated properties such as diffusion coefficients and viscosities differed by up to two and three orders of magnitude, respectively. This study signifies the importance of accurate salt structure generation for an accurate prediction of transport and thermophysical properties of multicomponent molten salts.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗
DOE OSTI2022

Design and Challenges of Edge Computing ASICs on Front-End Electronics

In situ or hardware-embedded data processing of raw signals, close to their source, in radiation detectors is expected to provide dramatic improvements in data quality and volumes. However, the implementation of artificial neural networks (ANNs) in the front-end electronics, and the design of custom integrated circuits (ASICs), comes with challenges. In addition, detectors have to operate with limited power budget and implement complex functionalities in a very dense space. They often are exposed to extreme conditions as they work in high-radiation environments and/or cryogenic temperatures. This paper presents examples of applications and design methodologies for in-situ ANNs, along with the challenges of retaining the fidelity of the trained networks. For illustration, we use the problem of estimating the energy deposited by the radiation from digitized waveforms. The proposed implementation starts with an ML algorithm trained in Qkeras and eventually leads to an equivalent ASIC implementation. The associated design challenges in realizing energy and area efficient implementations in CMOS processes are reviewed. Novel approaches that employ hybrid technologies (combination of CMOS with memristors), in-memory computing models, and bio-inspired spiking neural networks are also highlighted

47 OTHER INSTRUMENTATION↗
DOE OSTI2022

Spatial Graph Attention and Curiosity-driven Policy for Antiviral Drug Discovery

We developed Distilled Graph Attention Policy Network (DGAPN), a reinforcement learning model to generate novel graph-structured chemical representations that optimize user-defined objectives by efficiently navigating a physically constrained domain. The framework is examined on the task of generating molecules that are designed to bind, noncovalently, to functional sites of SARS-CoV-2 proteins. We present a spatial Graph Attention (sGAT) mechanism that leverages self-attention over both node and edge attributes as well as encoding the spatial structure --- this capability is of considerable interest in synthetic biology and drug discovery. An attentional policy network is introduced to learn the decision rules for a dynamic, fragment-based chemical environment, and state-of-the-art policy gradient techniques are employed to train the network with stability. Exploration is driven by the stochasticity of the action space design and the innovation reward bonuses learned and proposed by random network distillation. In experiments, our framework achieved outstanding results compared to state-of-the-art algorithms, while reducing the complexity of paths to chemical synthesis.

Wu, Yulun↗
DOE OSTI2022

SPATIAL GRAPH ATTENTION AND CURIOSITY-DRIVEN POLICY FOR ANTIVIRAL DRUG DISCOVERY

We developed Distilled Graph Attention Policy Network (DGAPN), a reinforcement learning model to generate novel graph-structured chemical representations that optimize user-defined objectives by efficiently navigating a physically constrained domain. The framework is examined on the task of generating molecules that are designed to bind, noncovalently, to functional sites of SARS-CoV-2 proteins. We present a spatial Graph Attention (sGAT) mechanism that leverages self-attention over both node and edge attributes as well as encoding the spatial structure - this capability is of considerable interest in synthetic biology and drug discovery. An attentional policy network is introduced to learn the decision rules for a dynamic, fragment-based chemical environment, and state-of-the-art policy gradient techniques are employed to train the network with stability. Exploration is driven by the stochasticity of the action space design and the innovation reward bonuses learned and proposed by random network distillation. In experiments, our framework achieved outstanding results compared to state-of-the-art algorithms, while reducing the complexity of paths to chemical synthesis.

Wu, Y↗
DOE OSTI2025

Electronic Configurational Transformation of Network Modifiers in Aluminate Glass above Megabar Pressures

Electronic responses of glasses under extreme pressures differ from those of crystalline analogs. Their distinct electronic environments are found in network formers with well-defined, covalent-bonded coordination environments (e.g., [4] Si and [4] Al) and in network modifiers with more disordered, ionic-bonded configurations (e.g., [5,6,7] Ca). Deciphering the evolution of the bonding environment of network modifier cations upon compression provides atomic insights into the pressure-driven hardening and transport properties of glasses. Despite the importance, in contrast to extensive efforts to uncover how network formers behave under pressure, considerable structural disorder around network modifiers makes it challenging to probe their electronic bonding environments under compression. Our understanding of the evolution of network modifiers above megabars is currently absent. Here, we report a discovery of highly densified electronic configurations of network modifier Ca in aluminate glass under extreme compression via the first inelastic X-ray scattering at the Ca L-edge up to 140 GPa. As evidenced by the prominent pressure-driven increases in electronic dispersion and delocalization, densified calcium environments are characterized by a decreased average Ca–O distance, the formation of highly coordinated calcium, a broader distribution of topological variables, and a greater distortion of Ca polyhedra above megabars. The spectral features for the Ca environments reveal significant electronic and bonding modifications, including pressure-driven increases in the ligand field interaction, the covalence characteristic of the Ca–O bond, and the electron–hole Coulomb interaction. These densification paths identify the electronic adaptation of network modifiers above megabars, shedding light on the origins of enhanced electron transport and the electron-storing capacity of glasses under pressure.

amorphous materials↗