Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “agent”

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 145 records · Page 8
All sourcesNASA NTRSDOE OSTI
NASA NTRS2003

Agent-Supported Mission Operations Teamwork

This slide presentation reviews the development of software agents to support of mission operations teamwork. The goals of the work was to make automation by agents easy to use, supervise and direct, manage information and communication to decrease distraction, interruptions, workload and errors, reduce mission impact of off-nominal situations and increase morale and decrease turnover. The accomplishments or the project are: 1. Collaborative agents - mixed initiative and creation of instructions for mediating agent 2. Methods for prototyping, evaluating and evolving socio-technical systems 3. Technology infusion: teamwork tools in mISSIons 4. Demonstrations in simulation testbed An example of the use of agent is given, the use of an agent to monitor a N2 tank leak. An incomplete instruction to the agent is handled with mediating assistants, or Intelligent Briefing and Response Assistant (IBRA). The IBRA Engine also watches data stream for triggers and executes Act-Whenever actions. There is also a Briefing and Response Instruction (BRI) which is easy for a discipline specialist to create through a BRI editor.

Malin, Jane T.↗
NASA NTRS2005

Monitoring Agents for Assisting NASA Engineers with Shuttle Ground Processing

The Spaceport Processing Systems Branch at NASA Kennedy Space Center has designed, developed, and deployed a rule-based agent to monitor the Space Shuttle's ground processing telemetry stream. The NASA Engineering Shuttle Telemetry Agent increases situational awareness for system and hardware engineers during ground processing of the Shuttle's subsystems. The agent provides autonomous monitoring of the telemetry stream and automatically alerts system engineers when user defined conditions are satisfied. Efficiency and safety are improved through increased automation. Sandia National Labs' Java Expert System Shell is employed as the agent's rule engine. The shell's predicate logic lends itself well to capturing the heuristics and specifying the engineering rules within this domain. The declarative paradigm of the rule-based agent yields a highly modular and scalable design spanning multiple subsystems of the Shuttle. Several hundred monitoring rules have been written thus far with corresponding notifications sent to Shuttle engineers. This chapter discusses the rule-based telemetry agent used for Space Shuttle ground processing. We present the problem domain along with design and development considerations such as information modeling, knowledge capture, and the deployment of the product. We also present ongoing work with other condition monitoring agents.

Semmel, Glenn S.↗
NASA NTRS2012

Optimal Routing and Control of Multiple Agents Moving in a Transportation Network and Subject to an Arrival Schedule and Separation Constraints

We address the problem of navigating a set of moving agents, e.g. automated guided vehicles, through a transportation network so as to bring each agent to its destination at a specified time. Each pair of agents is required to be separated by a minimal distance, generally agent-dependent, at all times. The speed range, initial position, required destination, and required time of arrival at destination for each agent are assumed provided. The movement of each agent is governed by a controlled differential equation (state equation). The problem consists in choosing for each agent a path and a control strategy so as to meet the constraints and reach the destination at the required time. This problem arises in various fields of transportation, including Air Traffic Management and train coordination, and in robotics. The main contribution of the paper is a model that allows to recast this problem as a decoupled collection of problems in classical optimal control and is easily generalized to the case when inertia cannot be neglected. Some qualitative insight into solution behavior is obtained using the Pontryagin Maximum Principle. Sample numerical solutions are computed using a numerical optimal control solver.

Sadovsky, A. V.↗
DOE OSTI2023

A Novel Multi-Agent Deep Reinforcement Learning-enabled Distributed Power Allocation Scheme for mmWave Cellular Networks

We consider the power allocation problem over shared spectrum for millimeter-Wave (mmWave) cellular downlink. Existing approaches usually find sub-optimal solutions by solving a non-convex optimization which leads to scalability issues due to centralized control. Therefore, distributed and adaptive approaches are desirable. Recently, model-free Deep Reinforcement Learning (DRL) has achieved success in such wireless resource management tasks. By modeling the radio environment as a Markov Decision Process (MDP) with the base stations (BSs) being the agents, power allocation can be automated at the agent level with comparable throughput performance to conventional centralized schemes. The multi-agent setting presents new challenges as the radio environment is impacted by the joint actions of the agents and is no longer stationary from any individual agent’s perspective. Existing literature bypasses this non-stationarity violation by ignoring it which may cause performance degradation. To tackle this issue, we propose a distributed continuous power allocation scheme based on a modified version of multi-agent Deep Deterministic Policy Gradient (MADDPG) that is tailored for the distributed multiple-agent setting. The proposed scheme employs a centralized-training distributed-execution framework where Q-functions are trained over subsets of BSs while each BS determines its transmit power based only on its own local observation. It admits constant per-BS communication and computation complexity and is thus scalable to large networks. Numerical evaluation shows that the proposed scheme adapts well to a wide range of interference conditions and can achieve comparable or better performance than several state-of-the-art non-learning approaches.

99 GENERAL AND MISCELLANEOUS↗
DOE OSTI2018

Global optimal consensus for discrete-time multi-agent systems with bounded controls

In this paper, we consider the global optimal consensus problem for discrete-time multi-agent systems with bounded control protocols over xed undirected communication networks. Each agent is described by a discrete-time single integrator and endowed with a quadratic objective function which is private to itself. For each agent, we develop two bounded distributed protocols: bounded proportional-integral (PI) protocol and bounded integral (I) protocol, based on information receiving from its neighboring agents through the communication network and the gradient of its own objective function. We show that the proposed bounded distributed protocols with properly chosen parameters solve the global optimal consensus problem if the communication network is connected, i.e., the multi-agent system achieves consensus at a state that minimizes the sum of the objective functions of all agents. The theoretical results are illustrated by numerical simulations.

Consensus, multi-agent system, optimal control↗
DOE OSTI2025

Automatic building energy model development and debugging using large language models agentic workflow

Building energy modeling (BEM) is a complex process that demands significant time and expertise, limiting its broader application in building design and operations. While Large Language Models (LLMs) agentic workflow have facilitated complex engineering processes, their application in BEM has not been specifically explored. This paper investigates the feasibility of automating BEM using LLM agentic workflow. Here, we developed a generic LLM-planning-based workflow that takes a building description as input and generates an error-free EnergyPlus building energy model. Our robust workflow includes four core agents: 1) Building Description Pre-Processing, 2) IDF Object Information Extraction, 3) Single IDF Object Generator Suite, and 4) IDF Debugging Agent. These agents divide the complex tasks into manageable sub-steps, enabling LLMs to generate accurate and reliable results at each stage. The case study demonstrates the successful translation of a building description into an error-free EnergyPlus model for the iUnit modular building at the National Renewable Energy Laboratory. The effectiveness of our workflow surpasses: 1) naive prompt engineering, 2) other LLM-based workflows, and 3) manual modeling, in terms of accuracy, reliability, and time efficiency. The paper concludes with a discussion on the interplay between foundational models and LLM agent planning design, advocating for the use of fine-tuned, specialized models to advance this field.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2026

Hierarchical Multi-agent Large Language Model Reasoning for Autonomous Heterogeneous Catalyst Discovery

Artificial intelligence is reshaping scientific exploration, but most methods automate procedural tasks without engaging in scientific reasoning, limiting autonomy in discovery. We demonstrate that hierarchical agentic large language model reasoning can efficiently drive simulation and scientific exploration. Across two chemical applications, CO adsorption on Cu surface transition metal adatoms and on M–N–C catalysts, reasoning-guided exploration reduces required atomistic simulations by up to 90% relative to heuristic or random selection. Comparisons across single-agent, multi-agent, and stochastic baselines show that hierarchical strategies yield more coherent and information-efficient search trajectories. Reasoning traces reveal chemically grounded decisions that cannot be explained by semantic bias or stochastic sampling. We realize these agentic reasoning strategies in Materials Agents for Simulation and Theory in Electronic-structure Reasoning (MASTER), a multimodal system that translates natural language into density functional theory workflows. Altogether, multi-agent collaboration accelerates heterogeneous catalyst discovery and marks a step toward more autonomous, reasoning-guided scientific exploration.

30 DIRECT ENERGY CONVERSION↗
DOE OSTI2019

Using a coupled agent-based modeling approach to analyze the role of risk perception in water management decisions

Abstract. Managing water resources in a complex adaptive natural–human system is a challenge due to the difficulty of modeling human behavior under uncertain risk perception. The interaction between human-engineered systems and natural processes needs to be modeled explicitly with an approach that can quantify the influence of incomplete/ambiguous information on decision-making processes. In this study, we two-way coupled an agent-based model (ABM) with a river-routing and reservoir management model (RiverWare) to address this challenge. The human decision-making processes is described in the ABM using Bayesian inference (BI) mapping joined with a cost–loss (CL) model (BC-ABM). Incorporating BI mapping into an ABM allows an agent's psychological thinking process to be specified by a cognitive map between decisions and relevant preceding factors that could affect decision-making. A risk perception parameter is used in the BI mapping to represent an agent's belief on the preceding factors. Integration of the CL model addresses an agent's behavior caused by changing socioeconomic conditions. We use the San Juan River basin in New Mexico, USA, to demonstrate the utility of this method. The calibrated BC-ABM–RiverWare model is shown to capture the dynamics of historical irrigated area and streamflow changes. The results suggest that the proposed BC-ABM framework provides an improved representation of human decision-making processes compared to conventional rule-based ABMs that do not take risk perception into account. Future studies will focus on modifying the BI mapping to consider direct agents' interactions, up-front cost of agent's decision, and upscaling the watershed ABM to the regional scale.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗
NASA NTRS2026

Multi-Agent Swarm State of the Art Report

The Next-Generation Multi-Agent Swarm (NGS) Study conducted by NASA’s Ames Research Center for NASA’s Space Technology Mission Directorate (STMD) will develop a comprehensive understanding of emerging multi-agent swarm capabilities. The study aims to identify existing swarm capabilities and asses their potential for persistent lunar space situational awareness, surface monitoring, and distributed autonomy demonstrations. A key objective is to inform the design of a next-generation multi-agent swarm that can perform autonomous distributed remote sensing, position, navigation, and timing (PNT) services, automated deployment that leverages autonomy, edge computing, and interoperable networking to enable cooperative operations without the need for immediate human operation. This study will address specific shortfalls identified by STMD, including intelligent multi-agent constellations, autonomy, edge computation, position, navigation, and timing for small spacecraft, small spacecraft propulsion, and space situational awareness (1625, 1438, 1433, 1557, 1431, 1430, 1589). The NASA Ames Mission Design Center (MDC) will provide subject matter expertise to support systems engineering trades, while experts in autonomy and spacecraft swarms in NASA’s Intelligent Systems Division will lead the study and focus on identifying emerging next-generation swarm capabilities. The study objectives include: capturing the current state-of-the-art for multi-agent swarm capabilities, evaluating technologies and creating technology roadmaps, and developing at least one new technology demonstration mission concept. This initial NGS study report surveys the current state of the art in technology areas relevant for the next-generation multi-agent swarm design. Our primary focus is on surveying relevant deployed space systems1, supplemented with selective analysis of relevant proposed missions and technology developments that have yet to fly.

agent↗
DOE OSTI2020

Closing the Loops on Solar Photovoltaics Modules: An Agent-Based Modeling Approach for the Study of Circular Economy Strategies

Solar photovoltaics (PV) installed capacity have grown exponentially since the early 2000s (average annual growth rate of 50%). With 4,700 GW projected installed capacity by 2050, the volume of PV panels waste is also expected to become substantial. Though renewables are a sine qua non to the establishment of a truly circular economy (CE), the issue arising from their end-of-life management needs to be resolved. The management of PV end-of-life also represent a singular opportunity to create value, from recovered valuable, rare or critical materials (e.g. silver, tellurium, indium). Moreover, the PV case illustrates some of the current barriers to CE; for instance, the need for a common definition of waste (PV are defined as e-waste in the European Union but as general waste in the United States) and potential loss of innovations' advantages (PV average efficiency has continuously grown the past 10 years). In this study, an agent-based modeling (ABM) approach is proposed to simulate PV end-of-life management in the United States. The model explores how the decisions of the various actors involved in handling PV waste affect the quantities of PV that are reused, recycled, or land-filled. In the model, manufacturers, residential, and nonresidential PV owners, installers, and recyclers are represented by agents while governmental policies and regulations are treated as exogenous variables. PV-market data are used to define agents' characteristics (e.g., installed PV capacities or producing costs). Agents' decision rules draw on the literature related to industrial symbiosis (IS) and peoples' waste behaviors as they appear as the most widely used model to implement CE principles at a meso level. Specifically, the concepts of mutual trust between IS actors and knowledge about the IS philosophy are yielded to define the agents' decision process. The primary outputs of the ABM are the costs and volume of waste associated with each end-of-life pathway. Preliminaries results indicate recycling of residential PV is affected by the cost, perceived difficulty of recycling behaviors, and social norms. Moreover, the type of network connecting agents and the initial recycling rate strongly influence results. The model also highlights the crucial role of recycling behavior adoption: in case of early failure of PV, recycled volumes increase by about 40%. This represents an additional 5.7 billion USD from potential recovered silver. Finally, although results for the PV case are presented here, the developed ABM aims at being a general tool for the study of CE strategies and the CE transition. Furthermore, steps of this research include extending the model to encompass circular economy strategies (e.g., design for recycling or lifetime extension) and validating its general architecture from various case studies.

14 SOLAR ENERGY↗
DOE OSTI2024

Operando visualization of porous metal additive manufacturing with foaming agents through high-speed x-ray imaging

Porous metals find extensive applications in soundproofing, filtration, catalysis, and energy-absorbing structures, thanks to their unique internal pore structure and high specific strength. In recent years, there has been an increasing interest in fabricating porous metals using additive manufacturing (AM), leveraging its unique advantages, including improved design freedom, spatial material control, and cost-effective small-batch production. In this study, we conducted pioneering operando visualization of AM porous metal using a laser powder bed fusion (L-PBF) setup combined with a high-speed synchrotron x-ray imaging system. Single track printing experiments using Ti6Al4V (Ti64) combined with titanium hydride (TiH 2 ) and sodium carbonate (Na 2 CO 3 ) as foaming agents, with varying mixing ratios were performed under different processing conditions. Here. the results elucidate the dynamic development of porosity formation. The average pore size is significantly influenced by the particle size of foaming agents when pore coalescence is absent. For all foaming agent content tested in the current study, the number of pores is found to be more sensitive to changes in laser power than in laser scanning speed. Increasing linear energy density (increasing laser power or reducing laser scanning speed) promotes the foaming agent activation thereby porosity formation. However, high linear energy density skews pore distribution towards the surface despite forming deeper melt pools. In addition, the impact of additional factors including foaming agent's laser absorptivity and decomposition kinetics with respect to AM time scales should be carefully considered to avoid ineffective activation of foaming agents during the AM of porous metals.

36 MATERIALS SCIENCE↗
DOE OSTI2026

Design and performance of AI agents interfacing with an atomic layer deposition tool

In this work, we introduce the design of an atomic layer deposition (ALD) reactor augmented with an AI interface for autonomous materials synthesis. Our modular design encapsulates the particularities of the hardware behind a Python interface that communicates with the ALD control software via transmission control protocol. This interface is compatible with model context protocol interfaces used in agentic frameworks. We have integrated our tool with a simple AI agent that leverages a large language model to transform user-supplied queries into ALD processes that are then run in our reactor. Our approach uses a JavaScript object notation schema to encode ALD processes. Our experimental results show that the AI interface does not impose a significant overhead to our control software, at least within our fastest 10 ms scale. We also carried out a detailed evaluation of the agent performance using leading models in two classes of tasks: basic instruction and process discovery tasks, where the agent is presented with a target material and needs to identify the correct ALD process compatible with the reactor configuration. Despite the simplicity of our agent design, we observed that most of the advanced models excelled at the instruction tasks. However, only recent models, such as o1, o3, GPT-5, and Claude Opus 4, performed well in process discovery tasks. We also observed significant variability in the response for the hardest challenges. While the results obtained are promising, we identify areas where AI research could improve the performance of agents for ALD.

47 OTHER INSTRUMENTATION↗
DOE OSTI2024

Multi-Agent Graph-Attention Deep Reinforcement Learning for Post-Contingency Grid Emergency Voltage Control

Grid emergency voltage control (GEVC) is paramount in electric power systems to improve voltage stability and prevent cascading outages and blackouts in case of contingencies. While most deep reinforcement learning (DRL)-based paradigms perform single agents in a static environment, real-world agents for GEVC are expected to cooperate in a dynamically shifting grid. Moreover, due to high uncertainties from combinatory natures of various contingencies and load consumption, along with the complexity of dynamic grid operation, the data efficiency and control performance of the existing DRL-based methods are challenged. To address these limitations, we propose a multi-agent graph-attention (GATT)-based DRL algorithm for GEVC in multi-area power systems. Here, we develop graph convolutional network (GCN)-based agents for feature representation of the graph-structured voltages to improve the decision accuracy in a data-efficient manner. Furthermore, a cutting-edge attention mechanism concentrates on effective information sharing among multiple agents, synergizing different-sized subnetworks in the grid for cooperative learning. We address several key challenges in the existing DRL-based GEVC approaches, including low scalability and poor stability against high uncertainties. Test results in the IEEE benchmark system verify the advantages of the proposed method over several recent multi-agent DRL-based algorithms.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2025

Multi-Agent Hierarchical Deep Reinforcement Learning for HVAC Control With Flexible DERs

As electricity consumption in commercial and residential buildings continues to rise, reducing energy costs presents an increasing challenge. Heating, ventilating, and air-conditioning (HVAC) systems, which typically account for 40%-50% of a building's energy use, are prime targets for energy savings. Intelligent control of HVAC temperature through the exploitation of HVAC load flexibility brings significant potential to reduce energy consumption and electricity expenses. The nonlinear models of HVAC systems challenge traditional control methods, while the uncertainty introduced by HVAC load flexibility complicates distributed energy resource (DER) management using conventional optimal dispatch techniques. In response to these challenges, we propose a hierarchical multi-agent deep reinforcement learning (DRL) approach. The lower-level agents focus on balancing comfort and energy conservation, while the upper-level DRL agents optimize the use of DERs to reduce peak demand based on the control outcomes of the HVAC by the lower-level agents. Here, in the upper-level agents, we incorporate a multi-agent structure based on ensemble learning, which acts based on historical and current data without relying on precise load forecasting to address the delayed rewarding issue in DRL. This allows for the effective reduction of energy costs. The proposed method is tested using a real-world microgrid comprising 413 buildings in Southern California, and the results demonstrate that our approach can significantly reduce overall electricity bills while ensuring the comfort of consumers and residents.

24 POWER TRANSMISSION AND DISTRIBUTION↗
DOE OSTI2022

Robust Decentralized Learning Using ADMM With Unreliable Agents

Many signal processing and machine learning problems can be formulated as consensus optimization problems which can be solved efficiently via a cooperative multi-agent system. However, the agents in the system can be unreliable due to a variety of reasons: noise, faults and attacks. Providing erroneous updates leads the optimization process in a wrong direction, and degrades the performance of distributed machine learning algorithms. This paper considers the problem of decentralized learning using ADMM in the presence of unreliable agents. First, we rigorously analyze the effect of erroneous updates (in ADMM learning iterations) on the convergence behavior of the multi-agent system. We show that the algorithm linearly converges to a neighborhood of the optimal solution under certain conditions and characterize the neighborhood size analytically. Next, we provide guidelines for network design to achieve a faster convergence to the neighborhood. Here, we also provide conditions on the erroneous updates for exact convergence to the optimal solution. Finally, to mitigate the influence of unreliable agents, we propose ROAD , a robust variant of ADMM, and show its resilience to unreliable agents with an exact convergence to the optimum.

97 MATHEMATICS AND COMPUTING↗
DOE OSTI2014

SU-E-I-84: MRI Relaxation Properties of a Pre-Clinical Hypoxia-Sensitive MRI Contrast Agent

Purpose: A possible hypoxia-sensitive MRI agent, hexamethyldisiloxane (HMDSO), has been tried to image oxygen level in proton-based MRI (Kodibagkar et al, NMR Biomed, 2008). The induced changes of T1 (or R1) value by the HMDSO as the oxygenation level changes are the principle that the hypoxia agent is based on: the R1 increases as the oxygen level increases. However, as reported previously, the range of R1 values (0.1–0.3 s-1, corresponding to 3–10 s of T1) is not in the range where a regular MRI technique can easily detect the change. In order for this agent to be widely applied in an MRI environment, more relaxation properties of this agent, including T1 in the rotating frame (T1rho) and T2, need to be explored. Here, the relaxation properties of this agent are explored. Methods: A phantom was made with HMDSO, water and mineral oil, each of which was prepared with oxygen and nitrogen, and was imaged in a 3T MRI system. The T1 properties were explored by the inversion recovery (TR=3000ms, TE=65ms) while varying the inversion time (TI), and also by the fast-field-echo (TR=2 ms, TE=2.8ms) while varying the flip angle (FA). T1rho was explored with a 5-pulse spin-locking technique (TR=5000ms, TE=10ms, spin-lock field=500Hz) while varying the spin-lock duration. T2 was explored with multi-shot TSE (TR=2500ms) while varying TE. Results: With the variable FA and TI, the signals of HMDSO with oxygen and nitrogen change in a similar way and do not respond well by the change of oxygen level, which confirms the large T1 value of HMDSO. The T1rho and T2, however, have a better sensitivity. Conclusion: For the possible pre-clinical hypoxia MRI agent (HMDSO), the detection of T1 (or R1) changes may be more challenging than the detection of other relaxation properties, particularly T2, as the oxygen level changes.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2015

Exogenous contrast agents for thermoacoustic imaging: An investigation into the underlying sources of contrast

Purpose: Thermoacoustic imaging at microwave excitation frequencies is limited by the low differential contrast exhibited by high water content tissues. To overcome this, exogenous thermoacoustic contrast agents based on gadolinium compounds, iron oxide, and single wall carbon nanotubes have previously been suggested and investigated. However, these previous studies did not fully characterize the electric, magnetic, and thermodynamic properties of these agents thus precluding identification of the underlying sources of contrast. To address this, measurements of the complex permittivity, complex permeability, DC conductivity, and Grüneisen parameter have been made. These measurements allowed the origins of the contrast provided by each substance to be identified. Methods: The electric and magnetic properties of the contrast agents were characterized at 3 GHz using two rectangular waveguide cavities. The DC conductivity was measured separately using a conductivity meter. Thermoacoustic signals were then acquired and compared to those generated in water. Finally, 3D electromagnetic simulations were used to decouple the different contributions to the absorbed power density. Results: It was found that the gadolinium compounds provided appreciable electric contrast but not originating from the gadolinium itself. The contrast was either due to dissociation of the gadolinium salt which increased ionic conductivity or its nondissociated polar fraction which increased dielectric polarization loss or a combination of both. In addition, very high concentrations were required to achieve appreciable contrast, to the extent that the Grüneisen parameter increased significantly and became a source of contrast. Iron oxide particles were found to produce low but measurable dielectric contrast due to dielectric polarization loss, but this is attributed to the coating of the particles not the iron oxide. Single wall carbon nanotubes did not provide measurable contrast of any type. Conclusions: It is concluded that gadolinium based contrast agents, iron oxide particles, and single walled carbon nanotubes have little intrinsic merit as thermoacoustic contrast agents. Simple electrolytes such as saline which yield high contrast based on ionic conductivity provide much higher dielectric contrast per unit solute concentration and are likely to be significantly more effective as contrast agents.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2026

matsim-agents v1.0

matsim-agents is a multi-agent AI framework for atomistic materials simulation and discovery. It orchestrates large language models (LLMs), machine-learned interatomic potentials (MLIPs), and DFT codes into a single agentic loop running on laptops and DOE leadership-class supercomputers. MULTI-AGENT ORCHESTRATION A LangGraph state machine with three nodes: a Planner that converts a natural-language research objective into structured tasks; an Executor that dispatches atomistic tools and loops until the queue is empty; and an Analyst that summarizes results into a human-readable report. State is checkpointed after every step and human-in-the-loop gates can be inserted at any edge. HYPOTHESIS-DRIVEN DISCOVERY CHAT An interactive REPL (matsim-agents chat) that couples LLM dialogue with atomistic simulation. Chemical formulas are automatically detected in conversation turns and trigger a full crystal-phase exploration: structure generation → relaxation → stability scoring → result injection back into the conversation, creating a closed hypothesis-refinement loop. CRYSTAL PHASE ENUMERATION Given a composition, the phase explorer enumerates prototypes by stoichiometry: elemental (fcc/bcc/hcp/sc/diamond), binary 1:1 (rocksalt/CsCl/zincblende/ wurtzite/fluorite/rutile), ternary 1:1:3 (cubic perovskite), ternary 1:2:4 (perovskite + spinel), quaternary 1:1:2:6 (Fm-3m double perovskite). 2-D prototypes (graphene, h-BN, MoS2 2H/1T) and multilayer stacking are also supported via --include-2d and --num-layers. SUPERCELL GENERATION AND SITE DECORATION Auto-tiling to a minimum atom count (--min-atoms), explicit NxNxN tiling (--supercell), symmetry-distinct site decorations (--n-orderings), and isotropic lattice-scale sweeps (--lattice-scales) for volume bracketing. MLFF RELAXATION AND STABILITY SCORING HydraGNN (multi-headed GNN) drives structure relaxation via ASE with FIRE, BFGS, or BFGSLineSearch. Stability output: delta-E/atom ranking across phases and a max-residual-force dynamical-stability proxy. Other MLIPs (MACE, NequIP, Orb) can be plugged in through the same interface. DFT BACKENDS Quantum ESPRESSO pw.x and VASP 6.6 are first-class labellers. Both have validated GPU builds and SLURM/PBS launchers for three DOE platforms: Frontier (AMD MI250X, ROCm), Aurora (Intel PVC, oneAPI), Perlmutter (NVIDIA A100, CUDA). QE produces ~100 binaries (pw.x, ph.x, epw.x, ...). VASP supports scf, relax, vc-relax, and vc-relax-shape run types. ACTIVE-LEARNING LOOP matsim-agents al run CONFIG.yaml drives an iterative HydraGNN-DFT loop: MD generates candidates → ensemble/MC-dropout uncertainty selects the most informative → DFT labels them in parallel inside one allocation → dataset grows → HydraGNN retrains → repeat. DFT backend is a single YAML toggle (dft.backend: vasp | qe). LLM-generated seed structures are supported (no curated POSCAR library needed). Config uses ${VAR}, ${VAR:-default}, ${VAR:?msg} shell-style substitution for cross-user/cross-site portability. LLM BACKENDS Ollama (local, default), vLLM (HPC multi-GPU serving), OpenAI, Anthropic, HuggingFace Transformers+Accelerate. Selected at runtime via flag or env var with no code changes. HPC PORTABILITY Same Python entry points run on Frontier (ROCm 7.2), Aurora (oneAPI), and Perlmutter (CUDA 12). DFT and ML stacks are never co-loaded in the same shell; they couple through the scheduler and filesystem. Advanced multi-node launchers (serve, discovery-chat, single-relaxation, active-learning, QE warm-start) are provided for all three platforms. CODABENCH COMPETITION BUNDLE A self-contained benchmark: 159 atomistic test structures across 11 material classes, 5 tasks (formation energy, forces, ML relaxation, AI-DFT relaxation, phase stability ranking), public/private leaderboard split (30/70), and four ready-to-run baselines: MACE-MP-0, HydraGNN, UMA, AllScAIP.

Lupo Pasini, Massimiliano [Oak Ridge National Labo↗