Pareto Prompt Optimization
Explore the source record for details and available documents.
97 MATHEMATICS AND COMPUTING↗
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.
Prompt Phrase Ordering Using Large Language Models in HPC: Evaluating Prompt Sensitivity
Large language models (LLMs) have demonstrated effective performance in domain-specific tasks, often requiring a well-designed prompt to guide their responses. However, optimizing the right prompt is challenging due to prompt sensitivity—the phenomenon where small changes in the prompt can lead to significant variations in performance. In this study, we evaluate prompt performance by examining all permutations of independent phrases to investigate prompt sensitivity and robustness. We used two datasets: the GSM8k dataset, which assesses mathematical reasoning, and a custom template prompt for summarizing database metadata. Our goal was to evaluate the performance across all permutations of a sequence of prompt phrases. The study was conducted using the llama3-instruct- 7B model hosted on Ollama, with computations parallelized in a high-performance computing environment. By comparing the average index of phrases in the best and worst-performing prompts, we found that the order of independent phrases within a prompt significantly impacts LLM performance. Additionally, we used Hamming distance to assess changes between phrase orderings, concluding that prompt modifications can dramatically affect scores, often by almost random chance. These findings support existing research on prompt sensitivity. We discuss the challenges of prompt optimization, noting that altering phrases in a successful prompt does not always result in another successful prompt.
97 MATHEMATICS AND COMPUTING↗
GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning
Large language models (LLMs) are increasingly adapted to downstream tasks via reinforcement learning (RL) methods like Group Relative Policy Optimization (GRPO), which often require thousands of rollouts to learn new tasks. We argue that the interpretable nature of language often provides a much richer learning medium for LLMs, compared to policy gradients derived from sparse, scalar rewards. To test this, we introduce GEPA (Genetic-Pareto), a prompt optimizer that thoroughly incorporates natural language reflection to learn high-level rules from trial and error. Given any AI system containing one or more LLM prompts, GEPA samples trajectories (e.g., reasoning, tool calls, and tool outputs) and reflects on them in natural language to diagnose problems, propose and test prompt updates, and combine complementary lessons from the Pareto frontier of its own attempts. As a result of GEPA's design, it can often turn even just a few rollouts into a large quality gain. Across six tasks, GEPA outperforms GRPO by 6% on average and by up to 20%, while using up to 35x fewer rollouts. GEPA also outperforms the leading prompt optimizer, MIPROv2, by over 10% (e.g., +12% accuracy on AIME-2025), and demonstrates promising results as an inference-time search strategy for code optimization. We release our code at https://github.com/gepa-ai/gepa.
97 MATHEMATICS AND COMPUTING↗
Pushing the Limits of Pulse Shape Discrimination in a Large Liquid Xenon Detector
The LUX-ZEPLIN (LZ) experiment is a direct-detection dark matter experiment, optimized to search for weakly interacting massive particles (WIMPs) through WIMP-nucleon interactions. The main challenge in dark matter detection is differentiating between WIMP signals and background events. In LZ, the ratio of ionization to scintillation signals (charge-to-light) is the primary method for rejecting electronic recoil (ER) background. Pulse shape discrimination (PSD) offers a method for additional ER backgrounds rejection in liquid xenon detectors. In this paper, the discrimination power of PSD with the LZ experiment is discussed. To precisely characterize the scintillation pulse shape, an analysis framework is developed to reconstruct the detection time of individual photons. Using LZ calibration data, the photon-timing prompt fraction discriminator is optimized and achieves ER leakage as low as $15\%$. For specific background processes such as $^{124}$Xe double electron capture, the leakage is reduced further to about $5\%$. PSD is combined with charge-to-light to form two-factor discrimination (TFD). The optimized TFD performance is compared with the performance of the charge-to-light method, with the corresponding false positive rate reduced by up to a factor of two for large scintillation pulses. Finally, PSD and TFD are applied to data from LZ's WS2024 run and their performance is summarized.
Akerib, D. S. [SLAC; KIPAC, Menlo Park]↗
Nanoscopic strain evolution in single-crystal battery positive electrodes
Single-crystal Ni-rich layered oxides (SC-NMC) with a grain-boundary-free configuration have effectively addressed the long-standing cracking issue of conventional polycrystalline Ni-rich materials (PC-NMC) in lithium-ion batteries, prompting a shift in optimization strategies. However, continued reliance on anisotropic lattice volume change—a well-established failure indicator in PC-NMC—as a metric for understanding strain and guiding compositional design for SC-NMC becomes controversial. Here, in this study, by leveraging multiscale diagnostic techniques, we unravelled the distinct nanoscopic strain evolution in SC-NMC during battery operation, challenging the conventional composition-driven strategies and mechanical degradation indicators used for PC-NMC. Through particle-level chemomechanical analysis, we reveal a decoupling between mechanical stability and lattice volume change in SC-NMC, identifying that structural instability in SC materials is primarily driven by multidimensional lattice distortions induced by kinetics-driven reaction heterogeneity and progressively deactivating chemical phases. Using this mechanical failure mode, we redefine the roles of cobalt and manganese in maintaining mechanical stability. Unlike cobalt’s detrimental role in PC-NMC, we find cobalt to be critical in enhancing the longevity of SC-NMC by mitigating localized strain along the extended diffusion pathway, whereas manganese exacerbates mechanical degradation.
36 MATERIALS SCIENCE↗
Machine Learning for Scalable and Optimal Load Shedding Under Power System Contingency
Prompt and effective corrective actions in response to unexpected contingencies are crucial for improving power system resilience and preventing cascading blackouts. The optimal load shedding (OLS) accounting for network limits has the potential to address the diverse system-wide impacts of contingency scenarios as compared to traditional local schemes. However, due to the fast cascading propagation of initial contingencies, real-time OLS solutions are challenging to attain in large systems with high computation and communication needs. In this paper, we propose a decentralized design that leverages offline training of a neural network (NN) model for individual load centers to autonomously construct the OLS solutions from locally available measurements. Our learning-for-OLS approach can greatly reduce the computation and communication needs during online emergency responses, thus preventing the cascading propagation of contingencies for enhanced power grid resilience. Numerical studies on both the IEEE 118-bus system and a synthetic Texas 2000-bus system have demonstrated the efficiency and effectiveness of our scalable OLS learning design for timely power system emergency operations.
24 POWER TRANSMISSION AND DISTRIBUTION↗
Visual Instance-aware Prompt Tuning
Visual Prompt Tuning (VPT) has emerged as a parameter-efficient fine-tuning paradigm for vision transformers, with conventional approaches utilizing dataset-level prompts that remain the same across all input instances. We observe that this strategy results in sub-optimal performance due to high variance in downstream datasets. To address this challenge, we propose Visual Instance-aware Prompt Tuning (ViaPT), which generates instance-aware prompts based on each individual input and fuses them with dataset-level prompts, leveraging Principal Component Analysis (PCA) to retain important prompting information. Moreover, we reveal that VPT-Deep and VPT-Shallow represent two corner cases based on a conceptual understanding, in which they fail to effectively capture instance-specific information, while random dimension reduction on prompts only yields performance between the two extremes. Instead, ViaPT overcomes these limitations by balancing dataset-level and instance-level knowledge, while reducing the amount of learnable parameters compared to VPT-Deep. Extensive experiments across 34 diverse datasets demonstrate that our method consistently outperforms state-of-the-art baselines, establishing a new paradigm for analyzing and optimizing visual prompts for vision transformers.
Xiao, Xi [ORNL] (ORCID:0009000009316982)↗
Bayesian Optimization of Catalysis with In-Context Learning
Large language models (LLMs) can perform accurate classification with zero or few examples through in-context learning (ICL), allowing the model to observe query-relevant examples at inference time and eliminating the need for additional weight updates to generalize beyond its original training data. We extend this capability to regression with uncertainty estimation using frozen LLMs (e.g., GPT-4o, Gemini), enabling Bayesian optimization (BO) in natural language without explicit model training or feature engineering. We apply this to materials discovery by representing materials as synthesis and testing procedures for use in natural language prompts. This Bayesian, design-first approach prioritizes optimization toward target material properties before detailed characterization, in contrast to conventional experimental workflows that often emphasize characterization of suboptimal materials. On benchmarks like aqueous solubility and oxidative coupling of methane (OCM), BO-ICL matches or outperforms Gaussian processes. In live experiments on the reverse water–gas shift (RWGS) reaction, BO-ICL identifies multimetallic catalysts that approach equilibrium CO yield within 6 and 10 iterations from a pool of 3,700 and 360,000 candidates, respectively. Our method redefines materials representation and accelerates discovery, with broad applications across catalysis, materials science, and AI.
Calibration↗
SmileyLlama: modifying large language models for directed chemical space exploration
Here we show that large language models (LLMs) can be transformed via supervised fine-tuning of engineered prompts into SmileyLlama for exploring the chemical space of drug molecules. We benchmark SmileyLlama against pretrained LLMs and chemical language models trained from scratch for generating valid and novel drug-like molecules, and use direct preference optimization to both improve SmileyLlama’s adherence to a prompt and as part of the iMiner reinforcement learning framework to predict molecules with optimized three-dimensional conformations and high binding affinity to drug targets. By training an LLM to speak directly as a chemical language model, while retaining most of its natural language capabilities, we show that SmileyLlama can reliably generate molecules with user-specified properties rather than acting only as a chatbot with knowledge of chemistry or as a virtual assistant. While SmileyLlama is geared toward drug discovery, the supervised fine-tuning/direct preference optimization/LLM framework can be extended to other chemical, biological and materials applications.
37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
Simulation of the radiological impact during selected space travel scenarios using the Monte Carlo code FLUKA
Radiation is one of the major challenges of space exploration and can negatively impact both biological and electronic systems, particularly in the case of long-term journeys or if the spaceship features inadequate shielding. Here, in this work, the cumulative dose levels from prompt radiation in the spacecraft are quantified alongside the residual dose contributions arising from activation of vessel components. The radiological impact was assessed for various space exploration scenarios, considering the same spaceship model featuring three shielding design variants. In each scenario, the radiation environment was generated with the Monte Carlo particle transport and interaction code FLUKA. These results can be used to quantify the contribution of prompt and residual dose in spacefaring ventures and help determine optimal radiation shielding needed to mitigate the overall radiological impact on both astronauts and equipment.
73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗
ARCS: Agentic Retrieval-Augmented Code Synthesis with Iterative Refinement
Agentic Retrieval-Augmented Code Synthesis with Iterative RefinementIn supercomputing, efficient and optimized code generation is essential to leverage high-performance systems effectively. We have developed Agentic Retrieval-Augmented Code Synthesis (ARCS), an advanced framework for accurate, robust, and efficient code generation, completion, and translation. ARCS integrates Retrieval-Augmented Generation (RAG) with Chain-of-Thought (CoT) reasoning to systematically break down and iteratively refine complex programming tasks. An agent-based RAG mechanism retrieves relevant code snippets, while real-time execution feedback drives the synthesis of candidate solutions. This process is formalized as a state-action search tree optimization, balancing code correctness with editing efficiency. Evaluations on the Geeks4Geeks and HumanEval benchmarks demonstrate that ARCS significantly outperforms traditional prompting methods in translation and generation quality. By enabling scalable and precise code synthesis, ARCS offers transformative potential for automating and optimizing code development in supercomputing applications, enhancing computational resource utilization
Bhattarai, Manish [Los Alamos National Labs]↗
Grid-Aware Charging and Operational Optimization for Mixed-Fleet Public Transit
The rapid growth of urban populations and the increasing need for sustainable transportation solutions have prompted a shift towards electric buses in public transit systems. However, the effective management of mixed fleets consisting of both electric and diesel buses poses significant operational challenges. One major challenge is coping with dynamic electricity pricing, where charging costs vary throughout the day. Transit agencies must optimize charging assignments in response to such dynamism while accounting for secondary considerations such as seating constraints. This paper presents a comprehensive mixed-integer linear programming (MILP) model to address these challenges by jointly optimizing charging schedules and trip assignments for mixed (electric and diesel bus) fleets while considering factors such as dynamic electricity pricing, vehicle capacity, and route constraints. We address the potential computational intractability of the MILP formulation, which can arise even with relatively small fleets, by employing a hierarchical approach tailored to the fleet composition. By using real-world data from the city of Chattanooga, Tennessee, USA, we show that our approach can result in significant savings in the operating costs of the mixed transit fleets.
Sen, Rishav↗
A Behavioral Robotics Approach to Radiation Mapping Using Adaptive Sampling
Radiation mapping is a desirable task to automate because of the inherent risks involved and its tedious nature. A novel system was designed to address this by combining various existing technologies, utilizing behavior-based robotics and Bayesian optimization. The system uses a quadruped robot equipped with a manipulator and gamma detector to take measurements at locations that are selected based on the uncertainty of a surrogate model used to estimate the true radiation field. The robot uses input from the world with depth cameras to avoid collisions with the robot’s body, and unreachable points for the end effector are addressed by both allowing for a soft collision with the environment to occur, prompting the system to abandon that point, and varying the exploration tendency of the optimization based on consecutive collisions. This approach provides unique traversability and adaptability over other strategies in the literature. Experiments were performed by placing a Cesium-137 source on the ground and varying geometric setups and an optimization parameter demonstrating the adaptability to diverse environments and the increased robustness resulting from the designed behavior. The results additionally demonstrate that dynamically adjusting the optimization algorithm’s exploration tendency based on the arm’s collision history improves the system’s ability to navigate cluttered environments and construct accurate radiation maps without getting stuck in unreachable areas.
Adams, Joel↗
Measurement of the inclusive WZ production cross section in pp collisions at $\sqrt{\textrm{s}}$ = 13.6 TeV
The inclusive WZ production cross section is measured in proton-proton collisions at a centre-of-mass energy of 13.6 TeV, using data collected during 2022 with the CMS detector, corresponding to an integrated luminosity of 34.7 fb −1 . The measurement uses multileptonic final states and a simultaneous likelihood fit to the number of events in four different lepton flavour categories: eee, ee μ , μμ e, and μμμ . The selection is optimized to minimize the number of background events, and relies on an efficient prompt lepton discrimination strategy. The WZ production cross section is measured in a phase space defined within a 30 GeV window around the Z boson mass, as σ total (pp → WZ) = 55.2 ± 1.2 (stat) ± 1.2 (syst) ± 0.8 (lumi) ± 0.3 (theo) pb. In addition, the cross section is measured in a fiducial phase space closer to the detector-level requirements. All the measurements presented in this paper are in agreement with standard model predictions.
72 PHYSICS OF ELEMENTARY PARTICLES AND FIELDS↗
Optimal Network Reconfiguration and Scheduling With Hardware-in-the-Loop Validation for Improved Microgrid Resilience
With the increased occurrence of various major extreme weather events, power outages and prompt power system restorations have recently drawn more attention to the resilience and recovery of power systems. From the perspective of a more resilient power delivery at the distribution grid, system restoration using network topology reconfiguration together with optimal scheduling of distributed energy resources are adopted in this paper. The proposed optimization model aims at minimizing the total load shedding cost and other operational costs, in which linearized topological constraints borrowed from graph theory and linearized DistFlow models are respectively used to maintain the radial network topology and power flow balance after system contingencies. To demonstrate the applicability of the proposed strategy, a real-world case study of a networked three-microgrid system in Adjuntas, Puerto Rico, is used with the consideration of different independent/interconnected microgrid scenarios, contingencies, and fairness settings. Furthermore, hardware-in-the-loop testing is conducted for the same three-microgrid network, where the closely matched results with the simulated ones have validated the effectiveness of the proposed restoration strategy, which is now ready to move one step forward towards field deployment. Finally, to test the proposed restoration strategy in a larger networked system, the modified IEEE-33 bus test distribution system is considered, and the results show a more resilient power delivery for critical loads under three and four line outages.
24 POWER TRANSMISSION AND DISTRIBUTION↗
CEC Quest: Long Duration Energy Storage Impact Analysis Tool
SAND2025-14389O CEC Quest is a Python tool with a user interface designed to analyze the greenhouse gas impacts of long-duration energy storage projects in California. The tool automates data collection from public sources and uses an Application Programming Interface (API) to enable users to download photovoltaic resource availability, marginal operating emissions rate, and utility rate data. It guides users in inputting parameters for a battery energy storage model and uploading site electrical load data, while also prompting for relevant analysis parameters like timestep and grid limits. CEC Quest performs monthly optimization of one year of data to assess impacts on the site’s electrical bill and the grid’s greenhouse gas emissions. Finally, it conducts a lifecycle analysis to evaluate changes over a defined quantification period, with results aggregated through automated report generation. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology & Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.
Rosewater, David [Sandia National Lab. (SNL-CA), L↗
Evaluating Polymer Properties with Different Additives for Carbon Capture and Other Applications
Anthropogenic climate change is one of this generation’s most pressing concerns, with the potential to completely alter the delicate balance we’ve struck with nature. Already, global temperatures have risen 1.29°C, leading to disrupted weather systems, extinctions, increased risks of wildfires, and sea level rise, to name a few effects. Carbon dioxide emission from the combustion of fossil fuels and other industrial activity is a large driver of this phenomenon, as it absorbs heat before it can be radiated away from Earth, trapping it. Carbon dioxide has reached unprecedented levels in our atmosphere, showing a 50% increase from preindustrial averages to a whopping 430 ppm. Thus, reducing the amount of carbon dioxide via carbon capture technology is an important endeavor that serves to benefit everyone. The Microencapsulated CO 2 Sorbent (MECS) team at Lawrence Livermore National Laboratory (LLNL) has turned to microencapsulation to approach this endeavor. Microcapsules provide an attractive approach to carbon capture, combining large surface areas for more efficient mass transfer, regenerative abilities, reduced solvent loss, and improved handling. Additionally, while existing carbon capture technology relies on industrial plants, capsules could present a modular approach to carbon capture, reducing the need for extensive physical infrastructure. The MECS team’s design consists of a polymer membrane that contains a liquid carbon sequestering sorbent, aqueous sodium carbonate. The carbon capturing reaction occurs in three distinct steps, the first of which is the dissolution of carbon dioxide into the sorbent solution and its conversion into carbonic acid (H 2 CO 3 ), shown in equations 1 and 2 respectively. Because this step hinges upon the ability of carbon dioxide to reach the solution inside the capsule, it is necessary that the microcapsule shell is permeable to carbon dioxide gas. The MECS team produces these microcapsules using the in-air droplet encapsulation apparatus (IDEA) shown in figure 1, which can produce uniform micron-scale droplets at speeds much faster than traditional single-dispersal microfluidic-based techniques. The IDEA Is 100 times faster than these current techniques and can reach up to 1000 times their speed when incorporating a multi-nozzle design. Additionally, because droplets are produced in-air via vibration, IDEA can decrease post-processing times and material waste by 99% and can fabricate microgels that are 10 to 100 times more viscous than can be produced via traditional microfluidics. While this design represents a breakthrough in the throughput, efficiency, and tunability of microcapsule production, it imposes a major constraint on the microcapsule curing process. Because microcapsule shells are crosslinked with UV light while falling 30 cm through the air, this gives them a reaction window of approximately 0.2 seconds. Thus, the system and shell formulations must be optimized such that the shells can be fully crosslinked within this very narrow window, prompting investigations into curing behavior.
36 MATERIALS SCIENCE↗
Large Language Model for Validation, Optical Calibration, and Learning (VOCAL) Distributed Temperature Sensing Interface
Distributed temperature sensing (DTS) using fiber optic sensors (FOS) offers a promising method for temperature measurements in advanced reactors, such as sodium fast reactors and molten salt cooled reactors. To support the calibration and validation of DTS measurements, Argonne National Laboratory developed the Validation, Optical Calibration, and Learning (VOCAL) software package. This report describes the integration of a local large language model (LLM) with a retrieval-augmented generation (RAG) system into the VOCAL interface to serve as an interactive user assistant. The LLM framework enhances the VOCAL platform’s accessibility to users by explaining interface components, clarifying inputs and outputs, and answering user queries dynamically in real-time. The accuracy of the LLM assistant performance was evaluated with 20 queries regarding the interface and its parameters using experimental data from the Thermal Hydraulic Experimental Test Article (THETA) facility. Results demonstrate that the LLM achieved a 95% accuracy rate, with a BERTScore of 0.8816 and SBERT value of 0.7417. Furthermore, validation of the RAG system within the LLM framework showed optimal accuracy with k-values between 1 and 2 using the k-refinement convergence test. The prompt perturbation analysis demonstrated good initial consistency for the RAG system, exhibiting the highest accuracy under punctuation variations and the greatest sensitivity under query reordering. Notably, the model’s errors were limited to data retrieval failures rather than factual hallucinations, reinforcing its baseline reliability. The integration of LLM provides a highly accurate, userfriendly enhancement to the VOCAL platform without disrupting its core computational capabilities for FOS calibration and validation.
Hong, Evan↗