Search NASA⌕ Search

SEARCH · Search NASA

Results for “Ensemble average”

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 19 records
All sourcesNASA NTRSDOE OSTI
DOE OSTI2024

The lion, the witch, and the wormhole: ensemble averaging the symmetric product orbifold

Abstract We consider the ensemble average of two dimensional symmetric product orbifold CFTs Sym N (𝕋 D ) over the Narain moduli space. We argue for a bulk dual given byNcopies of an abelian Chern-Simons theory coupled to topological gravity, endowed with a discrete gauge symmetry exchanging theNcopies. As a check of this proposal, we calculate the ensemble average of various partition and correlation functions of the symmetric product orbifold theory and compare the resulting expressions to gauge theory quantities in the bulk. We comment on the ensemble average of the tensionless string partition function on AdS 3 × S 3 × 𝕋 4 by considering the specific case ofD= 4 with the addition of supersymmetry.

Physics↗
DOE OSTI2024

A method for examining ensemble averaging forms during the transition to turbulence in HED systems for application to RANS models

This paper discusses a strategy to initialize a two-dimensional (2D) Reynolds-averaged Navier–Stokes model [LANL's Besnard–Harlow–Rauenzahn (BHR) model] in order to describe an unsteady transitional Richtmyer–Meshkov (RM)-induced flow observed in on-going high-energy-density ensemble experiments performed on the OMEGA-EP facility. The experiments consist of a nominal single-mode perturbation (initial amplitude a 0 ≈ 10 and wavelength $λ$ = 100μm) with target-to-target variations in the surface roughness subjected to the RM instability with delayed Rayleigh–Taylor in a heavy-to-light configuration. Our strategy leverages high-resolution three-dimensional (3D) implicit large eddy simulations (ILES) simulations to initialize BHR-relevant parameters and subsequently validate the 2D BHR results against the 3D ILES simulations. A suite of five 3D ILES simulations corresponding to five experimental target profiles is undertaken to generate an ensemble dataset. Using ensemble averages from the 3D simulations to initialize the turbulent kinetic energy in the BHR model ( K 0 ) demonstrates the ability of the model to predict the time evolution of the interface as well as the density-specific-volume covariance, b . To quantify the sensitivity of the BHR results to the choice of K 0 and the initial turbulent length scale, S 0 , we execute a parameter sweep spanning four orders of magnitude for both S 0 and K 0 , generating a parameter space consisting of 26 simulations. The Pearson's correlation coefficient is used as a measure of discrepancy between the 2D BHR and 3D ILES simulations and reveals that the ranges 8≲S 0 ≲20 μm and 10 9 ≲K 0 ≲10 10 cm 2 /s 2 produce predictions that agree best with the 3D ILES results.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2025

Nature of molybdenum carbide surfaces for catalytic hydrogen dissociation using machine-learned potentials: an ensemble-averaged perspective

Molybdenum carbides with an electronic structure similar to noble metals have gained attention as a promising low-cost catalyst for biomass valorization and the hydrogen evolution reaction. However, our fundamental understanding of the catalyst surface and how different phases of these catalysts behave at varying reaction conditions is limited to ground state density functional theory calculations as ab initio molecular dynamics (AIMD) is computationally prohibitive at relevant length and time scales. Here, in this work, we train a multi-atomic cluster expansion (MACE) machine-learned interatomic potentials (MLIP) to study hydrogen dissociation and dynamics over Mo, δ-MoC, α-Mo 2 C, and β-Mo 2 C surfaces at varying temperatures and hydrogen partial pressures. Our simulations identify unique and different molecular and atomic hydrogen adsorption sites on different surfaces that do not depend on the temperature. At low hydrogen pressures, the surface coverage is monolayer, which transitions to two-layer adsorption at higher pressures. We find that atomic hydrogen diffusion and recombinations are preferred over molybdenum atom hollow sites, while the diffusion over carbon-terminated facets was negligible, signifying particularly strong C–H interactions. In contrast, molecular hydrogen adsorption occurs mostly atop Mo or the bridging sites. At a comparable hydrogen loading, β-Mo 2 C (001) is the most active surface for hydrogen dissociation reaction. This work provides insights into the dynamic nature of the hydrogen dissociation chemistry and the diversity of hydrogen adsorption sites on molybdenum carbides.

08 HYDROGEN↗
DOE OSTI2023

A combined ensemble-volume average homogenization method for lattice structures with defects under dynamic and static loading

In the study of lattices structures, both experiments and numerical simulations are often conducted with small samples. Using combined ensemble and volume averaging, this work introduces a method to extract a macroscopic constitutive response of a lattice material from numerical simulations performed in periodic domains. The domain size needed to obtain statistically accurate results is investigated. Similar to molecular dynamics, the concept of the virial stress is introduced after homogenized equations are derived using the ensemble averaging method. Under static conditions, the virial stress is shown to agree with the volume averaged solid stress. Using the homogenization method, constitutive relations for this stress can be obtained from systems with uniform strains. Application of such obtained constitutive relations to more general cases results in an error proportional to the square of the ratio between the lattice length scale and the macroscopic length scale. Taking advantage of this property, numerical simulations are performed in systems with a uniform gradient of the average velocity. The volume average method is then used to accelerate convergence when studying lattices with defects. To avoid the artificial numerical time scale from the size of a representative volume element divided by the wave speed, a numerical scheme is developed to enforce a spatially uniform velocity gradient within the computational domain while allowing fluctuations of the velocity or displacement to develop naturally. To account for probability distribution of lattice defects, the stress is calculated as the ensemble-volume averaged value. For dynamic systems, energy dissipation properties are also studied.

36 MATERIALS SCIENCE↗
DOE OSTI2023

An Experimental Comparison of Cyclic Variations in Diesel–Natural Gas and POMDME–Natural Gas Dual Fuel Combustion

Abstract Cyclic variations in internal combustion engines are caused by various factors, including combustion mixture stratification, in-cylinder flows, local fluctuations in air-fuel ratio, etc. Cyclic variations have a profound impact on engine performance and emissions. In this study, cyclic variations in dual fuel combustion are analyzed, comparing diesel–natural gas (NG) and polyoxymethylene dimethyl ether (POMDME)-NG dual fuel combustion. Cyclic variability was initially quantified using the coefficient of variation of gross indicated mean effective pressure (IMEPg) computed from experimental cylinder pressure data. The cases analyzed in this study had a coefficient of variation (COV) of IMEPg greater than or around 5%, which was the lower limit of onset of instability for this engine. Experiments were performed at two fixed start of injection (SOI) of high-cetane fuel: 310 CAD and 350 CAD. For all experiments, a constant load of 5 bar IMEPg was maintained, and the intake boost pressure and rail pressure were fixed at 1.5 bar and 500 bar, respectively. For each case, 1000 cycles of cylinder pressure data were recorded, filtered, and processed using an in-house heat release analysis code for each cycle. A comparison between individual cycles and the “ensemble averaged cycle” was made for both diesel–NG and POMDME–NG combustion. For the early SOI of 310 CAD, the peak cylinder pressure fluctuations of individual cycle were found to be ± 15 bar for both fuel combinations, compared to the ensemble averaged cycle, and < 1/10th of the cycles had an IMEPg lower than 0.05 bar of the ensemble averaged cycle. However, the peak pressure fluctuations were found to be lower for POMDME–NG (±3 bar) than diesel–NG dual fuel combustion at 350 CAD SOI, indicating lower cyclic variations. The higher reactivity of POMDME helped reduce fluctuations in combustion phasing at the retarded SOI. The presence of cycles of deterioration and cycles of recovery were also observed with diesel–NG combustion for 310 CAD SOI, and the scatter in the IMEPg return map was similar for both fuel combinations. The IMEPg return map for POMDME–NG combustion was less scattered at the 350 CAD SOI.

Engineering↗
DOE OSTI2025

In Situ/Operando Probing of Dynamic Phase Structures of Alumina‐Supported Ultrasmall Copper‐Gold Alloy Nanoparticles Under Reaction Conditions

Abstract The ability to control phase structures and surface sites of ultrasmall alloy nanoparticles under reaction conditions is essential for preparing catalysts by design. This is, however, challenging due to limited understanding of the atomic‐scale phases and their correlation with the ensemble‐averaged structures and activities of catalysts during catalytic reactions. We reveal here a dynamic structural stability of alumina‐supported ultrasmall and equiatomic copper‐gold alloy nanoparticles under reaction conditions as a model system in the in situ/operando study. In situ atomic‐scale morphological tracking under oxygen reveals temperature‐dependent dynamic crystalline‐amorphous dual‐phase structures, showing dynamic stability over an elevated temperature range. This atomic‐scale dynamic phase stability coincides with a “conversion plateau” observed for carbon monoxide oxidation on the catalyst. It is substantiated by the stable lattice ordering/disordering structures and surface sites with oscillatory characteristics shown by operando ensemble‐average structural tracking of the catalyst during the oxidation reaction. The understanding of the atomic‐scale dynamic phase structures in correlation with the ensemble‐average dynamic ordering/disordering phase structures and surface sites provides fresh insights into the unique synergy of the supported alloy nanoparticles. This understanding has implications for the design and structural tuning of active and stable ultrasmall alloy catalysts under elevated temperatures.

Chemistry↗
DOE OSTI2024

Atomic Ordering-Induced Ensemble Variation in Alloys Governs Electrocatalyst On/Off States

The catalytic behavior of a material is influenced by ensembles—the geometric configuration of atoms. Traditional approaches, mainly utilizing solid-solution alloys in electrocatalysis, have often overlooked the challenges posed by concurrent changes in the electronic structure (i.e. d-band center) when the composition is altered. Here, this study introduces a methodology that distinctly separates the geometric effects (i.e. ensembles) from the electronic structure. We compare the reactivity of compositionally identical, but structurally different Pd 3 Bi ordered intermetallic and solid-solution alloys. Remarkably, we find that Pd 3 Bi intermetallics display nearly no reactivity for the methanol oxidation (MOR), while their solid-solution counterparts have significant reactivity. This highlights a unique case where materials with identical chemical compositions demonstrate drastically different catalytic behavior underscoring the critical importance of ensembles in electrocatalysis. Specifically, Pd 3 Bi intermetallics form smaller ensembles (average coordination number: 4.5 ± 1.6) with almost no measurable MOR activity at room temperature, in contrast to the solid-solution Pd 3 Bi that exhibit larger ensembles (average coordination number: 6.8 ± 0.9) and considerable MOR reactivity (0.5 mA cm −2 Pd ). An ordered Pd 3 Bi alloy, with an intermediate ensemble size (average coordination number: 5.3 ± 1.2), displays moderate MOR activity (0.1 mA cm −2 Pd ), further confirming the direct correlation between ensemble size and catalytic activity. Notably, all Pd 3 Bi alloys maintain similar electronic structures, because the chemical composition of the alloys is fixed, indicating that the differences in reactivity are predominantly from changes to the ensemble size. Our findings offer an approach for precisely controlling catalytic activity through manipulating the geometric configuration of the atoms within an alloy, paving the way for more efficient catalyst design.

alloys↗
DOE OSTI2025

Operator dynamics in Floquet many-body systems

We study operator dynamics in many-body quantum systems, focusing on generic features of systems that are ergodic, spatially extended, and lack conserved densities. Quantum circuits of various types provide simple models for such systems. We focus on Floquet quantum circuits, comparing their behavior with what has been found previously for circuits that are random in time. Floquet circuits, which have discrete time-translation symmetry, represent an intermediate case between circuits that are random in time and lack any symmetry, and systems with a time-independent Hamiltonian and continuous time-translation invariance. By making this comparison, one of our aims is to identify signatures of time-translation symmetry in Floquet operator dynamics. To characterize behavior we examine a variety of quantities in solvable models and numerically: operator autocorrelation functions; the partial spectral form factor; the out-of-time-order correlator (OTOC); and the paths in operator space that make the dominant contributions to the ensemble-averaged autocorrelation functions. Our most striking result is that ensemble-averaged autocorrelation functions show behavior that is distinctively different in Floquet systems compared to systems in which successive time-steps are independent. Specifically, while average autocorrelation functions decay on a microscopic timescale for circuits that are random in time, in Floquet systems they have a late-time tail with a duration that grows parametrically with the size of the operator support. In the simplest models this tail is separated from the initial decay by a minimum, so that the average autocorrelation function has an intermediate-time peak. The existence of these tails provides a way to understand deviations of the spectral form factor from random matrix behavior at times shorter than the Thouless time. In contrast to this feature in autocorrelation functions, we find no new aspects to the behavior of OTOCs for Floquet models compared to random-in-time circuits. We show that this difference between averaged autocorrelation functions and OTOCs can be understood in terms of the paths in operator space that contribute to the two quantities: paths for the former retain a limited support at late times, while paths for the latter are dominated by operator spreading. Published by the American Physical Society 2025

Yoshimura, Takato (ORCID:0000000309159846)↗
DOE OSTI2024

Resolving local ordering and structure in Mn x Ge 1- x Te alloys through thermodynamic ensembles of pair distribution functions

Characterizing local bonding environments in complex materials is essential for understanding and optimizing their properties. Equally as important is the ability to predict local motifs as a function of synthesis conditions, enhancing chemists’ ability to design properties into materials. In this study, we present an approach to leverage statistical mechanics to generate temperature- and energy-informed ensemble averaged pair distribution functions (PDFs). This method, which we have named Thermodynamic Ensemble Averages of PDFs for Ordering and Transformations (TEAPOT), utilizes density functional theory (DFT) to relax supercells while incorporating energetic penalties for local order, enabling accurate and computationally efficient analysis of local structure. We apply this method to the neutron PDF measurements of the pseudobinary MnTe–GeTe (MGT) alloy, demonstrating its capability to resolve complex local distortions and chemical ordering. Our results reveal detailed insights into phase transformations and local distortions driven by Mn substitution. For compositions that globally present as rock salt, our analysis reveals that Ge coordination geometry is heavily impacted by synthesis temperature. We propose that high temperature synthesis conditions promote a lowered Ge polyhedra distortion, promoting high charge carrier mobility due to the alignment of local and global structure. Incorporating statistical mechanics and computation into experimental analysis thus guides synthesis of tailored local structure.

36 MATERIALS SCIENCE↗
DOE OSTI2025

Comparative studies of cross-phase dynamics in turbulent momentum flux and particle flux at the tokamak edge

Turbulent transport events, including turbulent transport flux of momentum (i.e., turbulent momentum flux or Reynolds stress) and turbulent transport flux of particle (i.e., turbulent particle flux), have important effects on the confinement performance of magnetic confinement fusion devices. Poloidal Reynolds stress is the ensemble average of the product of radial velocity fluctuations and poloidal velocity fluctuations, i.e., $\langle {\widetilde{v}}_{r}{\widetilde{v}}_{\theta }\rangle$. Turbulent particle flux is the ensemble average of the product of radial velocity fluctuations and density fluctuations, i.e., $\langle \widetilde{n}{\widetilde{v}}_{r}\rangle$. Changes in either amplitude of fluctuations or cross phase between fluctuations can cause changes in turbulent transport. In this paper, cross-phase dynamics in the Reynolds stress and turbulent particle flux at the tokamak edge are studied in detail. Reynolds stress and turbulent particle flux are, respectively, written as the product of fluctuation amplitudes and an average cross-phase factor. The mathematical expressions of the average cross-phase factors are derived. The average cross-phase factors and the power spectra of cross phase are obtained using experimental measurement data. It is found that the cross-phase dynamics in Reynolds stress and particle flux are very different. Reynolds stress is found to be more sensitive to cross phase than particle flux is. In the strong $E\times B$ shear layer, spatial slips of cross phase lead to the obvious radial gradient of Reynolds stress. In the no/weak $E\times B$ shear region, the cross phase in Reynolds stress tends to lock. Here, phase locking refers to that the power spectra of phase tend to distribute around a fixed phase which does not change with radial position, while phase slip means that the power spectra of cross phase tend to distribute around a phase that varies with radial position. Phase slip or locking mainly describes the central phase weighted by the power spectra, while the phase scattering mainly describes the dispersion of the power spectrum distribution of the phase. The increased scattering of cross phase, which indicates the power spectra distribution of the phase is more dispersed, contributes to the decreased Reynolds stress for higher collisionality. The cross phase in particle flux tends to lock in both strong and no/weak shear regions. The degree of scattering of cross phase in the particle flux does not change obviously as collisionality increases. For higher collisionality, it is the increased density fluctuation amplitude rather than cross-phase dynamics that leads to the increased particle flux. The underlying physical mechanism that causes Reynolds stress and particle flux to exhibit different phase dynamics is discussed.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗
DOE OSTI2024

Probabilistic measures for biological adaptation and resilience

This paper introduces an approach to quantifying ecological resilience in biological systems, particularly focusing on noisy systems responding to episodic disturbances with sudden adaptations. Incorporating concepts from nonequilibrium statistical mechanics, we propose a measure termed “ecological resilience through adaptation,” specifically tailored to noisy, forced systems that undergo physiological adaptation in the face of stressful environmental changes. Randomness plays a key role, accounting for model uncertainty and the inherent variability in the dynamical response among components of biological systems. Our measure of resilience is rooted in the probabilistic description of states within these systems and is defined in terms of the dynamics of the ensemble average of a model-specific observable quantifying success or well-being. Our approach utilizes stochastic linear response theory to compute how the expected success of a system, originally in statistical equilibrium, dynamically changes in response to a environmental perturbation and a subsequent adaptation. Importantly, the resulting mathematical derivations allow for the estimation of resilience in terms of ensemble averages of simulated or experimental data. Finally, through a simple but clear conceptual example, we illustrate how our resilience measure can be interpreted and compared to other existing frameworks in the literature. The methodology is general but inspired by applications in plant systems, with the potential for broader application to complex biological processes.

60 APPLIED LIFE SCIENCES↗
DOE OSTI2026

Temporal Coarse Graining for Classical Stochastic Noise in Quantum Systems

Simulations of quantum systems with Hamiltonian classical stochastic noise can be challenging when the noise exhibits temporal correlations over a multitude of time scales, such as for 1/f noise in solid-state quantum information processors. Here we present an approach for simulating Hamiltonian classical stochastic noise that performs temporal coarse-graining by effectively integrating out the high-frequency components of the noise. We focus on the case where the stochastic noise can be expressed as a sum of Ornstein-Uhlenbeck processes. Temporal coarse-graining is then achieved by conditioning the stochastic process on a coarse realization of the noise, expressing the conditioned stochastic process in terms of a sum of smooth, deterministic functions and bridge processes with boundaries fixed at zero, and performing the ensemble average over the bridge processes. For Ornstein-Uhlenbeck processes, the deterministic components capture all dependence on the coarse realization, and the stochastic bridge processes are not only independent but taken from the same distribution with correlators that can be expressed analytically, allowing the associated noise propagators to be precomputed once for all simulations. This combination of noise trajectories on a coarse time grid and ensemble averaging over bridge processes has practical advantages, such as a simple concatenation rule, that we highlight with numerical examples.

Albash, Tameem [Sandia National Lab. (SNL-NM), Alb↗
DOE OSTI2023

Representative Phenomena of Cyclic Turbulent Combustion in High-Pressure Fuel Sprays

Abstract Cyclic variations in conventional diesel combustion engines can lead to large differences in engine out emissions even at steady operation. This study uses an optically accessible constant-pressure flow chamber to acquire fuel injections in quick succession to analyze mixing, auto-ignition, and combustion of diesel-surrogate n-heptane using multiple high-speed optical diagnostics. Prior studies have utilized fewer injections and/or they rely on analysis of ensemble average behavior. These approaches do not yield information on injection-to-injection variation or provide confidence in utilizing individual injection measurements for high-fidelity computational fluid dynamics(CFD) model validation. In this study, a large set of 500 injections is used to obtain global parameters including liquid length, vapor penetration length, ignition delay time, and lift-off length. Results for multiple injections are presented to illustrate large injection to injection variations. Potential sources for these variations are analyzed to conclude localized, small scale turbulence and rate of injection variations as the likely sources. Then, a statistical method based on z-scores is proposed and implemented to identify instantaneous injections that best represent the bulk data-set of jet boundaries measured independently by three different diagnostics. This synthesis of statistics-guided screening of data set and ensemble-average analysis offers higher confidence for CFD model validation relying upon both a representative single and average injection results.

42 ENGINEERING↗
DOE OSTI2023

A quantum trajectory picture of single photon absorption and energy transport in photosystem II

Here, we use quantum trajectory theory to study the dynamics of the first step in photosynthesis for a single photon interacting with photosystem II (PSII). By considering individual trajectories we are able to look beyond the ensemble average dynamics to compute the PSII system evolution conditioned upon individual photon counting measurements. Measurements of the transmitted photon beam strongly affects the system state, since detection of an outgoing photon confirms that the PSII must be in the electronic ground state, while a null measurement implies it is in an excited electronic state. We show that under ideal conditions, observing the null result transforms a state with a low excited state population to a state with nearly all population contained in the excited states. We study the PSII dynamics conditioned on such photon counting for both a pure excitonic model of PSII and a more realistic model with exciton-phonon coupling to a dissipative phononic environment. In the absence of such coupling, we show that the measured fluorescence rates show oscillations constituting a photon-counting witness of excitonic coherence. Excitonic coupling to the phonon environment has a strong effect on the observed rates of fluorescence, damping the oscillations. Addition of non-radiative decay and incoherent transitions to radical pair states in the reaction center to the phononic model allows extraction of a quantum efficiency of 92.5% from the long-time evolution, consistent with bulk experimental measurements.

Coherent states↗
DOE OSTI2024

HydraGNN_Predictive_GFM_2024 - Ensemble of predictive graph foundation models for ground state atomistic materials modeling

We provide the ensemble of fifteen pre-trained graph foundation models (GFMs) for atomistic materials modeling applications. Each one of the fifteen GFMs has been trained on five open-source datasets that (once aggregated) amount to over 154 million atomistic structures, which cover over two-thirds of the natural elements of the periodic table and that comprises a broad set of organic and inorganic compounds. This vast set of atomistic structures comprises ground state configurations that are dynamically stable (i.e., equilibrated structures with atomic forces approximately close to zero values) as well as dynamically unstable structures (i.e., non-equilibrium structures with non-negligible non-zero values of atomic forces). The ensemble of datasets aggregated does NOT include excited states. The datasets have been curated to remove atomistic structures with spectral norm of the force tensor above 100 eV/angstrom. Moreover, a linear term of the energy was computed for each dataset using a linear regression model that uses the chemical concentration of each natural element as regressor. The linear term predicted by the linear regression model has been subtracted from each original energy value to perform a re-alignment of the energy values across different electronic structures approximation theories performed to generate the diverse multi-source, multi-fidelity datasets. The folder "ADIOS_files" contains the set of pre-processed datasets in Adaptable I/O System (ADIOS) format (https://www.exascaleproject.org/research-project/adios/) that have been used for the development and training of GFMs in this work. The "ADIOS_files" directory contains 6 sub-directories named as follows: - ANI1x-v3.bp - MPTrj-v3.bp - OC2020-20M-v3.bp - OC2020-v3.bp - OC2022-v3.bp - qm7x-v3.bp Each sub-directory contains the pre-processed datasets converted in Adaptable I/O System (ADIOS) format (https://www.exascaleproject.org/research-project/adios/) that have been used to the development, training, and performance testing of the ensemble go predictive graph foundation models. Each GFM was developed using HydraGNN (https://github.com/ORNL/HydraGNN) as underlying graph neural network (GNN) architecture. The multi-task learning (MTL) capability of HydraGNN was used to simultaneously train the GFMs on labeled values for direct predictions of energy (a total system property of an atomistic structure that measures the chemical stability) and atomic forces (an atomic level property of an atomistic structure that measures the dynamical stability). The hyper parameters of the GFM have been tuned using scalable hyperparameter optimization (HPO) algorithms implemented in the software DeepHyper (https://github.com/deephyper/deephyper). The pre-training of each HPO trial was performed using distributed data parallelism (DDP) to scale the training across 128 compute nodes of the exascale OLCF supercomputer Frontier. Each HPO trial was trained only for 10 epochs and an early stopping was performed to avoid wasting significant computational resources on GNN architectures that were clearly underperforming. For each HPO trial, the 'omnistat' tool developed by (AMD Research - Advanced Micro Device) was used to measure the total energy consumption in kWh. The ensemble of GFMs was obtained by selecting the fifteen best performing HPO trials. Four models have been selected for their clear advantage in accuracy, and these are the GFMs with IDs 229, 156, 147, 260. Additional eleven models have been selected based on judicious balance between accuracy and energy consumption needed for training, and these are the GFMs with IDs 165, 78, 137, 1, 175, 171, 181, 67, 179, 167, 351. Each selected GFM of the ensemble was continued to cumulate a total of at most 30 epochs. In some cases, the total number of epochs actually performed was les than 30 due to two combined factors: (1) the size of the GFM (i.e., the number of model parameters to train) and (2) the total wall-clock time for which the computational resources could be allocated on OLCF-Frontier. The "Ensemble_of_models" directory contains 15 sub-directories named as follows: - gfm_0.229 - gfm_0.156 - gfm_0.147 - gfm_0.260 - gfm_0.165 - gfm_0.78 - gfm_0.137 - gfm_0.1 - gfm_0.175 - gfm_0.171 - gfm_0.181 - gfm_0.67 - gfm_0.179 - gfm_0.167 - gfm_0.351 Each one of these sub-directories refers to one of the fifteen HPO trials that have been selected to continue the pre-training with at most 30 epochs. With each sub-directory associated with a specific HPO trial, the following files can be found: - config.json: file for argument parsing to develop and train an HydraGNN architecture - gfm_0.ID_epoch_N.pk: file with model parameters for HPO ID trial after N epochs of training The ensemble of fifteen GFM architectures was used for (1) ensemble averaging to stabilize the predictions of energy and atomic forces after pre-training for post-processing analysis and (2) ensemble uncertainty quantification (UQ). The code used to develop, pre-train, and load the pre-trained models for post-processing analysis is available on the ORNL-GitHub at the following link: https://github.com/ORNL/HydraGNN/tree/Predictive_GFM_2024

36 MATERIALS SCIENCE↗
DOE OSTI2024

A novel conditional generative model for efficient ensemble forecasts of state variables in large-scale geological carbon storage

Integrating monitoring data to efficiently update reservoir pressure and CO 2 plume distribution forecasts presents a significant challenge in geological carbon storage (GCS) applications. Inverse modeling techniques are commonly used to fuse observational data and refine reservoir model parameters, thereby improving state variable forecasts. However, these techniques often rely on linear or Gaussian assumptions, which can limit their effectiveness in accurately predicting state variables. Moreover, simulating large-scale three-dimensional (3D) GCS problems is computationally expensive, making iterative runs in inverse problems prohibitive. To address these challenges, we propose a conditional generative model utilizing the score-based diffusion method for real-time 3D pressure and saturation field distribution predictions. Our approach involves solving the score function with a mini-batch-based Monte Carlo estimator to generate labeled data. This data is subsequently employed to train a fully connected neural network, enabling it to learn the conditional sample generator within a supervised learning framework. This method enables the rapid generation of a large ensemble of predictions, facilitating comprehensive uncertainty quantification of state variables. Here we applied our method to forecast the dynamic 3D distributions of pressure and saturation fields over a 30-year injection period. The statistical assessment with low root mean square error (RMSE) values demonstrates that our method can accurately predict the spatiotemporal distributions of both pressure and saturation fields. Moreover, the developed conditional generative model shows high computational efficiency by generating 100 ensemble forecasts of 3D state variables in less than 10 min. The consistency between ensemble averages and ground truth values further illustrates the model’s capability to capture state variable dynamics during the CO 2 plume injection process. Notably, the ground truth values fall within the ensemble forecasts, indicating that our uncertainty quantification effectively captures variability and potential noise in the observations. Thus, the developed conditional generative model proves to be a more efficient, accurate, and practical tool for GCS applications, facilitating timely risk analysis and informed decision-making.

58 GEOSCIENCES↗
DOE OSTI2025

High fidelity simulations of contaminant dispersion in an urban environment with comparison to magnetic resonance imaging measurements

The dispersion of a contaminant in an urban environment has the potential to impact a large population of people. In this work, a complex urban canopy flow based on the Oklahoma City downtown business district circa 2003 is studied using Magnetic Resonance Imaging (MRI) and high-fidelity Large Eddy Simulations (LES). MRI is a novel experimental technique that can provide high-resolution measurements in four dimensions (three spatial and temporal) for lab scale models. The experiments and simulations use the same geometry and boundary conditions providing a one-to-one comparison of the two methods. Results are presented on the time-averaged velocity and concentration fields, the temporal dynamics of the concentration plumes for a transient release, and a novel Cloud Identification Algorithm that can separate plumes produced by periodic contaminant releases used for ensemble averaging over many releases. The MRI and LES datasets both include millions of measurement voxels and the comparisons highlight the complex 3D nature of the flow including strong vertical velocities in spanwise street canyons and flow acceleration in streamwise street canyons. The concentration fields are qualitatively similar albeit the LES shows larger dispersion. A quantitative analysis with performance measures compares the datasets pointwise and demonstrates that the two 3D datasets are similar with respect to many measures including a fractional bias of 0.02 (ideal=0.0), correlation coefficient of 0.87 (ideal = 1.0), and the fraction points within a factor of 2 is 0.98 (ideal = 1.0). Plume analysis compares the arrival and residence time of contaminant and is found to vary significantly with location within the urban environment with arrival times between 0 and 1.25 and differences within the contaminant cloud less than 10% at most locations.

54 ENVIRONMENTAL SCIENCES↗
DOE OSTI2023

Infinite neural network quantum states: entanglement and training dynamics

We study infinite limits of neural network quantum states (∞-NNQS), which exhibit representation power through ensemble statistics, and also tractable gradient descent dynamics. Ensemble averages of entanglement entropies are expressed in terms of neural network correlators, and architectures that exhibit volume-law entanglement are presented. The analytic calculations of entanglement entropy bound are tractable because the ensemble statistics are simplified in the Gaussian process limit. A general framework is developed for studying the gradient descent dynamics of neural network quantum states (NNQS), using a quantum state neural tangent kernel (QS-NTK). For ∞-NNQS the training dynamics is simplified, since the QS-NTK becomes deterministic and constant. An analytic solution is derived for quantum state supervised learning, which allows an ∞-NNQS to recover any target wavefunction. Numerical experiments on finite and infinite NNQS in the transverse field Ising model and Fermi Hubbard model demonstrate excellent agreement with theory. ∞-NNQS opens up new opportunities for studying entanglement and training dynamics in other physics applications, such as in finding ground states.

quantum state supervised learning↗