Search NASA⌕ Search

SEARCH · Search NASA

Results for “modeling”

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 559 records · Page 31

Assessment of a tomography-informed polyhedral discrete element modelling approach for complex-shaped granular woody biomass in stress consolidation

The design of handling equipment for granular biomass primarily requires experiments. Discrete element models (DEM) can provide designers with detailed insight into the behaviour of granular materials. However, granular biomass comprising complex-shaped particles is difficult to model with DEM. A tomography-informed DEM approach and an exhaustive assessment of the ability of the approach to predict the bulk behaviour of milled pines using experimental data is presented. Nano-CT scan was conducted to obtain 3D particle surface geometries as the basis for particle shape approximation by a polyhedral model and a sphero-polyhedral model. These models were applied in the simulation of a compressibility test. Our parametric study showed that particle Young's modulus and restitution coefficient are the two main properties influencing the simulated bulk behaviour of the DEM particles and that the level of approximating DEM particle shape for real particles is critical for accurately replicating the bulk behaviour of milled pines. The polyhedral model demonstrated better suitability than its sphero-polyhedral counterpart for modelling pine particles. The polyhedral model calibrated in a compressibility test was then applied in the simulations of a friction test without additional parameter tuning. Interlocking was dominant in the shear of bulk pine particles. Remarkably, the polyhedral model predicted the frictional behaviour of the pine particles when compared to the experimental results. The limitations of the model, as well as possible ways for enhancement, are discussed. This work provided novel insights into the suitability of complex-shaped DEM models for granular woody biomass.

Biomass flow↗

Role of Troposphere-Convection-Land Coupling in the Southwestern Amazon Precipitation Bias of the Community Earth System Model version 1 (CESM1)

The Amazon forest plays an important role in the tropical circulation and global water and biogeochemical cycles. However, the common Amazon precipitation biases in global climate models undermine our ability to understand future changes in the Amazon and their global consequences. The Community Earth System Model exhibits a persistent wet season dry bias over the southwestern Amazon. The bias is also characterized by inaccurate diurnal timing and intensity of rainfall events. Analysis of a diverse set of simulations suggests that the bias originates from local processes. A known land model bias in the partitioning of surface energy fluxes is found not to contribute significantly to the rainfall bias. Key model processes are identified to be the low-level divergent circulation controlling horizontal moisture flow and the sensitivity of the modeled deep convection to the lower tropospheric moisture. The lower tropospheric moisture divergence dictates the spatiotemporal distributions of convective precipitation through parameterized entrainment mixing, which generally improves the rainfall diurnal cycle. However, it also limits convective precipitation over the areas with frequent and large-scale divergence in the lower troposphere. The southwestern Amazon is such an area, particularly in the late afternoon, although uncertainty exists in the low-level flow in the simulation and reanalyses. One way to sustain rainfall in the presence of low-level divergence is a convective trigger based on both the moisture and temperature changes by large-scale processes, which is shown to help in reducing the dry bias.

Sakaguchi, Koichi↗

A Model Intercomparison Study of Aerosol‐Cloud‐Turbulence Interactions in a Cloud Chamber: 1. Model Results

This study presents the first model intercomparison of aerosol‐cloud‐turbulence interactions in a controlled cloudy Rayleigh‐Bénard Convection chamber environment, utilizing the Pi Chamber at Michigan Technological University. We analyzed simulated cloud chamber‐averaged statistics of microphysics and thermodynamics in a warm‐phase, cloudy environment under steady‐state conditions at varying aerosol injection rates. Simulation results from seven distinct models (DNS, LES, and a 1D turbulence model) were compared. Our findings demonstrate that while all models qualitatively capture observed trends in droplet number concentration, mean radius, and droplet size distributions at both high and low aerosol injection rates, significant quantitative differences were observed. Notably, droplet number concentrations varied by over two orders of magnitude between models for the same injection rates, indicating sensitivities to the model treatments in droplet activation and removal and wall fluxes. Furthermore, inconsistencies in vertical relative humidity profiles and in achieving steady‐state liquid water content suggest the need for further investigation into the mechanisms driving these variations. Despite these discrepancies, the models generally reproduced consistent power‐law relationships between the microphysical variables. This model intercomparison underscores the importance of controlled cloud chamber experiments for validating and improving cloud microphysical parameterizations. Recommendations for future modeling studies are also highlighted, including constraining wall conditions and processes, investigating droplet/aerosol removal (including sidewall losses), and conducting simplified experiments to isolate specific processes contributing to model divergence and reduce model uncertainties.

54 ENVIRONMENTAL SCIENCES↗

Machine Learning Augmented Predictive and Generative Model for Rupture Life in Ferritic and Austenitic Steels

The Larson-Miller parameter (LMP) offers an efficient and fast scheme to estimate the creep rupture life of alloy materials for high temperature applications. However, owing to poor generalizability and dependence on the constant C, which is typically not known a-priori, estimations using the Larson-Miller parameter often result in suboptimal performance for a wide range of materials. At best it is useful in comparing alloys of similar composition. In this work, three machine learning (ML) schemes were developed for rupture life prediction for 9-12% Cr ferritic-martensitic steels and austenitic stainless steels, i.e., a hierarchical model to parameterize LMP using the LMP constant C to compute rupture life, a hierarchical model to parameterize both C and LMP to compute rupture life, and a direct prediction of rupture life. Specifically, we show that the third scheme, using a gradient boosting algorithm, can be used to train ML models for very accurate prediction of rupture life in a variety of alloys (Pear-son Correlation Coefficient > 0.9 for 9-12% Cr and > 0.8 for austenitic stainless steels). In addition, the Shapley value was used to quantify feature importance, making the model interpretable by identifying the effect of various features on the model performance. Furthermore, a variational autoencoder-based generative model was built by conditioning on the experimental dataset to sample hypothetical synthetic candidate alloys from the learnt joint distribution not existing in both 9-12 % Cr ferritic-martensitic steel and austenitic stainless steel datasets. Finally, based on the predictive and generative model, a reinforcement learning strategy has been proposed to guide experimentalists into designing better heat resistant alloys.

Mamun, Md Osman G.↗

Guidelines for the formulation of Lagrangian stochastic models for particle simulations of single-phase and dispersed two-phase turbulent flows

In this paper, we establish a set of criteria which are applied to discuss various formulations under which Lagrangian stochastic models can be found. These models are used for the simulation of fluid particles in single-phase turbulence as well as for the fluid seen by discrete particles in dispersed turbulent two-phase flows. The purpose of the present work is to provide guidelines, useful for experts and non-experts alike, which are shown to be helpful to clarify issues related to the form of Lagrangian stochastic models. A central issue is to put forward reliable requirements which must be met by Lagrangian stochastic models and a new element brought by the present analysis is to address the single- and two-phase flow situations from a unified point of view. For that purpose, we consider first the single-phase flow case and check whether models are fully consistent with the structure of the Reynolds-stress models. In the two-phase flow situation, coming up with clear-cut criteria is more difficult and the present choice is to require that the single-phase situation be well-retrieved in the fluid-limit case, elementary predictive abilities be respected and that some simple statistical features of homogeneous fluid turbulence be correctly reproduced. This analysis does not address the question of the relative predictive capacities of different models but concentrates on their formulation since advantages and disadvantages of different formulations are not always clear. Indeed, hidden in the changes from one structure to another are some possible pitfalls which can lead to flaws in the construction of practical models and to physically unsound numerical calculations. A first interest of the present approach is illustrated by considering some models proposed in the literature and by showing that these criteria help to assess whether these Lagrangian stochastic models can be regarded as acceptable descriptions. A second interest is to indicate how future developments can be safely built, which is also relevant for stochastic subgrid models for particle-laden flows in the context of Large Eddy Simulations.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

IMPROVED MODELING OF THE MASS DISTRIBUTION OF DISK GALAXIES BY THE EINASTO HALO MODEL

Analysis of the rotation curves (RCs) of spiral galaxies provides an efficient diagnostic for studying the properties of dark matter halos and their relations with baryonic material. Since the cored pseudo-isothermal (Iso) model usually provides a better description of observed RCs than does the cuspy Navarro-Frenk-White (NFW) model, there have been concerns that the {Lambda}CDM primordial density fluctuation spectrum may not be the correct one. We have modeled the RCs of galaxies from The H I Nearby Galaxy Survey (THINGS) with the Einasto halo model, which has emerged as the best-fitting model of the halos arising in dissipationless cosmological N-body simulations. We find that the RCs are significantly better fit with the Einasto halo than with either Iso or NFW halo models. In our best-fit Einasto models, the radius of density slope -2 and the density at this radius are highly correlated. The Einasto index, which controls the overall shape of the density profile, is near unity on average for intermediate and low mass halos. This is not in agreement with the predictions from {Lambda}CDM simulations. The indices of the most massive halos are in rough agreement with those cosmological simulations and appear correlated with the halo virial mass. We find that a typical Einasto density profile declines more strongly in its outermost parts than any of the Iso or NFW models whereas it is relatively shallow in its innermost regions. The core nature of those regions of halos thus extends the cusp-core controversy found for the NFW model with low surface density galaxies to the Einasto halo with more massive galaxies like those of THINGS. The Einasto concentrations decrease as a function of halo mass, in agreement with trends seen in numerical simulations. However, they are generally smaller than values expected for simulated Einasto halos. We thus find that, so far, the Einasto halo model provides the best match to the observed RCs and can therefore be considered as a new standard model for dark matter halos.

79 ASTRONOMY AND ASTROPHYSICS↗

Surrogate modelling the Baryonic Universe II: On forward modelling the colours of individual and populations of galaxies

ABSTRACT Among the properties shaping the light of a galaxy, the star formation history (SFH) is one of the most challenging to model due to the variety of correlated physical processes regulating star formation. In this work, we leverage the stellar population synthesis model fsps, together with SFHs predicted by the hydrodynamical simulation IllustrisTNG and the empirical model universemachine, to study the impact of star formation variability on galaxy colours. We start by introducing a model-independent metric to quantify the burstiness of a galaxy formation model, and we use this metric to demonstrate that universemachine predicts SFHs with more burstiness relative to IllustrisTNG. Using this metric and principal component analysis, we construct families of SFH models with adjustable variability, and we show that the precision of broad-band optical and near-infrared colours degrades as the level of unresolved short-term variability increases. We use the same technique to demonstrate that variability in metallicity and dust attenuation presents a practically negligible impact on colours relative to star formation variability. We additionally provide a model-independent fitting function capturing how the level of unresolved star formation variability translates into imprecision in predictions for galaxy colours; our fitting function can be used to determine the minimal SFH model that reproduces colours with some target precision. Finally, we show that modelling the colours of individual galaxies with per cent-level precision demands resorting to complex SFH models, while producing precise colours for galaxy populations can be achieved using models with just a few degrees of freedom.

Chaves-Montero, Jonás (ORCID:0000000295534261)↗

Graph reinforcement learning for exploring model spaces beyond the standard model

We present a methodology for performing scans of beyond the standard model (BSM) parameter spaces with reinforcement learning. We identify a novel procedure using graph neural networks that is capable of exploring spaces of models without the user specifying a fixed particle content, allowing broad classes of BSM models to be explored—in theory, the technique is applicable to nearly any model space with a prespecified gauge group. We provide a generic procedure by which a suitable graph grammar can be developed for any BSM model that features user-specified symmetry groups and a finite number of different possible particle species, the use of which is applicable to a variety of machine learning tasks over the actions of BSM theories beyond our particular reinforcement learning use case. As a proof of concept, we construct the graph grammar for theories with vectorlike leptons that may or may not be charged under a dark U ( 1 ) group, inspired by portal matter extensions of the sub-GeV vector portal/kinetic mixing simplified dark matter models. We then use this graph grammar to create a reinforcement learning environment tasked with creating models with these vectorlike leptons that are consistent with a list of a variety of precision observables. The reinforcement learning agent succeeds in developing models that can address the observed muon anomalous magnetic moment discrepancy while remaining consistent with flavor violation and electroweak precision observables, including both constructions that have previously been studied as well as new models that have not, to our knowledge, previously been identified. By inspecting the resulting ensembles of models that the agent produces and experimenting with different configurations for our reinforcement learning environment and graph grammar, we also infer various lessons about the development of these environments that can be transferable to reinforcement learning scans of more complicated model spaces and comment on future directions for the development of this technique into a more mature tool. Published by the American Physical Society 2025

Wojcik, George N.↗

A revised partiality model and post-refinement algorithm for X-ray free-electron laser data

An updated partiality model and post-refinement algorithm for XFEL snapshot diffraction data is presented and confirmed by observing anomalous density for S atoms at an X-ray wavelength of 1.3 Å. Research towards using X-ray free-electron laser (XFEL) data to solve structures using experimental phasing methods such as sulfur single-wavelength anomalous dispersion (SAD) has been hampered by shortcomings in the diffraction models for X-ray diffraction from FELs. Owing to errors in the orientation matrix and overly simple partiality models, researchers have required large numbers of images to converge to reliable estimates for the structure-factor amplitudes, which may not be feasible for all biological systems. Here, data for cytoplasmic polyhedrosis virus type 17 (CPV17) collected at 1.3 Å wavelength at the Linac Coherent Light Source (LCLS) are revisited. A previously published definition of a partiality model for reflections illuminated by self-amplified spontaneous emission (SASE) pulses is built upon, which defines a fraction between 0 and 1 based on the intersection of a reflection with a spread of Ewald spheres modelled by a super-Gaussian wavelength distribution in the X-ray beam. A method of post-refinement to refine the parameters of this model is suggested. This has generated a merged data set with an overall discrepancy (by calculating the R{sub split} value) of 3.15% to 1.46 Å resolution from a 7225-image data set. The atomic numbers of C, N and O atoms in the structure are distinguishable in the electron-density map. There are 13 S atoms within the 237 residues of CPV17, excluding the initial disordered methionine. These only possess 0.42 anomalous scattering electrons each at 1.3 Å wavelength, but the 12 that have single predominant positions are easily detectable in the anomalous difference Fourier map. It is hoped that these improvements will lead towards XFEL experimental phase determination and structure determination by sulfur SAD and will generally increase the utility of the method for difficult cases.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Greenland Surface Mass Balance as Simulated by the Community Earth System Model. Part I: Model Evaluation and 1850–2005 Results

The modeling of the surface mass balance (SMB) of the Greenland Ice Sheet (GIS) requires high-resolution models in order to capture the observed large gradients in the steep marginal areas. Until now, global climate models have not been considered suitable to model ice sheet SMB owing to model biases and insufficient resolution. This study analyzes the GIS SMB simulated for the period 1850–2005 by the Community Earth System Model (CESM), which includes a new ice sheet component with multiple elevation classes for SMB calculations. The model is evaluated against observational data and output from the regional model Regional Atmospheric Climate Model version 2 (RACMO2). Because of a lack of major climate biases, a sophisticated calculation of snow processes (including surface albedo evolution) and an adequate downscaling technique, CESM is able to realistically simulate GIS surface climate and SMB. CESM SMB agrees reasonably well with in situ data from 475 locations ($r$ = 0.80) and output from RACMO2 ($r$ = 0.79). The simulated mean SMB for 1960–2005 is 359 ± 120 Gt yr -1 in the range of estimates from regional climate models. The simulated seasonal mass variability is comparable with mass observations from the Gravity Recovery and Climate Experiment (GRACE), with synchronous annual maximum (May) and minimum (August–September) and similar amplitudes of the seasonal cycle. CESM is able to simulate the bands of precipitation maxima along the southeast and northwest margins, but absolute precipitation rates are underestimated along the southeastern margin and overestimated in the high interior. The model correctly simulates the major ablation areas. Total refreezing represents 35% of the available liquid water (the sum of rain and melt).

54 ENVIRONMENTAL SCIENCES↗

A Physically Based Runoff Routing Model for Land Surface and Earth System Models

A new physically based runoff routing model, called the Model for Scale Adaptive River Transport (MOSART), has been developed to be applicable across local, regional, and global scales. Within each spatial unit, surface runoff is first routed across hillslopes and then discharged along with subsurface runoff into a ‘‘tributary subnetwork’’ before entering the main channel. The spatial units are thus linked via routing through the main channel network, which is constructed in a scale-consistent way across different spatial resolutions. All model parameters are physically based, and only a small subset requires calibration.MOSART has been applied to the Columbia River basin at 1/ 168, 1/ 88, 1/ 48, and 1/ 28 spatial resolutions and was evaluated using naturalized or observed streamflow at a number of gauge stations. MOSART is compared to two other routing models widely used with land surface models, the River Transport Model (RTM) in the Community Land Model (CLM) and the Lohmann routing model, included as a postprocessor in the Variable Infiltration Capacity (VIC) model package, yielding consistent performance at multiple resolutions. MOSART is further evaluated using the channel velocities derived from field measurements or a hydraulic model at various locations and is shown to be capable of producing the seasonal variation and magnitude of channel velocities reasonably well at different resolutions. Moreover, the impacts of spatial resolution on model simulations are systematically examined at local and regional scales. Finally, the limitations ofMOSART and future directions for improvements are discussed.

runoff routing, scale adaptive↗

Two-Layer Dynamic Recycling Model (2L-DRM): Learning from Moisture Tracking Models of Different Complexity

Atmospheric moisture tracking models are used to identify and quantify sources and sinks of water in the atmospheric branch of the hydrologic cycle. These models are primarily used to investigate the origin of moisture resulting in precipitation for particular regions around the globe. Moisture tracking models vary widely in their level of complexity, depending on the number of physical processes represented. Complex models are comprehensive in their physical representation, but computationally much more expensive than simple models, which only focus on specific physical processes and use simplifying assumptions. We present the mathematical derivation of the new two-layer dynamical recycling model (2L-DRM), a simple analytical moisture tracking model that relaxes the vertically-integrated formulation of the original one-layer DRM (Dominguez et al., 2006). By comparing the simple DRM to a very complex moisture tracking model that uses water vapor tracers embedded within the Weather Research and Forecast regional climate model (WRF-WVT) for the North American Monsoon region, we pinpoint the absence of vertical wind shear as the main deficiency in the simple DRM. When comparing both simple models (DRM and 2L-DRM) to the WRF-WVT (which we treat as ‘truth’), the 2L-DRM better captures the spatial extent, the net amount and the temporal variability of precipitation that originates from oceanic and local terrestrial sources. The 2L-DRM is well suited to study the large-scale climatological sources of moisture, and for these applications, performs on-par with the much more complex and computationally demanding WRF-WVT model.

Dominguez, Francina↗

Grounding our Understanding of the Impacts of Boreal Forest Expansion on Shallow Cumulus Clouds with a Simple Modeling Framework

Abstract The expansion of the boreal forest poleward is a potentially important driver of feedbacks between the land surface and Arctic climate. A growing body of work has highlighted the importance of differences in evaporative resistance between different possible future Arctic land covers, which in turn alters humidity and cloudiness in the boundary layer, for these feedbacks. While thus far this problem has been studied primarily with complex Earth system models, we turn to a locally focused, idealized model capable of diagnosing and testing the sensitivity of first-order processes connecting vegetation, the atmospheric boundary layer, and low clouds in this critical region. This allows us to benchmark the mechanisms and results at the center of predictions from larger-scale simulations. A surface dominated by broadleaf trees, characterized by higher albedo and lower surface evaporative resistance, drives cooling and moistening of the boundary layer relative to a surface of needleleaf trees, characterized by lower albedo and higher surface evaporative resistance. Differences in evaporative resistance between these hypothetical Arctic vegetation covers are of equal importance to changes in albedo for the initial response of the boundary layer to boreal expansion, even with our idealized approach. However, compensation between the elevation of the lifting condensation level (LCL) and more rapid growth of the mixed layer over higher evaporative resistance surfaces can minimize changes in the favorability of shallow clouds over different land cover types under some conditions. We then perform two tests on the sensitivity of this compensating effect, to changes in water availability, represented first by a reduction in boundary layer humidity and then by both a reduction in humidity and soil moisture available to our vegetation surface. Finally, given the importance of this potential LCL–mixed-layer height compensation in our idealized modeling results, we look to determine its relevance in observational data from a field campaign in boreal Finland. These observations do confirm that such a coupling plays an important role in cumulus-topped boundary layers over a needleleaf forest surface. While our results confirm some underlying mechanisms at the center of prior work with Earth system models, they also provide motivation for future work to constrain the impact of boreal forest expansion. This will include both large eddy simulations to examine the impact of processes and feedbacks not resolved by a mixed-layer model, as well as a more systematic evaluation and comparison of relevant observations at the site in Finland and sites from prior boreal field campaigns. Significance Statement Clouds and vegetation are both important components of the climate system that interact across a range of scales. These interactions are central to understanding how changes at the land surface feedback on climate. For example, if a forest expands or recedes, diagnosing how that will impact clouds will determine whether you predict warming or cooling temperatures from that shift in the forest area. These predictions are often made with complex Earth system models, but we look to a more idealized representation of the land–atmosphere system to diagnose how shallow clouds should respond to changes in surface properties with different scenarios of boreal forest expansion at a more foundational level. This both grounds our understanding of previous analysis and provides helpful direction for future studies of this relevant and impactful land cover change.

Meteorology & Atmospheric Sciences↗

Measurements and Model Improvement: Insight into NWP Model Error Using Doppler Lidar and Other WFIP2 Measurement Systems

Abstract Doppler-lidar wind-profile measurements at three sites were used to evaluate NWP model errors from two versions of NOAA’s 3-km-grid HRRR model, to see whether updates in the latest version 4 reduced errors when compared against the original version 1. Nested (750-m grid) versions of each were also tested to see how grid spacing affected forecast skill. The measurements were part of the field phase of the Second Wind Forecasting Improvement Project (WFIP2), an 18-month deployment into central Oregon–Washington, a major wind-energy-producing region. This study focuses on errors in simulating marine intrusions, a summertime, 600–800-m-deep, regional sea-breeze flow found to generate large errors. HRRR errors proved to be complex and site dependent. The most prominent error resulted from a premature drop in modeled marine-intrusion wind speeds after local midnight, when lidar-measured winds of greater than 8 m s −1 persisted through the next morning. These large negative errors were offset at low levels by positive errors due to excessive mixing, complicating the interpretation of model “improvement,” such that the updates to the full-scale versions produced mixed results, sometimes enhancing but sometimes degrading model skill. Nesting consistently improved model performance, with version 1’s nest producing the smallest errors overall. HRRR’s ability to represent the stages of sea-breeze forcing was evaluated using radiation budget, surface-energy balance, and near-surface temperature measurements available during WFIP2. The significant site-to-site differences in model error and the complex nature of these errors mean that field-measurement campaigns having dense arrays of profiling sensors are necessary to properly diagnose and characterize model errors, as part of a systematic approach to NWP model improvement. Significance Statement Dramatic increases in NWP model skill will be required over the coming decades. This paper describes the role of major deployments of accurate profiling sensors in achieving that goal and presents an example from the Second Wind Forecast Improvement Program (WFIP2). Wind-profile data from scanning Doppler lidars were used to evaluate two versions of HRRR, the original and an updated version, and nested versions of each. This study focuses on the ability of updated HRRR versions to improve upon predicting a regional sea-breeze flow, which was found to generate large errors by the original HRRR. Updates to the full-scale HRRR versions produced mixed results, but the finer-mesh versions consistently reduced model errors.

Meteorology & Atmospheric Sciences↗

Physiological fidelity or model parsimony? The relative performance of reverse-toxicokinetic modeling approaches

Physiologically-based toxicokinetic (PBTK) models are often developed to facilitate in vitro to in vivo extrapolation (IVIVE) using a top-down, compartmental approach, favoring architectural simplicity over physiological fidelity despite the lack of general guidelines relating model design to dynamical predictions. Here we explore the impact of design choice (high vs. low fidelity) on chemical distribution throughout an animal’s organ system. Results We contrast transient dynamics and steady states of three previously proposed PBTK models of varying complexity in response to chemical exposure. The steady states for each model were determined analytically to predict exposure conditions from tissue measurements. Steady state whole-body concentrations differ between models, despite identical environmental conditions, which originates from varying levels of physiological fidelity captured by the models. These differences affect the relative predictive accuracy of the inverted models used in exposure reconstruction to link effects-based exposure data with whole-organism response thresholds obtained from in vitro assay measurements. Conclusions Our results demonstrate how disregarding physiological fideltiy in favor of simpler models affects the internal dynamics and steady state estimates for chemical accumulation within tissues, which, in turn, poses significant challenges for the exposure reconstruction efforts that underlie many IVIVE methods. Developing standardized systems-level models for ecological organisms would not only ensure predictive consistency among future modeling studies, but also ensure pragmatic extrapolation of in vivo effects from in vitro data or modeling exposure-response relationships.

59 BASIC BIOLOGICAL SCIENCES↗

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↗

Surrogate modeling of Cellular-Potts agent-based models as a segmentation task using the U-Net neural network architecture

The Cellular-Potts model is a powerful and ubiquitous framework for developing computational models for simulating complex multicellular biological systems. Cellular-Potts models (CPMs) are often computationally expensive due to the explicit modeling of interactions among large numbers of individual model agents and diffusive fields described by partial differential equations (PDEs). In this work, we develop a convolutional neural network (CNN) surrogate model using a U-Net architecture that accounts for periodic boundary conditions. We use this model to accelerate the evaluation of a mechanistic CPM previously used to investigate in vitro vasculogenesis. The surrogate model was trained to predict 100 computational steps ahead (Monte-Carlo steps, MCS), accelerating simulation evaluations by a factor of 562 times compared to single-core CPM code execution on CPU. Over short timescales of up to 3 recursive evaluations, or 300 MCS, our model captures the emergent behaviors demonstrated by the original Cellular-Potts model such as vessel sprouting, extension and anastomosis, and contraction of vascular lacunae. This approach demonstrates the potential for deep learning to serve as a step toward efficient surrogate models for CPM simulations, enabling faster evaluation of computationally expensive CPM simulations of biological processes.

97 MATHEMATICS AND COMPUTING↗

A Finite Element Model for Simulation of Carbon Dioxide Sequestration

We present a hydro-mechanical model, followed by stress, deformation, and shear-slip failure analysis for geological sequestration of carbon dioxide (CO2). The model considers the poroelastic effects by taking into account of the two-way coupling between the geomechanical response and the fluid flow process. Analytical solutions for pressure and deformation fields were derived for a typical geological sequestration scenario in our previous work. A finite element approach is introduced here for numerically solving the hydro-mechanical model with arbitrary boundary conditions. The numerical approach was built on an open-source finite element code Elmer, and results were compared to the analytical solutions. The shear-slip failure analysis was presented based on the numerical results, where the potential failure zone is identified. Information is relevant to the prediction of the maximum sustainable injection rate or pressure. The effects of caprock permeability on the fluid pressure, deformation, stress, and the shear-slip failure zone were also quantitatively studied. It was shown that a larger permeability in caprock and base rock leads to a larger uplift but a smaller shear-slip failure zone.

54 ENVIRONMENTAL SCIENCES↗