Search NASA⌕ Search

SEARCH · Search NASA

Results for “sampling methods”

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

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

At least 289 records · Page 16
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

Impact of survey spatial variability on galaxy redshift distributions and the cosmological 3 × 2-point statistics for the Rubin Legacy Survey of Space and Time (LSST)

We investigate the impact of spatial survey non-uniformity on the galaxy redshift distributions for forthcoming data releases of the Rubin Observatory Legacy Survey of Space and Time (LSST). Specifically, we construct a mock photometry data set degraded by the Rubin OpSim observing conditions, and estimate photometric redshifts of the sample using a template-fitting photo-z estimator, BPZ, and a machine learning method, FlexZBoost. We select the Gold sample, defined as $i\lt 25.3$ for 10 yr LSST data, with an adjusted magnitude cut for each year and divide it into five tomographic redshift bins for the weak lensing lens and source samples. We quantify the change in the number of objects, mean redshift, and width of each tomographic bin as a function of the coadd i-band depth for 1-yr (Y1), 3-yr (Y3), and 5-yr (Y5) data. In particular, Y3 and Y5 have large non-uniformity due to the rolling cadence of LSST, hence provide a worst-case scenario of the impact from non-uniformity. We find that these quantities typically increase with depth, and the variation can be $10\!-\!40~{{\rm per\ cent}}$ at extreme depth values. Using Y3 as an example, we propagate the variable depth effect to the weak lensing $3\times 2$ pt analysis, and assess the impact on cosmological parameters via a Fisher forecast. We find that galaxy clustering is most susceptible to variable depth, and non-uniformity needs to be mitigated below 3 per cent to recover unbiased cosmological constraints. There is little impact on galaxy–shear and shear–shear power spectra, given the expected LSST Y3 noise.

cosmology↗
DOE OSTI2025

Comparison of microprecipitation methods for polonium source preparation for alpha spectrometry

Detection of radioactive isotopes of polonium is important for understanding natural processes, management and assessment of radioactive waste, and nuclear forensics applications. Further, the most common methods for preparation of polonium samples for alpha spectrometry are electrodeposition and spontaneous deposition which are time consuming. Here, we compare three approaches utilizing rapid microprecipitation from bismuth phosphate, copper sulfide, or tellurium alongside traditional spontaneous deposition methods. From these experiments, results show that copper sulfide microprecipitation recoveries are similar to spontaneous deposition on silver and less time consuming with an approximate five-fold decrease in preparation time, including in the presence of complex matrices like seawater.

38 RADIATION CHEMISTRY, RADIOCHEMISTRY, AND NUCLEA↗
DOE OSTI2026

Sample Preparation Laboratory (SPL) Best Practices: CGD as the Primary Acceptance Method

The traditional paradigm applied to the construction of nuclear safety related systems, structures, and components (SSCs), like SPL, involves an owner/licensee hiring an EPC firm with an NQA 1 program to manage the engineering, procurement, and construction of the desired SSC. This traditional approach would have posed significant barriers to the successful completion of the SPL. As a smaller capital project, in the $\$$100–$\$$200 million range, there was concern that employing one of the larger EPC companies would escalate costs and extend the projected schedule beyond what would be feasible for the Department of Energy (DOE). The traditional acceptance method employed for nuclear facilities with credited safety functions requires the use of an EPC firm applying American Society of Mechanical Engineers (ASME) NQA 1 nuclear quality assurance (NQA) to accept safety SSCs. In lieu of this approach, INL elected to apply its own NQA 1 program to hire an EPC firm operating a commercial quality program. This method used commercial grade dedication (CGD) as outlined in NQA 1 which had not been applied at this scale. Use of the commercial supply chain was identified as a key factor to the successful completion of SPL. Favorable conditions enabling the use of CGD included: • INL serving as both the designer and the design authority • Design created by an NQA 1 qualified source • A mature CGD program • Availability of a CGD subject matter expert (SME) • A strong construction management program. This white paper elaborates on the methodology employed in the construction of the SPL, the successes achieved, and best practices implemented.

42 - ENGINEERING↗
DOE OSTI2025

Denoising Autoencoder for Reconstructing Sensor Observation Data and Predicting Evapotranspiration: Noisy and Missing Values Repair and Uncertainty Quantification

Abstract Machine learning (ML) methods applied in scientific research often deal with interrelated features in high‐dimensional data. Reducing data noise and redundancy is needed to increase prediction accuracy and efficiency especially when dealing with data from field sensors. We explored an unsupervised learning method, the denoising autoencoder (DAE), to extract the underlying data structure from noisy raw data in the context of predicting hydrologic quantities from multiple field sensors. These sensors have intrinsic instrumental noise and occasional malfunctions that cause missing values. Our DAE neural network reconstructed meteorological sensor data containing noise and missing values to predict evapotranspiration in a mountainous watershed. The DAE reconstructed the sensor variables with a mean coefficient of determination value of 0.77 across 15 dimensions representing individual sensors. It reduced variance and bias uncertainties compared to a classical autoencoder model. The reconstruction quality varied across dimensions depending on their cross‐correlation and alignment with the underlying data structure. Uncertainties arising from the model structure were overall higher than those resulting from data corruption. We attached the DAE structure to a downstream ET‐prediction neural network in three formats and achieved reasonably accurate ET predictions . The use of the DAE notably reduced variance uncertainty in ET prediction. However, excessive variance reduction may be accompanied by an increase in bias due to the intrinsic bias‐variance tradeoff. Our method of evaluating and reducing uncertainties in aggregated data from different sources can be used to improve predictive models, process understanding, and uncertainty quantification for better water resource management. Plain Language Summary We present a machine learning method, namely the denoising autoencoder, which reduces the effects of data noise and missing values typically present in scientific data sets collected through sensor measurements. This method selects the most relevant information from noisy raw data collected by the instruments and fills in missing values. To demonstrate the effectiveness of our method, we applied it to predict evapotranspiration, a hydrologic variable that represents the water moved from the land surface to the atmosphere through a combination of evaporation and plant water use (transpiration). We also used a random sampling technique (the Monte Carlo method) to compare the uncertainty in the predictions when using the raw and noisy data versus the reconstructed data. The denoising process produced more accurate predictions of evapotranspiration with less uncertainty. Improved predictions of evapotranspiration can lead to a better understanding and accounting of water budgets. This ML approach is broadly suitable for a wide variety of applications that involve noisy sensor data with missing values. Key Points We used a denoising autoencoder (DAE) neural network to reduce noise in meteorological and soil sensor observations by on average We used Monte Carlo sampling to estimate the bias and variance of all model outputs, including uncertainty sources from data and the model We attached the DAE component to a downstream neural network to predict ET with the variance reduced by , compared to that without the DAE

denoising autoencoder↗
DOE OSTI2024

Hyperspectral Detection of the Fluorescence Shift between Chirality-Sorted Empty and Water-Filled Single-Wall Carbon Nanotube Enantiomers

Single-wall carbon nanotubes (SWCNTs) have extraordinary electronic and optical properties that depend strongly on their exact chiral structure and their interaction with their inner and outer environment. The fluorescence (PL) of semiconducting SWCNTs, for instance, will shift depending on the molecules with which the SWCNT’s hollow core is filled. These interaction-induced shifts are challenging to resolve on the ensemble level in samples containing a mixture of different filling contents due to the relatively large inhomogeneous line width of the ensemble SWCNT PL compared to the size of these shifts. To circumvent this inhomogeneous broadening, single-tube spectroscopy and hyperspectral imaging are often applied, which until now required time-consuming statistical studies. Here, we present hyperspectral PL microscopy combined with automated SWCNT segmenting based on either principal component analysis or a convolutional neural network, capable of both spatially and spectrally resolving the PL along the length of many individual SWCNTs at the same time and automatically fitting peak positions and line widths of individual SWCNTs. The methodology is demonstrated by accurately determining the emission shifts and line widths of thousands of left- and right-handed empty and water-filled SWCNTs coated with a chiral surfactant, resulting in four statistical distributions which cannot be resolved in ensemble spectroscopy of unsorted samples. The results demonstrate a robust method to quickly probe ensemble properties with single-enantiomer spectral resolution. Moreover, it promises to be an absolute quantitative method to characterize the relative abundances of SWCNTs with different handedness or filling content in macroscopic samples, simply by counting individual species.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2025

Third-order photon correlations extract single-nanocrystal multiexciton properties in solution

Colloidal semiconductor nanocrystals are considered promising materials for high-flux optical applications, including lasing, light-emitting diodes, biological imaging, and quantum optics. In high-flux applications, multiexcitons can significantly contribute to emission, influencing its brightness, spectral purity, and kinetics. As a result, understanding and controlling multiexciton emission in colloidal nanocrystal materials is of the utmost importance. In the past, single-nanocrystal photon correlation methods have been applied to understand biexciton and triexciton efficiencies, lifetimes, and spectra. While powerful, such methods suffer from user selection bias and require stable emission from single nanocrystals. To compensate for this shortcoming, second-order correlation methods were developed to extract sample-averaged biexciton properties from a solution of nanocrystals. Until now, however, the analogous third-order solution photon correlation methods remained unexplored. In this work, we present a pair of third-order photon correlation techniques to obtain the sample-averaged single-nanocrystal triexciton quantum yield and lifetime in a solution-phase experiment. These techniques derive from the relationship between the Poisson probability of nanocrystal photon absorption and the intrinsic probability of nanocrystal photon emission. We validate the theoretical background of these techniques by creating a numerical model to simulate the diffusion and emission of many nanocrystals in solution. Our simulations confirm that the average triexciton quantum yield and triexciton lifetime can be extracted from a solution of nanocrystals. These techniques will enable researchers to gain a better understanding of the fundamental multiexciton properties of colloidal nanocrystals.

Horowitz, Jonah R. [Massachusetts Institute of Tec↗
DOE OSTI2025

Creep suppression and fatigue in bio-based composites manufactured via conventional and large format additive manufacturing processes

The emergence of novel extrusion-based additive manufacturing (AM) processes has prompted the development of new thermoplastic composite feedstocks, and broadening sustainability initiatives have driven the development of bio-based and recyclable material for AM feedstocks. Poly(lactic acid) (PLA) with wood flour (WF) is one composite system that has been demonstrated in numerous AM applications, as well as traditional processing methods (i.e., compression and injection molding); however, there has been a need to understand how the variation in processing methodology impacts the material performance of these bio-based feedstocks from a fundamental perspective, with particular emphasis on creep for an extended application use-life. Herein, PLA/WF is explored as a feedstock material for large format additive manufacturing (LFAM) and the performance of additively manufactured materials is compared to those produced via more traditional processing methods. It is also demonstrated that the addition of WF decreases the material’s coefficient of thermal expansion (CTE) while increasing its Young’s modulus, susceptibility to water uptake, and creep fatigue resistance. Essentially, the addition of 20 wt% WF results in a 92 % decrease in rubbery regime CTE while simultaneously resulting in a 14 % increase in modulus, 190 % increase in water uptake, and a 31 % decrease in residual strain after cyclic creep tests. The processing method was also found to play a large role in the final part performance, with the printed material increasing the crystallinity by 183 % and 214 % compared to its compression and injection molded counterparts. Furthermore, the porosity of printed samples increased by two orders of magnitude compared to samples prepared via traditional processing methods.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Untargeted metabolite data from a root surface in a rhizobox

Raw data is provided from samples analyzed using separate reverse phase chromatographic methods on a high performance liquid chromatograph with mass spectrometry. These porewater samples were collected from a microdialysis which generated samples along the surface of a growing A. sative root (all_hc). This data was used to answer questions connecting rhizosphere metabolite (putatively identified metabolites, hc_putative_norm) changes over time (root growth) with changes in the surrounding rhizosphere biogeochemistry (DOC, redox, pH). Rhizosphere biogeochemistry values are provided in the hc_putative_norm file as averages over their respective range of time that they were collected at. The hc_putative_norm file also contains all normalized values over only the intensity values collected for putatively identified metabolites.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2025

An Investigation of Thermal Properties of 2D Materials [Dissertation]

Studying the thermal conductivity of 2D materials is important due to the applications of 2D materials in fields such as thermal management, thermoelectricity, renewable energy, and sensors. As such, measurements of the thermal conductivity of these 2D materials become important to measure. Thermal conductivity is often difficult to measure for 2D materials due to their atomically thin nature and many experimental methods for doing so requiring contact with the sample, which can alter the thermal properties. A non-contact method for calculating the thermal conductivity of 2D materials supported on substrates in order to model the thermal conductivity of 2D materials for devices, is proposed and experimentally performed in this dissertation. The optothermal Raman technique is a useful non-contact diagnostic technique useful in determining the thermal conductivity of 2D materials. The optothermal Raman typically does not account for heat losses due to convection or radiation or substrate resistance, which are shown to be important factors to consider when developing an optothermal Raman model. Additionally, the calculation of the interfacial thermal conductance between the bottom surface of the sample and the top surface of the substrate, plays an important role in determining the final value of the thermal conductivity of a supported sample, and will yield differing results based on whether or not the conductance is calculated using an approach such as the Diffuse Mismatch Model (DMM) or calculated directly by varying the laser heating profile (usually done by changing the laser objective). This is shown to be the case for both graphene on Ni, graphene on Cu, and SnSe 2 on Cu. In addition to experimentally calculating the thermal conductivity of a 2D material with the optothermal Raman technique, the thermal conductivity of 2D materials can also be calculated using computational methods. The three-phonon method is a method which can be used to simulate phonon scattering processes and determine the thermal conductivity of semiconductors, wherein phonon scattering is the dominant mechanism which determines the thermal conductivity. The three-phonon method uses relaxation times for phonon scattering with other phonons, electrons, and other material system elements, such as isotopes or material defects, in order to create a single-mode relaxation time approximation (SMRTA), which is used to calculate the final value of the thermal conductivity. An important consideration when determining the thermal conductivity of a 2D material using this method is the device geometry, which is reflected in this work as the phonon-boundary scattering relaxation time. This inclusion is important along with the inclusion of phonon-electron scattering in accurately determining the thermal conductivity of a 2D material. In both the optothermal Raman experiments and the three-phonon method computations, strain is shown to have a demonstrable effect on the thermal conductivity of 2D materials. When a 1.1% strain was applied to the mechanical properties of SnSe, the three-phonon processes yielded a lower thermal conductivity than the no-strain case. For the optothermal Raman experiments, the strain induced in the Cu substrate and transferred to a single-layer graphene (SLG) sample yields a trend where the thermal conductivity of the SLG decreases with respect to strain applied. In the case where the interfacial thermal conductance was calculated directly, the conductance increased with respect to strain applied. This presents strain as a reliable and viable method for tuning the thermal properties of 2D materials for device applications.

36 MATERIALS SCIENCE↗
DOE OSTI2024

Two-Stage Estimation and Variance Modeling for Latency-Constrained Variational Quantum Algorithms

The quantum approximate optimization algorithm (QAOA) has enjoyed increasing attention in noisy, intermediate-scale quantum computing with its application to combinatorial optimization problems. QAOA has the potential to demonstrate a quantum advantage for NP-hard combinatorial optimization problems. As a hybrid quantum-classical algorithm, the classical component of QAOA resembles a simulation optimization problem in which the simulation outcomes are attainable only through a quantum computer. The simulation that derives from QAOA exhibits two unique features that can have a substantial impact on the optimization process: (i) the variance of the stochastic objective values typically decreases in proportion to the optimality gap, and (ii) querying samples from a quantum computer introduces an additional latency overhead. In this paper, we introduce a novel stochastic trust-region method derived from a derivative-free, adaptive sampling trust-region optimization method intended to efficiently solve the classical optimization problem in QAOA by explicitly taking into account the two mentioned characteristics. The key idea behind the proposed algorithm involves constructing two separate local models in each iteration: a model of the objective function and a model of the variance of the objective function. Exploiting the variance model allows us to restrict the number of communications with the quantum computer and also helps navigate the nonconvex objective landscapes typical in QAOA optimization problems. In conclusion, we numerically demonstrate the superiority of our proposed algorithm using the SimOpt library and Qiskit when we consider a metric of computational burden that explicitly accounts for communication costs.

Derivative-free Optimization↗
DOE OSTI2025

Evaluating the feasibility of LA-ICP-TOF-MS for the analysis of environmental particle collections

Laser ablation-inductively coupled plasma-time-of-flight-mass spectrometry (LA-ICP-TOF-MS) was employed to rapidly analyze environmental particle samples collected using aerosol contaminate extractors (ACE). The ACE particle collectors were placed at various distances (0.5, 1.3, and 4.5 km) from a source that released Ru-bearing particles. Samples for measurement were then generated (as sub-samples) from the ACE collection plates via particle “lift off” with gunshot residue (GSR) tabs. The LA-ICP-TOF-MS method was employed such that 10+ samples could be analyzed in a single unattended analytical session. A 3 × 1 mm area of individual GSR tab samples were analyzed in less than 30 minutes. This provided spatially resolved elemental and isotopic measurements of the particulate content and confirmed the presence of Ru-bearing particles within the complex background environmental particle loading. As anticipated, measurements showed collectors closest to the source had the highest concentration of the released Ru-bearing particles, while all collectors, regardless of distance, contained similar levels of background particles (e.g., Fe and Sr). Sequential scanning electron microscopy – automated particle analysis (SEM-APA) and LA-ICP-TOF-MS analysis was employed for method validation and a demonstration of the multi-modal approach. The same 2-dimensional region was analyzed by both methods and the particles identified via SEM-APA were also detected using LA-ICP-TOF-MS, with 100% accuracy. Overall, LA-ICP-TOF-MS demonstrated its utility for rapid elemental and isotopic particle analysis from environmental air samples.

Manard, Benjamin T. [Oak Ridge National Laboratory↗
DOE OSTI2025

A novel methodology for assessing the hygroscopicity of aerosol filter samples

Abstract. Due to US regulations, concentrations of hygroscopic inorganic sulfate and nitrate have declined in recent years, leading to an increased importance of the hygroscopic nature of organic matter (OM). The hygroscopicity of OM is poorly characterized because only a fraction of the multitude of organic compounds in the atmosphere is readily measured, and there is limited information on their hygroscopic behaviors. Hygroscopicity of aerosol is traditionally measured using a humidified tandem differential mobility analyzer (HTDMA) or electrodynamic balance (EDB). EDB measures water uptake by a single particle. For ambient and chamber studies, HTDMA measurements provide water uptake and particle size information but not chemical composition. To fill this information gap, we developed a novel methodology to assess the water uptake by particles collected on Teflon filters. This method uses the same filter sample for both hygroscopicity measurements and chemical characterization, thereby providing an opportunity to link the measured hygroscopicity with ambient particle composition. To test the method, hygroscopic measurements were conducted in the laboratory for ammonium sulfate, sodium chloride, glucose, and malonic acid, which were collected on 25 mm Teflon filters using an aerosol generator and sampler. Constant-humidity solutions (CHSs), including potassium chloride, barium chloride dihydrate, and potassium sulfate, were employed in a saturated form to maintain the relative humidity (RH) at approximately 84 %, 90 %, and 97 % in small chambers. Our preliminary experiments revealed that, without the pouch, water uptake measurements were not feasible due to rapid water loss during weighing. Additionally, we observed some absorption by the aluminum pouch itself. To account for this, concurrent measurements were conducted for both the loaded and the blank filters at each RH level. Thus, the dry loaded and blank Teflon filters were placed in aluminum pouches with one side open and in RH-controlled chambers for more than 24 h. The wet loaded samples and wet blanks were then weighed using an ultramicrobalance to determine the water uptake by the respective compound and the blank Teflon filter. The net amount of water absorbed by each compound was calculated by subtracting the water uptake of the blank filter from that of the wet loaded filter. Hygroscopic parameters, including the water-to-solute (W / S) ratio, molality, mass fraction solute (mfs), and growth factors (GFs), were calculated from the measurements. The results obtained are consistent with those reported by the Extended Aerosol Inorganics Model (E-AIM) and previous studies utilizing HTDMA and EDB for these compounds, highlighting the accuracy of this new methodology. This new approach enables the hygroscopicity and chemical composition of individual filter samples to be assessed so that in complex mixtures, such as chamber and ambient samples, the total water uptake can be parsed between the inorganic and organic components of the aerosol.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2026

Improvement of Drop‐Hammer Impact Testing for Safety Assessment of High Explosives Using 10‐mg Samples

Here, in this study, we established an improved method for drop-hammer impact testing of small quantities of high explosives (10 mg). We performed about seven hundred impact tests under various experimental conditions (e.g., sandpaper vs bare anvil, different sample masses, drop-weights, and striker diameters) to determine an optimal set of conditions and reaction detection methods (e.g., gas analysis, video, and sound recordings) that give the most statistically reliable results with 10 mg samples. We used both Frequentist and Bayesian statistical approaches to compare estimates of the drop height (DH50) that initiates a reaction 50% of the time, and to quantify the associated uncertainty. Gas analysis proved to be the most reliable reaction detection method, showing unambiguous rises in HE decomposition products (e.g., CO 2 ) even when the other indicators (e.g., sound, video) were inconclusive. The impact tests performed with a bare anvil showed much better reproducibility than those conducted with sandpaper, reducing the largest uncertainty observed in the data sets by a factor of 1.7. The DH 50 values obtained from three different sample masses (10, 20, and 35 mg) fell within the uncertainties of the measurements. We demonstrated the improved procedure (i.e., 10-mg samples, gas analysis, bare anvil, and Bayesian approach) on a variety of PETN samples having different surface areas and thermal histories.

PETN↗
DOE OSTI2026

Full-stack Quantification of Variability in Predicting Ion Transport Properties using Machine-learned Interatomic Potentials

Machine-learned interatomic potentials (MLIPs) have become the state-of-the-art for performing accurate, scalable molecular dynamics (MD) simulations. It is therefore crucial to understand and quantify the reliability of MLIPs for downstream property predictions. Uncertainty in predicted properties can arise from limitations in first-principles training data, intrinsic MLIP model errors in representing the data, and the statistical noise introduced during subsequent MD simulations. Using ion transport in Li7P3S11 as a case study, we systematically assess the impact of training set size and selection, neural network stochasticity, and MD sampling statistics on predicted diffusivity and activation energy. We find that when using equivariant MLIP architectures with standard MD protocols, uncertainty arising from MD sampling dominates over model-induced errors. In contrast, MLIP errors relative to the underlying first-principles data are consistently minor. Given this, there are two main routes to improving the accuracy of predictions based on MLIP potentials: adopting higher accuracy reference data generation methods, and improving the MD sampling statistics.

36 MATERIALS SCIENCE↗
DOE OSTI2026

A comprehensive framework to assess elemental mercury in the Department of Energy: A descriptive analysis

Objective: This study investigated the relationship between personal breathing zone air samples and associated biological monitoring urine samples collected in the U.S. Department of Energy Operations. Methods: We performed standard descriptive analyses of the air sample and BEI monitoring data. We also provide a list of direct-reading instruments used in mercury field assessments. Results: A total of 2,330 air samples and 265 BEI data were analyzed. These data were grouped into 16 job titles, excluding categories with fewer than 10 samples, which resulted in 11 job titles. Conclusions: We conclude that industrial hygiene airborne sample data alone may not provide a complete assessment in exposure determination. We suggested the necessity of incorporating biological monitoring to determine the exposure.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2024

A novel means to generate high pressure

Diamond anvil cells are the most popular means of generating pressures above 2 GPa. However, in many experiments, such as nuclear magnetic resonance and x-ray absorption, the metallic pressurizing gasket (which confines much of the sample) represents an occluding barrier that requires a low Z gasket material (e.g., Be), a split gasket, or other means to enable better coupling of the sample to electromagnetic radiation. In this paper, we demonstrate a novel method for generating high pressures that confines the sample just above the plane of the gasket by using a diamond with a laser hole drilled into the center of the tip. The sample is then confined by the hole, which is sealed by a flat gasket that fits over the hole. When load is applied to the diamonds, metal flows from the deformed gasket into the hole thereby pressurizing the sample similarly to how a piston pressurizes gas inside a cylinder. The pressurized sample is above the metallic gasket plane just inside the tip of the diamond, and thus easily accessible via x rays or visible light that skims just above the plane of the gasket providing an enhanced aperture of radiation collection. Furthermore, we have demonstrated the utility of this method by obtaining Raman spectra of SnC 2 O 4 and x-ray diffraction spectra of seleno-DL-cystine, all at high pressures.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗
DOE OSTI2026

FunDiff: diffusion models over function spaces for physics-informed generative modeling

Recent advances in generative modeling-particularly diffusion models and flow matching-have been widely used for synthesizing discrete data such as images and videos. However, adapting these models to physical applications remains challenging, as the quantities of interest are continuous functions governed by complex physical laws. To address this, we introduce FunDiff, an efficient and robust framework for generative modeling in function spaces. FunDiff combines a latent diffusion process with a function autoencoder architecture to handle input functions with varying discretizations, generates continuous functions that can be evaluated at arbitrary locations, and seamlessly incorporate physical priors. These priors are enforced through architectural constraints or physics-informed loss functions, ensuring that generated samples satisfy fundamental physical laws. We theoretically establish minimax optimality guarantees for density estimation in function spaces, demonstrating that diffusion-based estimators achieve optimal convergence rates under suitable regularity conditions. We further demonstrate the practical effectiveness of FunDiff across diverse applications in fluid dynamics and solid mechanics. Empirical results indicate that our method can generate physically consistent samples with high fidelity to the target distribution, and exhibit robustness to noisy and low-resolution data.

Wang, Sifan [Yale University, New Haven, CT (Unite↗