SEARCH · Engineering Papers

Results for “reference models”

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.

SourceResearch connectionFrom yearThrough year

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

At least 721 records · Page 40

All sources NASA NTRS DOE OSTI

DOE OSTI2024

New ASME Section III, Division 5 Creep-Fatigue Design Rules Based on EPP and SMT Approaches

The integrated elastic-perfectly plastic (EPP) and simplified model test (SMT) creep-fatigue (CF) design methodology, referred to as the EPP-SMT method, is being developed as an alternative for CF evaluation in the design of pressure boundary components for high-temperature reactors. This report reviews the conceptual basis of the EPP-SMT methodology, summarizes the SMT experimental development efforts and results, and provides the technical basis for finalizing the EPP-SMT CF design curves for Alloy 617, based on a combined experimental and numerical approach conducted in FY 2024. This report presents the effect of hold time on the CF design curves for Alloy 617 at elevated temperatures. It includes proposed EPP-SMT CF design curves and tabulated values for continuous cycling, along with the effects of maximum hold time, for the use of this EPP-SMT CF evaluation method.

36 MATERIALS SCIENCE↗

DOE OSTI2025

Optimal Control for Fast Frequency Response and Black-Start using Embedded Storages with Grid-Forming Control

The grid-forming inverter is regarded as the solution for integrating high levels of renewable resources into future power systems. Ensuring the stable operation of grids necessitates that grid-forming inverters offer fast frequency response. This report introduces an optimal control method that coordinates the embedded storage within the grid-forming control model with conventional synchronous generators. The optimized active power reference for the embedded storage is generated using receding horizon optimization control, aiming to keep the center of inertia frequency within acceptable limits. The effectiveness of the proposed control is verified through testing in the IEEE 39-bus system. In addition, with grid-forming capability, we will also investigate the application of using mobile embedded storages as black-start units to provide cranking power to energize non-blackstart generators in a black-start process.

24 POWER TRANSMISSION AND DISTRIBUTION↗

DOE OSTI2025

Data and code from: Multivariate bayesian regression model for predicting disposed ash composition at U.S. coal fired power stations

This dataset contains the code and data files needed for implementation of a Multivariate Bayesian Regression model, described in Jin et al. (2025), for the historical prediction of the chemical composition of disposed coal ash at U.S. coal fired power plants as a function of annualized coal purchase data. The integrated coal supply data file (CoalSupplyDataset.csv) represents a compilation of monthly fuel purchase records for the period 1973-2022 at major U.S. power stations. These records were obtained from the U.S. Energy Information Administration. The CSV file also contains, for each coal purchase record, the coal region of the mine as defined by the U.S. Geological Survey. Data entry errors and data gaps in the EIA records were corrected as described in Jin et al. This CSV file represents the integrated coal supply data after corrections were made. The model structure and fitting parameters are encoded in pickle file format (Bayesian.pkl). The model was developed with the coal supply data and coal ash composition data, apportioned according to the Stratified Shuffle Split for training and testing subsets. The model was built using Python and the PyMC library. Reference Publication: Jin, Z.; Huang, J.; Hower, J.C.; Hsu-Kim, H.(2025). Predictive Assessment of the Chemical Composition of Coal Ash in Reserve at U.S. Disposal Sites. Environmental Science & Technology.

Coal ash composition↗

DOE OSTI2022

Modeling Seismic Recordings of High-Frequency Guided Infrasound on Mars

NASA's Interior Exploration using Seismic Investigations, Geodesy and Heat Transport (InSight) mission records several high-frequency (>0.5 Hz) dispersive seismic signals on Mars. These signals are due to the acoustic-to-seismic coupling of infrasound generated by the entry and impact of meteorites. This dispersion property is due to infrasound propagating in a structured atmosphere, and we refer to this dispersive infrasound as guided infrasound. We propose to model the propagation of guided infrasound and the seismic coupling to the ground analytically; we use a 1D layered atmosphere on a three-layer solid subsurface medium. The synthetic ground movements fit the observed dispersive seismic signals well and the fitting indicates that the regolith beneath InSight is about 40-m in thickness. We also examine and validate the previously-published subsurface models derived from InSight ambient seismic vibration data.

58 GEOSCIENCES↗

DOE OSTI2024

Equivalent-Circuit Models for Grid-Forming Inverters under Unbalanced Steady-State Operating Conditions: Preprint

Positive- and negative-sequence equivalent-circuit models are put forth to capture the operation of grid-forming (GFM) inverters in unbalanced steady-state operating conditions acknowledging the impact of current limiting. The particular control architecture examined adopts droop control (for primary control), nested inner-current and outer-voltage control (in the stationary reference frame), and it is adaptable to two different types of current limiting (current-reference saturation and virtual-impedance limiting). We anticipate the proposed models to be of interest in modeling, analysis, and simulation of GFM inverters in unbalanced settings that may arise, e.g., in the face of faults. Validation of the equivalent-circuit models is pursued via comparison with full-order electromagnetic-transient (EMT) simulations for representative balanced and unbalanced faults.

equivalent-circuit models↗

DOE OSTI2022

Production of Fischer-Tropsch Synfuels at Nuclear Plants

A case study analysis was performed to evaluate nuclear-powered synthetic fuel production in the midwestern United States (U.S.). A Fischer-Tropsch (FT) fuel synthesis plant design was used as the basis for the analysis. The FT plant design was configured to produce a product slate consisting of diesel fuel, jet fuel, and motor gasoline blend stocks from carbon dioxide (CO 2 ) and hydrogen (H 2 ) feedstocks. The CO 2 feedstock for the FT plant was assumed to be sourced from biorefineries in the region around a Midwest light water reactor (LWR) nuclear power plant (NPP). The analysis specifies that power from the LWR is used to produce H 2 via high-temperature steam electrolysis and to operate the FT synfuel production plant. Capital costs were estimated for the FT plant while capital costs for the electrolysis plant were based on previous Idaho National Laboratory (INL) studies. In addition to labor and maintenance costs for the FT and electrolysis plants, operating costs also include the costs for CO 2 feedstock transport. An analysis was performed to determine the cost of transporting CO 2 from the distributed biorefinery sources to the centralized fuel synthesis plant as a function of the synfuel plant capacity and corresponding CO 2 demand. The primary revenue streams are associated with sales of the synthetic fuel products. The synthetic fuel products will likely follow the same market trends as the conventional fuel products. The synfuel price data was thus based on projections made by the U.S. Energy Information Administration (EIA) 2021 Annual Energy Outlook (AEO) for conventional fuel products minus federal and state taxes, as well as marketing and distribution costs. The economic analysis also considered cases that included and excluded revenues from the 2022 Inflation Reduction Act (IRA) clean hydrogen production tax credit (PTC) of $\$ $3.00/kg for the first ten years of operation. The economic analysis calculated the net present value (NPV) for cases involving steady-state synfuel production for comparison with the NPV for a business-as-usual case in which NPP continues to sell only electric power to the grid. A synfuel production “Reference Case” was considered in addition to sensitivity cases in which the plant capacity, electricity price, and synthetic fuel product prices were perturbed. The synfuel production Reference Case considered a scenario in which the electrolysis and synfuel plants utilized a combined electrical load of 1000 megawatt electrical (MWe) from the LWR with the balance of the LWR power output being sold to the electric grid. The economic analysis suggests that the synfuel production Reference Case evaluated in this analysis would lead to considerable economic potential for near-term deployment of a nuclear-based synfuel production plant. Specifically, the economic analysis suggests that the deployment of a 1000 megawatt (MW) nuclear-powered synfuel plant could result in a NPV increase of approximately $\$ $1.7 billion for a case with no clean synfuel price premium relative to conventional petroleum fuels when accounting for the additional revenues from the 2022 IRA clean hydrogen PTCs of $\$ $3/kg. Sensitivity analysis was performed to evaluate the effect of perturbation of selected model input parameters on the NPV for the synfuel production Reference Case. The sensitivity analysis indicates that the plant capacity has the largest impact on the differential NPV, with a smaller synfuel production capacity resulting in a decrease in revenue when a larger fraction of the power from the NPP is sold to the grid and a smaller fraction of the power is used to produce synthetic fuel products. The synfuel product pricing has the next largest impact on the differential NPV, with lower synfuel prices resulting in decreased NPV from decreased synfuel sales revenue while higher synfuel prices result in increased NPV from increased synfuel sales revenue. Electricity pricing has a smaller effect on the NPV than the fuel sales price since, in the Reference Case, most of the energy from the NPP is used for synfuel production and a smaller amount of the system revenues are associated with electrical power sales. However, the electricity price sensitivity does indicate that the Synfuel Integrated Energy System (IES) would have a greater NPV than the business-as-usual case (e.g., grid power sales only) when electricity market prices are low, suggesting that synfuel production could provide a strategy for decreasing the economic risks to NPPs posed by a loss of revenues attributed to falling electricity market prices.

10 SYNTHETIC FUELS↗

DOE OSTI2022

Validation of wind resource and energy production simulations for small wind turbines in the United States

Abstract. Due to financial and temporal limitations, the small wind community relies upon simplified wind speed models and energy production simulation tools to assess site suitability and produce energy generation expectations. While efficient and user-friendly, these models and tools are subject to errors that have been insufficiently quantified at small wind turbine heights. This study leverages observations from meteorological towers and sodars across the United States to validate wind speed estimates from the Wind Integration National Dataset (WIND) Toolkit, the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5), and the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), revealing average biases within ±0.5 m s−1 at small wind hub heights. Observations from small wind turbines across the United States provide references for validating energy production estimates from the System Advisor Model (SAM), Wind Report, MyWindTurbine.com, and Global Wind Atlas 3 (GWA3), which are seen to overestimate actual annual capacity factors by 2.5, 4.2, 11.5, and 7.3 percentage points, respectively. In addition to quantifying the error metrics, this paper identifies sources of model and tool discrepancies, noting that interannual fluctuation in the wind resource, wind speed class, and loss assumptions produces more variability in estimates than different horizontal and vertical interpolation techniques. The results of this study provide small wind installers and owners with information about these challenges to consider when making performance estimates and thus possible adjustments accordingly. Looking to the future, recognizing these error metrics and sources of discrepancies provides model and tool researchers and developers with opportunities for product improvement that could positively impact small wind customer confidence and the ability to finance small wind projects.

17 WIND ENERGY↗

DOE OSTI2025

Evidence of medium response to hard probes using correlations of Z bosons with hadrons in heavy ion collisions

The first measurement of pseudorapidity and azimuthal angle distributions relative to the momentum vector of a Z boson for low transverse momentum ($p_\mathrm{T}$) charged hadrons in lead-lead (PbPb) collisions is presented. By studying the hadrons produced in an event with a high-$p_\mathrm{T}$ Z boson (40 $\lt$ $p_\mathrm{T}$ $\lt$ 350 GeV), the analysis probes how the quark-gluon plasma (QGP) medium created in these collisions affects the parton recoiling opposite to the Z boson. Utilizing PbPb data at a nucleon-nucleon center-of-mass energy $\sqrt{s_{_\mathrm{NN}}}$ = 5.02 TeV from 2018 with an integrated luminosity of 1.67 nb$^{-1}$ and proton-proton (pp) data at the same energy from 2017 with 301 pb$^{-1}$, the distributions are examined in bins of charged-hadron $p_\mathrm{T}$. A significant modification of the distributions for charged hadrons in the range 1 $\lt$ $p_\mathrm{T}$ $\lt$ 2 GeV in PbPb collisions is observed when compared to reference measurements from pp collisions. The data provide new information about the correlation between hard and soft particles in heavy ion collisions, which can be used to test predictions of various jet quenching models. The results are consistent with expectations of a hydrodynamic wake created when the QGP is depleted of energy by the parton propagating through it. Based on comparisons of PbPb data with pp references and predictions from theoretical models, this Letter presents the first evidence of medium-recoil and medium-hole effects caused by a hard probe.

CMS↗

DOE OSTI2025

The approximate second order coupled-cluster method based on a size-consistent Brillouin–Wigner partitioning

We present a variant of the approximate second order coupled-cluster method (CC2) with a two-parameter size-consistent Brillouin–Wigner (BW-s) partitioning instead of a Møller–Plesset (MP) partitioning for the unperturbed Hamiltonian, which we refer to as BWs-CC2. The computational complexity of this model scales identically to CC2 with molecular size. Conventional CC2 and its regularized BWs-CC2 variants, as well as conventional MP2 and two of its regularized BW-s2 variants, were assessed on a 535 element database spanning thermochemistry, non-covalent interactions, barrier heights, and isomerization energies. To ensure a well-defined model chemistry, the assessment was performed using internally stable spin-polarized Hartree–Fock (HF) orbitals in the finite aug-cc-pVQZ basis without counterpoise corrections. As a result of using stable orbitals, contrary to conventional wisdom, we find that CC2 substantially outperforms MP2 on molecules with significantly spin contaminated reference orbitals without a significant increase in error on systems with a spin-pure reference, showing the value of its single substitutions. While no single choice of regularization parameters can be optimal for all datasets, we find that BWs-CC2 generally outperforms both CC2 and BW-s2 with a single judicious parameter choice. Additional tests on dipole moments and bond lengths of diatomics provide further support for the utility of this choice. Furthermore, the main outliers and poorest performing cases are associated with large amounts of spin-contamination in the HF reference, which is indicative of systems with either strong correlation or extensive artificial symmetry breaking. Overall, these findings argue that the perception of the quality of the CC2 ground state should be reevaluated and that it can be further improved upon by the soundly based BWs-CC2 variant with the recommended parameter choice.

Correlation energy↗

DOE OSTI2025

Transfer learning-based soybean LAI estimations by integrating PROSAIL, UAV, and PlanetScope imagery

Accurate Leaf Area Index (LAI) estimations at the soybean plot scale is achievable using high-resolution Unmanned Aerial Vehicle (UAV) imagery and field measurement samples. However, the limited coverage of UAV flights restricts large-scale remote sensing monitoring in expansive soybean fields. This study leverages the broad coverage and 3-m resolution of PlanetScope satellite imagery to extend LAI prediction from UAV to satellite scales through transfer learning, using UAV-scale LAI estimates as a benchmark to validate cross-scale consistency. To address this challenge, this study proposed the LAI-TransNet, a two-stage transfer learning framework designed for precise and scalable soybean LAI prediction across large areas, demonstrating its effectiveness in cross-scale monitoring. In Stage 1, a UAV-scale benchmark is established using PROSAIL-simulated UAV reflectance data (UAV-Sim) and field-measured soybean LAI. Traditional machine learning, deep learning, and transfer learning models are trained on a hybrid UAV-Sim and field-measured dataset (UAV-Sim_Measured), with the transfer learning model CNN-TL, fine-tuned using pre-trained weights derived from UAV-Sim, achieving the highest accuracy (R 2 = 0.81, RMSE = 0.64 m 2 /m 2 , rRMSE = 11.5 %). In Stage 2, LAI-TransNet is developed by fine-tuning the CNN-TL model on PlanetScope simulated data (PS-Sim), preprocessed via cross-domain mapping to align UAV and satellite spectral features. Real PlanetScope imagery is corrected for reflectance consistency with reference to UAV imagery spectral profiles. LAI-TransNet outperforms other deep learning models trained directly on PS-Sim (R 2 = 0.69 vs. 0.60–0.63), ensuring robust cross-scale consistency. In conclusion, by bridging UAV and satellite scales, LAI-TransNet enables large-scale soybean LAI monitoring, enhancing precision agriculture management through improved monitoring with the PlanetScope imagery.

Leaf area index (LAI)↗

DOE OSTI2020

Static and dynamic properties of multi-ionic plasma mixtures

Complex plasma mixtures with three or more components are often encountered in astrophysics or in inertial confinement fusion (ICF) experiments. For mixtures containing species with large differences in atomic number Z, the modeling needs to consider at the same time the kinetic theory for low-Z elements combined with the theory of strongly coupled plasma for high-Z elements, as well as all the intermediate situations that can appear in multi-component systems. For such cases, we study the pair distribution functions, self-diffusions, mutual diffusion and viscosity for ternary mixtures at extreme conditions. These quantities can be produced from first principles using orbital free molecular dynamics at the computational expense of very intensive simulations to reach good statistics. Utilizing the first-principles results as reference data, we assess the merit of a global analytic model for transport coefficients, "Pseudo-Ions in Jellium" (PIJ), based on an iso-electronic assumption (iso-n e ). With a multi-component hypernetted-chain integral equation, we verify the quality of the iso-n e prescription for describing the static structure of the mixtures. This semi-analytical modeling compares well with the simulation results and allows one to consider plasma mixtures not accessible to simulations. Further, applications are given for the mix of materials in ICF experiments. A reduction of a multicomponent mixture to an effective binary mixture is also established in the hydrodynamic limit and compared with PIJ estimations for ICF relevant mixtures.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

DOE OSTI2025

Wind Turbine Sound Setbacks and Supply Curves: Ordinances and Extrapolated Trends, 110 Hub Height, 130 Rotor Diameter

This dataset provides a comprehensive set of wind turbine sound setbacks from every residential structure in the contiguous United States (CONUS). A sound setback is defined as the minimum required distance between a residential structure and a hypothetical turbine installation site to ensure that modeled sound levels received at the residence do not exceed local sound ordinances, which are commonly expressed in A-weighted decibels (dBA). Therefore, sound setbacks are a local spatial assessment combining multiple factors, including the sound pressure curve as a function of the observer location (distance and direction) relative to the turbine, local sound regulations, and the geographical distribution of residential structures. The dataset is organized into multiple scenario-based products, detailed as follows: 1. Existing and extrapolated sound setbacks. An existing scenario characterizes sound setbacks only in states or counties that have implemented sound regulations as of 2022. The extrapolated scenarios extend a constant sound threshold to counties that lack explicit sound regulations, with thresholds ranging from 35 to 60 dBA, in 5-dBA increments reflecting the variation observed in current sound ordinances. 2. Sound setbacks in directional and worst scenarios. The directional scenario accounts for the distance and orientation of residential structures relative to a hypothetical turbine location, utilizing the turbine's sound emissions in that specific direction. In contrast, the worst scenario takes loudest sound level at each distance step from the turbine, irrespective of directional considerations, which aligns with current industry practice. 3. Supply curves for Open and Reference Access scenarios. This dataset includes supply curves generated by the reV model, which integrates each of the above sound setbacks into both Open and Reference siting scenarios. In addition, two Open and Reference baselines scenarios were included which do not consider sound setbacks for comparative analysis. All sound setback data are stored in TIF files, with partial maps of the data provided in PNG format. The values in the sound setback raster range from 0 to 1, representing the fraction of developable land within a 90 meter by 90 meter pixel due to sound ordinances. A value of 0 indicates areas where wind energy development is prohibited, while a value of 1 signifies areas fully permissible. The wind turbine parameters used in the sound modeling are based on the land-based turbine from International Energy Agency (IEA), featuring a rated electrical power of 3.4 MW, a rotor diameter of 130 meters, and a hub height of 110 meters. The atmospheric conditions, including wind speed/direction, turbulence, air temperature, relative humidity, and air pressure, that drive the sound generation are obtained from the WIND Toolkit dataset.

Array↗

DOE OSTI2025

ResStock Technical Reference Documentation (V.3.3.0)

ResStock™ is the best-in-class building stock energy model for simulating and publishing energy use, utility bills, and greenhouse gas emissions from the residential sector of the U.S. ResStock answers two primary questions: (1) How is energy used in the U.S. residential building stock? and (2) What is the timeseries aggregate impact of energy technologies? Specifically, ResStock quantifies energy use across geographical locations, demographic groups, building types, fuels, end uses, and time of day. Additionally, it details the impact of efficiency or electrification measures: total changes in the amount of energy used by measure; where or in what use cases efficiency or electrification upgrade measures save energy; when or at what times of day savings occur; and which building stock or demographic segments have the biggest savings potential. This model, and the datasets it produces, are foundational to identifying pathways to affordable and equitable decarbonization of the U.S. residential buildings sector.

32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATI↗

DOE OSTI2021

RELAP5-3D Modeling of High Temperature Test Facility (HTTF)

HTTF at Oregon State University (OSU) Reference: General Atomics? modular high-temperature gas-cooled reactor Helium cooled, electrically heated Prismatic graphite blocks in the core and reflectors Alumina ceramic blocks are used to simulate the core and top and bottom reflectors One-fourth scale in length and diameter Most of the coolant channels in the core are full scale Lower pressure compared to the prototype reactor Over 500 instruments Designed primarily to investigate depressurized conduction cooldown (DCC) transients

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2021

Biomass supply chain equipment for renewable fuels production: A review

The production of renewable fuels is a critical component of global strategies to reduce greenhouse gas (GHG) emissions. Moreover, the collection of raw materials for its production can provide added benefits such as reduction of wildfire risk, additional income for farmers, and decreased disposal costs. Although there is substantial literature on design and modeling of supply chains, the authors were unable to find a single reference with the information needed for the selection and cost estimation of each type of equipment involved in the supply chain. Therefore, the goal of this research is to gather information necessary for the construction and utilization of models that might drive the identification of a feasible supply chain to produce renewable fuels at a commercial scale. The primary objectives are to 1) understand the supply chain of critical feedstocks for renewable fuels production; 2) identify the equipment commercially available for collection and adequation of feedstock; and 3) consolidate information regarding equipment cost, energy consumption, and efficiency, as well as feedstock storage and transportation systems. This paper provides a compilation for five feedstock types studied for sustainable aviation fuel production: 1) agricultural residues and grasses, 2) forest residues, 3) urban wood waste, 4) oilseeds, 5) fats, oils & greases. All the technologies involved from the field to the gate of the preprocessing or conversion unit were reviewed. The information on fats, oils & greases supply chains and equipment purposely designed for forest thinning and pruning was very limited.

: Feedstock, Collection and Adequation, Renewable ↗

DOE OSTI2020

Area-Preserving Mapping of 3D Carotid Ultrasound Images Using Density-Equalizing Reference Map

Carotid atherosclerosis is a focal disease at the bifurcations of the carotid artery. To quantitatively monitor the local changes in the vessel-wall-plus-plaque thickness (VWT) and compare the VWT distributions for different patients or for the same patients at different ultrasound scanning sessions, a mapping technique is required to adjust for the geometric variability of different carotid artery models. In this work, we propose a novel method called density-equalizing reference map (DERM) for mapping 3D carotid surfaces to a standardized 2D carotid template, with an emphasis on preserving the local geometry of the carotid surface by minimizing the local area distortion. The initial map was generated by a previously described arc-length scaling (ALS) mapping method, which projects a 3D carotid surface onto a 2D non-convex L-shaped domain. A smooth and area-preserving flattened map was subsequently constructed by deforming the ALS map using the proposed algorithm that combines the density-equalizing map and the reference map techniques. This combination allows, for the first time, one-to-one mapping from a 3D surface to a standardized non-convex planar domain in an area-preserving manner. Evaluations using 20 carotid surface models show that the proposed method reduced the area distortion of the flattening maps by over 80% as compared to the ALS mapping method.

60 APPLIED LIFE SCIENCES↗

DOE OSTI2025

FEDERATED LEARNING ON STOCHASTIC NEURAL NETWORKS

Federated learning is a machine learning paradigm that leverages edge computing on client devices to optimize models while maintaining user privacy by ensuring that local data remain on the device. However, since all data are collected by clients, federated learning is susceptible to latent noise in local datasets. Factors such as limited measurement capabilities or human errors may introduce inaccuracies in client data. To address this challenge, we propose the use of a stochastic neural network as the local model within the federated learning framework. Stochastic neural networks not only facilitate the estimation of the true underlying states of the data but also enable the quantification of latent noise. We refer to our federated learning approach, which incorporates stochastic neural networks as local models, as federated stochastic neural networks. In this work we will present numerical experiments demonstrating the performance and effectiveness of our method, particularly in handling nonindependent and identically distributed data.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2024

Evaluating the Effectiveness of Retrieval-Augmented Large Language Models in Scientific Document Reasoning

Despite the dramatic progress in Large Language Model (LLM) development, LLMs often provide seemingly plausible but not factual information, often referred as hallucinations. Retrieval-augmented LLMs provide a non-parametric approach to solve these issues by retrieving relevant information from external data sources and augment the training process. These models helps to trace evidence from an externally provided knowledge base allowing the model predictions to be better interpreted and verified. In this work, we critically evaluate these models in their ability to perform in scientific document reasoning tasks. To this end, we tuned multiple such model variants with science-focused instructions and evaluated them on a scientific document reasoning benchmark for the usefulness of the retrieved document passages. Our findings suggest that models justify predictions in science tasks with fabricated evidence and leveraging scientific corpus as pretraining data does not alleviate the risk of evidence fabrication.

• Artificial intelligence (AI) / machine learning ↗