Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “prompt optimization”

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 37 records · Page 2

Market optimization and technoeconomic analysis of hydrogen-electricity coproduction systems

Decarbonization efforts across North America, Europe, and beyond rely on variable renewable energy sources such as wind and solar, as well as alternative fuels, such as hydrogen, to support the sustainable energy transition. These advancements have prompted a need for more flexibility in the electric grid to complement non-dispatchable energy sources and increased demand from electrification. Integrated energy systems are well suited to provide this flexibility, but conventional technoeconomic modeling paradigms neglect the time-varying dynamic nature of the grid and thus undervalue resource flexibility. In this work, we develop a computational optimization framework for dynamic market-based technoeconomic comparison of integrated energy systems that coproduce low-carbon electricity and hydrogen (e.g., solid oxide fuel cells, solid oxide electrolysis) against technologies that only produce electricity (e.g., natural gas combined cycle with carbon capture) or only produce hydrogen. Our framework starts with rigorous physics-based process models, built in the open-source Institute for the Design of Advanced Energy Systems (IDAES) modeling and optimization platform, for six energy process concepts. Using these rigorous models and a workflow to optimally design each technology, the framework is shown to be capable of evaluating new and emerging technologies in varying energy markets under a plethora of future scenarios (i.e., renewables penetration, carbon tax, etc.). Ultimately, our framework finds that solid oxide fuel cell-based coproduction systems achieve positive profits for 85% of the analyzed market scenarios. From these market optimization results, we use multivariate linear regression (R 2 values up to 0.99) to determine which electricity price statistics are most significant to predict the optimized annual profit of each system. The proposed framework provides a powerful tool for directly comparing flexible, multi-product energy process concepts to help discern optimal technology and integration options.

08 HYDROGEN↗

Spatial-Temporal PV Hosting Capacity Estimation and Evaluation

Evaluating Photovoltaic Hosting Capacity (PVHC) is an essential step in the process of integrating solar energy into power grids, particularly when focusing on the distribution network (DN) as the primary integration target. PVHC needs to be investigated, especially in cases where the grids are unbalanced, and their operational conditions vary spatially and temporally. This motivation prompted us to propose a scalable model tailored to this application. In this paper, we applied linearization to the alternating current optimal power flow (AC-OPF) and solar inverters, transforming the original problem into a mixed-integer linear programming (MILP) problem. Additionally, we accounted for the battery energy storage system (BESS) as a time-coupling factor for calculating PVHC. We then compared the PVHC results between the IEEE-13 bus and SMART-DS San Francisco (SFO) cases and discussed the extent to which BESS can enhance the PVHC of a DN. Furthermore, we designed a web-based graphical visualization for the SFO case, enabling user interaction with raw data and simulation results on a map through a graphical user interface (GUI). In summary, our results and findings provide valuable insights for future three-phase unbalanced AC-OPF PVHC practices and their visualization.

AC-optimal power flow↗

DAMSA Experiment Conceptual Design White Paper

DAMSA (DArk Messenger Searches at an Accelerator) is a novel short-baseline accelerator experiment aimed at probing short-lived physics processes, including searches for evidence of a dark sector of particle physics and well-motivated Standard Model signals. Motivated by open questions in neutrino physics and the absence of conclusive evidence for conventional weakly interacting massive particles, DAMSA targets MeV-to-sub-GeV dark-sector messengers with feeble couplings that can be produced in abundance at the PIP-II LINAC. By employing an ultra-short baseline of order one meter, DAMSA is uniquely positioned to overcome the beam-dump "ceiling" that limits sensitivity to promptly decaying particles in longer-baseline experiments. The conceptual design emphasizes a beam-dump production scheme combined with a compact detector optimized for rare decays while mitigating intense neutron-induced backgrounds inherent to high-power proton beams. To validate the experimental strategy and detector technologies, the Little DAMSA Path-Finder (LDPF) proof-of-concept experiment is proposed, focusing on axion-like particles decaying to two photons and operating with 300 MeV electron beams at FAST. Successful realization of LDPF will establish the feasibility of the DAMSA approach, enabling a broad and powerful program to explore short-lived new physics and precision Standard Model processes in a previously inaccessible regime. This conceptual design document outlines the technical details of DAMSA's physics goals, the beam facility proposals, key experimental challenges and how to overcome them, and the proposed experimental staging campaigns.

Bhattarai, Prithak [Texas U., Arlington]↗

Exploring thermal equilibria of the Fermi-Hubbard model with variational quantum algorithms

Here, this study investigates the thermal properties of the repulsive Fermi-Hubbard model with chemical potential using variational quantum algorithms, crucial in comprehending particle behaviour within lattices at heightened temperatures in condensed matter systems. Conventional computational methods encounter challenges, especially in managing chemical potential, prompting exploration into Hamiltonian approaches. Despite the promise of quantum algorithms, their efficacy is hampered by coherence limitations when simulating extended imaginary time evolution sequences. To overcome these constraints, this research focuses on optimizing variational quantum algorithms to probe the thermal properties of the Fermi-Hubbard model. Physics-inspired circuit designs are tailored to alleviate coherence constraints, facilitating a more comprehensive exploration of materials at elevated temperatures. Our study demonstrates the potential of variational algorithms in simulating the thermal properties of the Fermi-Hubbard model while acknowledging limitations stemming from error sources in quantum devices and encountering barren plateaus.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

Distinguishing prompt-collapse binary neutron star mergers from binary black Holes: Tidal effects and remnant properties

We study the properties of remnants formed in prompt-collapse binary neutron star mergers. We consider nonspinning neutron star binaries over a range of total masses and mass ratios across a set of 22 equations of state, totaling 107 numerical relativity simulations. We report the final mass and spin of the systems (including the accretion disk and ejecta) to be constrained in a narrow range—0.98 ≲ 𝑀 𝑓 /𝑀 ≲ 0.99 for the mass and 0.85 ≲ 𝑎 𝑓 ≲ 0.95 for the dimensionless spin—regardless of the binary configuration and matter effects. This sets them apart from binary black hole merger remnants. We assess the detectability of the postmerger signal in a future 40 km Cosmic Explorer observatory and find that the signal-to-noise ratio in the postmerger of an optimally located and oriented binary at a distance of 100 Mpc can range from <1 to 8, depending on the binary configuration and equation of state, with a majority of them greater than 4 in the set of simulations that we consider. We also consider the distinguishability between prompt-collapse binary neutron star and binary black hole mergers with the same masses and spins. We find that Cosmic Explorer will be able to distinguish such systems primarily via the measurement of tidal effects in the late inspiral. Neutron star binaries with reduced tidal deformability $\tilde{Λ}$ as small as ∼ 3.5 can be identified up to a distance of 100 Mpc, while neutron star binaries with $\tilde{Λ}$ ∼ 22 can be identified to distances greater than 250 Mpc. This is larger than the distance up to which the postmerger will be visible. Finally, we discuss the possible implications of our findings for the equation of state of neutron stars from the gravitational wave event GW230529.

79 ASTRONOMY AND ASTROPHYSICS↗

Fast ion studies in the extended high-performance high β P plasma on EAST

Comprehending and optimizing fast ion behaviors is critical for the enhancement of performance in Experimental Advanced Superconducting Tokamak (EAST). This study explores the potential benefits of several factors that can improve the fast ion confinement. First, experiments show the change in the direction of the NBI2 from counter-I p to co-I p leads to a significant reduction in fast ion losses. TRANSP/NUBEAM simulation and tomography results based on fast-ion D-alpha measurements reveal that after the neutral beam injection (NBI) upgrade, the beam ion prompt loss is reduced by approximately 50%. Second, the upgraded ion cyclotron resonant frequency (ICRF) antenna at the N-port features twice the coupling resistance of the original antennas at EAST. This improved ICRF power coupling has enhanced the synergistic heating effect of NBI + ICRF, where the ICRF wave field accelerates beam ions at the harmonics. Experiments demonstrate that NBI + ICRF synergistic not only enhances plasma neutron yield and β P , but also accelerates beam ions to hundreds of keV. Further, the electron density and the neutral beam voltage have been optimized to reduce the fast ion slowing-down time and beam ion losses. Experimental and simulation results indicate that increasing the electron density reduces beam ion losses and enhances the bootstrap current fraction. While higher beam voltage results in a slight decrease in beam power absorption, it can increase the fraction of bootstrap current. With the understanding of these optimization of fast ion confinement, experiments have demonstrated fully non-inductive operation at high density (n e /n G ∼ 0.67, β P ∼ 3.1, β N ∼ 2.1, H 98,y2 ∼ 1.2) even without the support of co-I p beam NBI2. This investigation presents a potential regime to enhance fast ion confinement and extend performance in the high β P plasma for future experiments.

EAST tokamak↗

Towards Next-Generation Urban Decision Support Systems through AI-Powered Construction of Scientific Ontology Using Large Language Models—A Case in Optimizing Intermodal Freight Transportation

The incorporation of Artificial Intelligence (AI) models into various optimization systems is on the rise. However, addressing complex urban and environmental management challenges often demands deep expertise in domain science and informatics. This expertise is essential for deriving data and simulation-driven insights that support informed decision-making. In this context, we investigate the potential of leveraging the pre-trained Large Language Models (LLMs) to create knowledge representations for supporting operations research. By adopting ChatGPT-4 API as the reasoning core, we outline an applied workflow that encompasses natural language processing, Methontology-based prompt tuning, and Generative Pre-trained Transformer (GPT), to automate the construction of scenario-based ontologies using existing research articles and technical manuals of urban datasets and simulations. From these ontologies, knowledge graphs can be derived using widely adopted formats and protocols, guiding various tasks towards data-informed decision support. The performance of our methodology is evaluated through a comparative analysis that contrasts our AI-generated ontology with the widely recognized pizza ontology, commonly used in tutorials for popular ontology software. We conclude with a real-world case study on optimizing the complex system of multi-modal freight transportation. Our approach advances urban decision support systems by enhancing data and metadata modeling, improving data integration and simulation coupling, and guiding the development of decision support strategies and essential software components.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗

Case Report: Differential diagnosis of hematuria in the emergency department: emphasizing double J stent-inferior vena cava fistula

Introduction Hematuria, a common clinical indicator of genitourinary tract pathology, arises from diverse etiologies including calculi, infections, malignancies, trauma, and iatrogenic causes. Initial evaluation requires hemodynamic assessment, identification of underlying causes, and urinary drainage optimization. This report highlights a rare case of iatrogenic hematuria secondary to double-J stent migration into the inferior vena cava. Case presentation A Chinese male presented with acute left flank pain and gross hematuria persisting for 4 h. Diagnostic imaging revealed a left ureteral stone, prompting double-J stent placement at a local hospital. Despite intervention, hematuria worsened, necessitating abdominal CT. Imaging identified proximal migration of the left double-J stent into the inferior vena cava, with no evidence of vascular injury. Due to concerns regarding inadequate drainage and infection risk, conservative management without catheter clamping was initiated prior to referral. Definitive treatment involved ureteroscopic stent removal under direct visualization at our institution, resulting in rapid symptom resolution. Conclusion This case emphasizes three critical clinical insights: (1) Persistent postoperative hematuria warrants consideration of iatrogenic causes, particularly following urologic device placement. (2) Imaging modalities, especially CT, are indispensable for detecting atypical stent migration. (3) Comprehensive history-taking must include prior urologic interventions to guide differential diagnosis. While double-J stent migration into major vessels remains exceptionally rare, its recognition prevents delayed management of potentially life-threatening complications. Clinicians should maintain heightened vigilance for device-related hematuria in patients with refractory symptoms post-procedurally, ensuring prompt imaging evaluation and multidisciplinary intervention when indicated.

Qi, Wenqi↗

AI in Science Communication

Generative AI has brought innovations across multiple fields, offering great tools for enhanced communication and efficiency. This project focused on developing a custom AI chatbot using OpenAI's Chat GPT (GPT-4o) to support the Fermilab communications team. An analysis identified Chat GPT as the optimal choice, leading to the adoption of its team version and the implementation of a real-time JSON schema for website scanning. Four distinct personas were created to tailor responses to specific audiences, and Fermilab's published content was uploaded to ensure tone consistency. The training involved iterative prompt trials, resulting in a responsive and effective communication assistant. Initial evaluations indicate that the custom GPT shows promise.

Valle, Diego↗

AI in Science Communication

Generative AI has brought great innovations across multiple fields, offering great tools for enhanced communication and efficiency. This project focused on developing a custom AI chatbot using OpenAI's Chat GPT (GPT-4o) to support the Fermilab communications team. An analysis identified Chat GPT as the optimal choice, leading to the adoption of its team version and the implementation of a real-time JSON schema for website scanning. Four distinct personas were created to tailor responses to specific audiences, and Fermilab's published content was uploaded to ensure tone consistency. The training involved iterative prompt trials, resulting in a responsive and effective communication assistant. Initial evaluations indicate that the custom GPT shows promise.

Valle, Diego↗

Beam-dump ceiling and its experimental implications: The case of a portable experiment

We generalize the nature of the so-called beam-dump “ceiling” beyond which the improvement on the sensitivity reach in the search for fast-decaying mediators dramatically slows down, and we point out its experimental implications that motivate tabletop-sized beam-dump experiments for the search. Light (bosonic) mediators are well-motivated new-physics particles, as they can appear in dark-sector portal scenarios and models to explain various laboratory-based anomalies. Due to their low mass and feebly interacting nature, beam-dump-type experiments, utilizing high-intensity particle beams, can play a crucial role in probing the parameter space of such visibly decaying mediators—in particular, the “prompt decay” region, where the mediators feature relatively large coupling and mass. We present a general and semianalytic proof that the ceiling effectively arises in the prompt-decay region of an experiment and show its insensitivity to data statistics, background estimates, and systematic uncertainties, considering a concrete example, the search for axion-like particles interacting with ordinary photons at three benchmark beam facilities: PIP-II at FNAL, and SPS and LHC-dump at CERN. We then identify optimal criteria to perform a cost-effective and short-term experiment to reach the ceiling, demonstrating that very short-baseline compact experiments enable access to the parameter space unreachable thus far.

72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗

Compatibility of molten plutonium with wrought and additively manufactured metal crucibles

Understanding plutonium’s interaction with metals is crucial for optimizing pyrochemical operations, nuclear fuel containment, and various actinide processing techniques. Traditionally, tantalum crucibles are employed for plutonium processing due to their high durability, excellent temperature stability, and low solubility in plutonium. However, tantalum faces challenges such as plutonium wetting and diffusion, making surface coatings particularly important for crucibles in pyrochemical applications to enhance corrosion resistance against plutonium. Tantalum is also expensive and difficult to machine, prompting the need for advanced manufacturing techniques to address these challenges. Here, in this work, we investigate the interaction of Pu with tantalum and titanium crucibles fabricated using both traditional machining methods and laser powder bed fusion (LPBF) additive manufacturing (AM). LPBF-AM is an advanced technique that allows for the creation of complex geometries from traditionally difficult-to-machine metals by using a high-powered laser to build parts. Previous studies of conventional manufactured tantalum have utilized oxidation and carburization of the surface to mitigate plutonium wetting; however, no studies of surface modified LPBF-AM material have been undertaken. These studies are crucial, given the typical differences in the grain structure between conventional and LPBF-AM materials. All crucibles underwent differential scanning calorimetry to confirm the melting of plutonium. Subsequently, the crucibles were sectioned and mounted in epoxy for microstructural analysis using optical microscopy and scanning electron microscopy. This investigation, comparing the performance of wrought vs AM metal crucibles, provides a basis for future tooling applications in actinide processing techniques and can address the challenges associated with traditional machining, particularly in pyrochemical applications.

Actinides↗

Optimization of the light detection system of the ICARUS detector

The Short Baseline Neutrino (SBN) Program at Fermilab is designed to investigate short-baseline neutrino oscillations and test the hypothesis of sterile neutrinos, motivated by several experimental anomalies observed over the past decades. Within this program, the ICARUS experiment plays a key role. It employs the world’s largest Liquid Argon Time Projection Chamber (LArTPC) and serves as the farthest and most sensitive SBN detector for studying muon and electron neutrino oscillations. A crucial subsystem of the ICARUS detector is the Light Detection System (LDS), which captures the prompt scintillation light produced by neutrino interactions in the 600-ton active liquid Argon volume. This system provides precise timing information that is essential for event reconstruction, the trigger system, and cosmic background rejection. The LDS is composed of 360 Hamamatsu R5912-MOD 8-inch photomultiplier tubes (PMTs), operating under cryogenic conditions ($\sim 87 \ K$) inside the detector’s cryostats. During the detector’s operation at FNAL, a degradation in PMT gain has been observed, attributed to aging under low-temperature conditions. In collaboration with ICARUS teams from INFN Pavia and Catania, I developed an experimental setup to study the temperature-dependent behavior of the PMTs, performing gain measurements both at room temperature and down to $-70°C$ using a climatic chamber at INFN Catania. The results indicate that while the PMTs maintain stable gain at room temperature, a significant and permanent gain reduction occurs at low temperatures. Although $-70°C$ is still warmer than liquid Argon temperatures, the findings clearly demonstrate a gain-dependent performance degradation. The thesis also discusses mitigation strategies implemented in the ICARUS detector to address this issue and presents a simplified model to describe and simulate the observed behavior.

Saia, Clara [Catania U.] (ORCID:0009000464102417)↗

Time projection chamber for GADGET II

The established Gaseous Detector with Germanium Tagging (GADGET) detection system is used to measure weak, low-energy 𝛽-delayed proton decays. It consists of the Gaseous Proton Detector equipped with a MICROMEGAS (MM) readout to detect protons and other charged particles calorimetrically, surrounded by the Segmented Germanium Array (SeGA) for high-resolution detection of prompt 𝛾 rays. To upgrade GADGET's Proton Detector to operate as a compact time projection chamber (TPC) for the detection, three-dimensional imaging and identification of low-energy 𝛽-delayed single- and multiparticle emissions mainly of interest to astrophysical studies. A new high granularity MM board with 1024 pads has been designed, fabricated, installed, and tested. A high-density data acquisition system based on generic electronics for TPCs (GET) has been installed and optimized to record and process the gas avalanche signals collected on the readout pads. The TPC's performance has been tested using a 220 Rn 𝛼-particle source and cosmic-ray muons. In addition, decay events in the TPC have been simulated by adapting the attpcroot data analysis framework. Furthermore, a novel application of two-dimensional convolutional neural networks for GADGET II event classification is introduced. The optimization of data throughput is also addressed. The GADGET II TPC is capable of detecting and identifying 𝛼 particles as well as measuring their track direction, range, and energy. The extracted energy resolution of the GADGET II TPC using P10 gas is about 5.4% at 6.288 MeV ( 220 Rn 𝛼 events), computed using charge integration. Based on a systematic simulation study, we estimated the detection efficiency of the GADGET II TPC for protons and 𝛼 particles, respectively. It has also been demonstrated that the GADGET II TPC is capable of tracking minimum-ionizing particles (i.e., cosmic-ray muons). From these measurements, the electron drift velocity was measured under typical operating conditions. In addition to being one of the first generation of micropattern gaseous detectors (MPGDs) to utilize a resistive anode applied to low-energy nuclear physics, the GADGET II TPC will also be the first TPC surrounded by a high-efficiency array of high-purity germanium 𝛾-ray detectors. As a result, the TPC of GADGET II has been designed, fabricated, and tested and is ready for operation at the Facility for Rare Isotope Beams for radioactive-beam-line experiments.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Conceptual design study of neutron detectors for safeguards measurement of an irradiated pebble

Nuclear material control and accounting (MC&A) of pebble-bed reactors (PBRs) is challenging because a PBR utilizes hundreds of thousands of identical, unmarked pebbles that are continuously recirculated through the core. To develop tools that enable the implementation of international safeguards, especially in the context of MC&A of spent pebbles, we designed and simulated three neutron detection concepts to determine fissile content in individual pebbles: a differential die-away (DDA) detector, a californium interrogation prompt neutron (CIPN) detector, and a passive neutron albedo reactivity (PNAR) detector using Monte Carlo calculations. Burnup calculations were performed on the spent pebbles from the PBMR-400 classic PBR. The varying neutron and gamma source terms, and isotopic compositions in the spent pebbles calculated at various burnup levels were used in the neutron detector models. DDA was found to be sensitive to the number of passes a pebble has had through the core and to the fissile content contained in a spent pebble. Optimization in the DDA design further increased the neutron count rates and thus reduced counting uncertainty. Meanwhile, passive neutron counting using the same detector body could distinguish pebbles with different numbers of passes, but its response was dominated by neutron-emitting actinides and was not sensitive to fissile content. On the other hand, the PNAR technique was not viable for a single pebble but performed reasonably for a 27-pebble array, which suggested potential use for verification measurements of containers filled with 27 or more spent pebbles.

CIPN↗

Agentic framework for programmatic crystal structure generation using a fine-tuned worker–supervisor large language model

Platinum group metals (PGMs) underpin many catalytic technologies but face severe supply constraints, motivating the search for alternative materials and computational methods to accelerate discovery. While atomistic simulation tools such as Pymatgen and ASE have streamlined structure manipulation, they require detailed inputs, limiting accessibility for experimentalists and slowing early-stage exploration. Here, in this study, we present an AI-driven agentic framework that orchestrates worker–supervisor large language models (LLMs). The worker translates natural-language prompts of varying abstraction into valid crystallographic structures using a compact LLM fine-tuned with low-rank adaptation on a curated text–code–CIF dataset, emphasizing energy-efficient training. Benchmarking against the baseline CodeGen-350M-mono model shows that fine-tuning reduces hallucination rates from 100% to as low as 5% and improves structural match accuracy to up to 82% for fully specified inputs. Accuracy declines with decreasing prompt detail but remains nontrivial even when only stoichiometry and space group are provided, underscoring the LLM’s capacity for crystallographic inference. The supervisor Claude LLM evaluates the outputs and triggers iterative refinement through the worker’s built-in structure manipulation capabilities (e.g., supercell scaling, strain, vacancy, and substitution operations). We further demonstrate use cases for technologically relevant catalysts, including IrO 2 , pyrochlore Pb 2 Ir 2 O 7 , Ni 2 FeO 4 , and Ni 3 Mo, where the framework generates physically consistent structures that can be refined via geometry optimization. This work introduces a low-energy, language-driven pathway for integrating human and machine intelligence in materials design, paving the way for AI-assisted synthesis planning and high-throughput screening of complex oxides.

AI agent↗

Comparative Evaluation of Control-Oriented Heavy Duty Vehicle Air Drag Coefficient Models

Heavy-duty vehicles (HDVs) are a significant source of fuel consumption and greenhouse gas emissions, prompting solutions such as HDV platooning to mitigate these negative impacts through air drag reduction. The intervehicle distance in an HDV platoon needs to be carefully selected, such that the platoon-level energy efficiency and safety considerations can be well balanced. Underlying this problem lies in accurately modeling the relationship between HDV air drag coefficient and intervehicle distance. Through comprehensive evaluation and comparison, we analyze five control-oriented HDV air drag coefficient models, including the polynomial model, rational polynomial model, rational model, semi-quadratic model, and ridge model. Leveraging Scipy Curve-Fit toolbox and our previously compiled air drag coefficient datasets, we optimally identify the parameters inside each model. The calibrated models are then thoroughly evaluated via five complementary metrics. The comparison results reveal that the semi-quadratic model has the highest overall performance, while the widely adopted rational model only exhibits suboptimal performance.

Best, Micah↗

Agentic AI vs ML-Based Autotuning: A Comparative Study for Loop Reordering Optimization

High Performance Computing (HPC) applications rely heavily on code optimizations to achieve good performance on modern CPU and GPU architectures. Traditional Machine Learning auto-tuning approaches have demonstrated success in exploring high-dimensional spaces, but they often require expensive compile-run evaluations and lack adaptability for large HPC applications. The recent advances in Large Language Models (LLMs) and Agentic AI systems raise intriguing questions about the potential of these approaches to address specific optimization methodologies. This work aims to answer an essential question for the HPC community: “How Agentic AI Systems Compare to Traditional ML Autotuning Techniques?” To address this question, we present a comparative analysis between a traditional ML-based optimization approach and an Agentic AI system, evaluating their respective capabilities and limitations for loop-level optimization. In addition, we introduced a new Agentic AI system named LoopGen-AI using three different Large Language Models: GPT-4.1, Claude 4.0, and Gemini 2.5. A key finding is that LoopGen-AI achieves competitive per-formance with only a few program runs, the reasoning logs from the agents revealed that their decisions rely heavily on the combination of semantic understanding of the target kernel with dynamic feedback from the environment, highlighting a promising new dimension in performance tuning. In contrast, ML-based autotuners focus on statistical exploration, and require orders of magnitude more runs to reach peak performance. Additionally, our analysis shows that prompt engineering, particularly using Persona + Context Manager patterns, significantly impacts the effectiveness of Agentic AI. Our results indicate that while Agentic AI systems are not yet a complete replacement for ML-based autotuners, it can effectively complement traditional methods.

Rosas, Miguel Romero↗