Search NASA⌕ Search

SEARCH · Search NASA

Results for “Ensemble”

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 37 records · Page 2
All sourcesNASA NTRSDOE OSTI
NASA NTRS1992

The macroscopic behavior of cumulus ensembles simulated by a cumulus ensemble model

A number of simulation experiments are performed with the 2D UCLA cumulus ensemble model to study the macroscopic behavior of cumulus convection under a variety of different large-scale and underlying surface conditions. Specifically, the modulation of cumulus activity by the imposed large-scale processes and the eddy kinetic energy (EKE) budget are investigated in detail. In all simulations, cumulus convection is rather strongly modulated by large-scale advective processes in spite of the existence of some nonmodulated high-frequency fluctuations. The modulation exhibits some phase delays, however, when the basic wind shear is strong. This is presumably associated with the existence of mesoscale convective organization. The EKE budget analysis shows that the net eddy buoyancy generation rate is nearly zero for a wind range of cumulus ensembles.

Xu, Kuan-Man↗
NASA NTRS2014

Propellant Handler's Ensemble (PHE) Aka Self-Contained Atmospheric Protective Ensemble (SCAPE), Ventilator Improvement Study Project

The overall objective for this project is to evaluate two candidate alternatives for the existing Propellant Handler's Ensemble (PHE) escape ventilator. The new candidate ventilators use newer technology with similar quantities of air at approximately half the weight of the current ventilator. Ventilators are typically used to ingress/egress a hazardous work area when hard line air is provided at the work area but the hose is not long enough to get the operator to and from the staging area to the work area. The intent of this test is to verify that the new ventilators perform as well as or better than the current ventilators in maintaining proper oxygen (O2) and carbon dioxide (CO2) levels in the PHE during a typical use for the rated time period (~10 minutes). We will evaluate two new units comparing them to the existing unit. Subjects will wear the Category I version of the Propellant Handler's Ensemble with the rear suit pouch snapped.

PHE↗
DOE OSTI2024

Analysis of a Computational Framework for Bayesian Inverse Problems: Ensemble Kalman Updates and MAP Estimators under Mesh Refinement

This paper analyzes a popular computational framework to solve infinite-dimensional Bayesian inverse problems, discretizing the prior and the forward model in a finite-dimensional weighted inner product space. We demonstrate the benefit of working on a weighted space by establishing operator-norm bounds for finite element and graph-based discretizations of Matérn-type priors and deconvolution forward models. For linear-Gaussian inverse problems, we develop a general theory to characterize the error in the approximation to the posterior. We also embed the computational framework into ensemble Kalman methods and MAP estimators for nonlinear inverse problems. Furthermore, our operator-norm bounds for prior discretizations guarantee the scalability and accuracy of these algorithms under mesh refinement.

Bayesian inverse problem↗
DOE OSTI2026

HydraGNN_OPF_GFM_2026 - Ensemble of predictive graph foundation models for power grid applications

This dataset supports research on graph foundation models for optimal power flow (OPF) on electric grids using HydraGNN. It contains heterogeneous graph representations of PGLib-OPF cases spanning systems from 14 to 13,659 buses, together with packed HDF5 datasets for pretraining, feasibility classification, and N-1 contingency analysis. The release includes OPF solution data, downstream fine-tuning datasets, pretrained HeteroSAGE and HeteroHEAT model checkpoints, hyperparameter-optimization summaries across multiple heterogeneous GNN architectures, and aggregated fine-tuning results for sample-efficiency studies. The dataset is designed to enable scalable training, evaluation, and transfer-learning studies for OPF surrogate modeling, including node-level AC-OPF solution prediction, graph-level prediction, feasibility classification, operating-condition generalization, and contingency-response tasks.

24 POWER TRANSMISSION AND DISTRIBUTION↗
NASA NTRS2020

Designing an Optimal Ensemble Strategy for GMAO S2S Forecast System

The NASA Global Modeling and Assimilation Office (GMAO) Sub-seasonal to Seasonal (S2S) prediction system is being readied for a major upgrade. An important factor in successful extended range forecasting is the definition of the ensemble. Our overall strategy is to run a relatively large ensemble of about 40 members up to 3 months (focusing on the sub-seasonal forecast problem), after which we sub-sample the ensemble, and continue the forecast with about 10 members (up to 12 months). Here we present the results of our testing of various ways to generate the initial perturbations and the validation of a stratified sampling approach for choosing the members of the smaller ensemble. For the initialization of the ensemble we propose a combination of lagged and burst initial conditions. To generate perturbations for the burst ensemble members we used scaled differences of pairs of analysis states (chosen randomly from the corresponding season) separated by 1-10 days. We consider perturbing separately the atmosphere and the ocean, or both. By varying the separation times between the analysis states, we are able to produce perturbations that resemble well-known modes of variability. Focusing on the ENSO SST indices, we found that all types of perturbations are important for the ensemble spread with, however, considerable differences in the timing of the impacts on spread for the atmospheric and oceanic perturbations.Our initial (larger) ensemble size was determined so as to maximize the skill of predicting some of the leading modes of boreal winter atmospheric modes (namely the NAO, PNA and AO). Since it is not feasible for us to run with the larger ensemble beyond about 3 months, we employ a stratified sampling procedure that identifies the emerging directions of error growth to subset the ensemble. By comparing the results from the stratified ensemble with that of the randomly sampled ensemble of the same size, we find that the former provides substantially better estimates the mean of the original large ensemble.

Borovikov, Anna↗
DOE OSTI2020

GMRES with embedded ensemble propagation for the efficient solution of parametric linear systems in uncertainty quantification of computational models

In a previous work, embedded ensemble propagation was proposed to improve the efficiency of sampling-based uncertainty quantification methods of computational models on emerging computational architectures. It consists of simultaneously evaluating the model for a subset of samples together, instead of evaluating them individually. A first approach introduced to solve parametric linear systems with ensemble propagation is ensemble reduction. In Krylov methods for example, this reduction consists in coupling the samples together using an inner product that sums the sample contributions. Ensemble reduction has the advantages of being able to use optimized implementations of BLAS functions and having a stopping criterion which involves only one scalar. However, the reduction potentially decreases the rate of convergence due to the gathering of the spectra of the samples. In this paper, we investigate a second approach: ensemble propagation without ensemble reduction in the case of GMRES. This second approach solves each sample simultaneously but independently to improve the convergence compared to ensemble reduction. This raises two new issues which are solved in this paper: the fact that optimized implementations of BLAS functions cannot be used anymore and that ensemble divergence, whereby individual samples within an ensemble must follow different code execution paths, can occur. We tackle those issues by implementing a high-performing ensemble GEMV and by using masks. The proposed ensemble GEMV leads to a similar cost per GMRES iteration for both approaches, i.e. with and without reduction. For illustration, we study the performances of the new linear solver in the context of a mesh tying problem. Furthermore, this example demonstrates improved ensemble propagation speed-up without reduction.

BLAS↗
DOE OSTI2023

Predictive performance of multi-model ensemble forecasts of COVID-19 across European nations

Background: Short-term forecasts of infectious disease burden can contribute to situational awareness and aid capacity planning. Based on best practice in other fields and recent insights in infectious disease epidemiology, one can maximise the predictive performance of such forecasts if multiple models are combined into an ensemble. Here, we report on the performance of ensembles in predicting COVID-19 cases and deaths across Europe between 08 March 2021 and 07 March 2022. Methods: We used open-source tools to develop a public European COVID-19 Forecast Hub. We invited groups globally to contribute weekly forecasts for COVID-19 cases and deaths reported by a standardised source for 32 countries over the next 1–4 weeks. Teams submitted forecasts from March 2021 using standardised quantiles of the predictive distribution. Each week we created an ensemble forecast, where each predictive quantile was calculated as the equally-weighted average (initially the mean and then from 26th July the median) of all individual models’ predictive quantiles. We measured the performance of each model using the relative Weighted Interval Score (WIS), comparing models’ forecast accuracy relative to all other models. We retrospectively explored alternative methods for ensemble forecasts, including weighted averages based on models’ past predictive performance. Results: Over 52 weeks, we collected forecasts from 48 unique models. We evaluated 29 models’ forecast scores in comparison to the ensemble model. We found a weekly ensemble had a consistently strong performance across countries over time. Across all horizons and locations, the ensemble performed better on relative WIS than 83% of participating models’ forecasts of incident cases (with a total N=886 predictions from 23 unique models), and 91% of participating models’ forecasts of deaths (N=763 predictions from 20 models). Across a 1–4 week time horizon, ensemble performance declined with longer forecast periods when forecasting cases, but remained stable over 4 weeks for incident death forecasts. In every forecast across 32 countries, the ensemble outperformed most contributing models when forecasting either cases or deaths, frequently outperforming all of its individual component models. Among several choices of ensemble methods we found that the most influential and best choice was to use a median average of models instead of using the mean, regardless of methods of weighting component forecast models. Conclusions: Our results support the use of combining forecasts from individual models into an ensemble in order to improve predictive performance across epidemiological targets and populations during infectious disease epidemics. Our findings further suggest that median ensemble methods yield better predictive performance more than ones based on means. Our findings also highlight that forecast consumers should place more weight on incident death forecasts than incident case forecasts at forecast horizons greater than 2 weeks. Funding: AA, BH, BL, LWa, MMa, PP, SV funded by National Institutes of Health (NIH) Grant 1R01GM109718, NSF BIG DATA Grant IIS-1633028, NSF Grant No.: OAC-1916805, NSF Expeditions in Computing Grant CCF-1918656, CCF-1917819, NSF RAPID CNS-2028004, NSF RAPID OAC-2027541, US Centers for Disease Control and Prevention 75D30119C05935, a grant from Google, University of Virginia Strategic Investment Fund award number SIF160, Defense Threat Reduction Agency (DTRA) under Contract No. HDTRA1-19-D-0007, and respectively Virginia Dept of Health Grant VDH-21-501-0141, VDH-21-501-0143, VDH-21-501-0147, VDH-21-501-0145, VDH-21-501-0146, VDH-21-501-0142, VDH-21-501-0148. AF, AMa, GL funded by SMIGE - Modelli statistici inferenziali per governare l'epidemia, FISR 2020-Covid-19 I Fase, FISR2020IP-00156, Codice Progetto: PRJ-0695. AM, BK, FD, FR, JK, JN, JZ, KN, MG, MR, MS, RB funded by Ministry of Science and Higher Education of Poland with grant 28/WFSN/2021 to the University of Warsaw. BRe, CPe, JLAz funded by Ministerio de Sanidad/ISCIII. BT, PG funded by PERISCOPE European H2020 project, contract number 101016233. CP, DL, EA, MC, SA funded by European Commission - Directorate-General for Communications Networks, Content and Technology through the contract LC-01485746, and Ministerio de Ciencia, Innovacion y Universidades and FEDER, with the project PGC2018-095456-B-I00. DE., MGu funded by Spanish Ministry of Health / REACT-UE (FEDER). DO, GF, IMi, LC funded by Laboratory Directed Research and Development program of Los Alamos National Laboratory (LANL) under project number 20200700ER. DS, ELR, GG, NGR, NW, YW funded by National Institutes of General Medical Sciences (R35GM119582; the content is solely the responsibility of the authors and does not necessarily represent the official views of NIGMS or the National Institutes of Health). FB, FP funded by InPresa, Lombardy Region, Italy. HG, KS funded by European Centre for Disease Prevention and Control. IV funded by Agencia de Qualitat i Avaluacio Sanitaries de Catalunya (AQuAS) through contract 2021-021OE. JDe, SMo, VP funded by Netzwerk Universitatsmedizin (NUM) project egePan (01KX2021). JPB, SH, TH funded by Federal Ministry of Education and Research (BMBF; grant 05M18SIA). KH, MSc, YKh funded by Project SaxoCOV, funded by the German Free State of Saxony. Presentation of data, model results and simulations also funded by the NFDI4Health Task Force COVID-19 ( https://www.nfdi4health.de/task-force-covid-19-2 ) within the framework of a DFG-project (LO-342/17-1). LP, VE funded by Mathematical and Statistical modelling project (MUNI/A/1615/2020), Online platform for real-time monitoring, analysis and management of epidemic situations (MUNI/11/02202001/2020); VE also supported by RECETOX research infrastructure (Ministry of Education, Youth and Sports of the Czech Republic: LM2018121), the CETOCOEN EXCELLENCE (CZ.02.1.01/0.0/0.0/17-043/0009632), RECETOX RI project (CZ.02.1.01/0.0/0.0/16-013/0001761). NIB funded by Health Protection Research Unit (grant code NIHR200908). SAb, SF funded by Wellcome Trust (210758/Z/18/Z).

60 APPLIED LIFE SCIENCES↗
NASA NTRS2020

Ensemble PM2.5 Forecasting during the 2018 Camp Fire Event Using the HYSPLIT Transport and Dispersion Model

Biomass burning releases a vast amount of aerosols into the atmosphere, often leading to severe air quality and health problems. Prediction of the air quality effects from biomass burning emissions is challenging due to uncertainties in fire emission, plume rise calculation, and other model inputs/processes. Ensemble forecasting is increasingly used to represent model uncertainties. In this paper, an ensemble forecast was conducted to predict surface PM2.5 during the 2018 California Camp Fire event using the NOAA HYSPLIT dispersion model at 0.1- degree horizontal resolution. Different combinations of four satellite-based fire emission datasets (FEER, FLAMBE, GBBEPx and GFAS), two plume rise schemes (Briggs and Sofiev), various meteorology inputs and model setup options were used to create the forecast ensemble, for a total of 112 experiments. The performance of each ensemble member and the ensemble mean were evaluated using ground-based observations, with four statistical metrics and an overall rank. The ensemble spread of the 112 members reached 1000 μg/m3, highlighting the large uncertainty in wildfire forecast. The ensemble mean displayed the best performance. Each fire emission product contributed to one or more members among the top ten performers, revealing the forecasting dependence on both the quality of fire emissions data and model representation of emission, transport and removal processes. In addition, an ensemble size reduction technique was introduced. With the help of this technique, the ensemble size was reduced from 112 to 28 members and still produced an ensemble mean that yielded comparable or even better performance to that of the full ensemble.

Biomass Burning↗
NASA NTRS2020

Designing an Optimal Ensemble Strategy for GMAO S2S Forecast System

GMAO Sub/Seasonal prediction system (S2S) is being readied for a major upgrade to GEOS-S2S Version 3. An important factor in successful extended range forecast is the definition of an ensemble For initialization of the ensemble we propose a combination of lagged and burst initial conditions. We plan to run a relatively large ensemble of 40 members for sub-seasonal forecast (up to 3 months), at which point we sub-sample the ensemble, and continue the forecast with 10 members (up to 12 months). Here we present the results of the extensive testing of various ways to generate the perturbations to the initial conditions and the validation of the stratified sampling strategy we chose.To generate perturbations for the burst ensemble members we used scaled differences of pairs of analysis states separated by 1-10 days, randomly chosen from a corresponding season. We considered perturbing separately only the atmospheric fields or only the ocean or both of the forecast initial conditions. Considering varying separation times between the analysis states, we were able to produce perturbations sampling various modes of variability. Focusing on the ENSO SST indices, we found that all types of perturbations are important for the ensemble spread.Our ensemble size for sub-seasonal forecasts was determined as to maximize the skill of predicting some of the leading modes of boreal winter atmospheric modes, NAO, PNA and AO. It is not feasible to run equally large ensemble for seasonal forecasts. Using a stratified sampling procedure we can identify the emerging directions of error growth. By comparing the stratified ensemble with randomly sampled ensemble of the same size, we were able to show that the former better estimates the mean of the original large ensemble.

Borovikov, Anna↗
NASA NTRS2016

Lessons from Climate Modeling on the Design and Use of Ensembles for Crop Modeling

Working with ensembles of crop models is a recent but important development in crop modeling which promises to lead to better uncertainty estimates for model projections and predictions, better predictions using the ensemble mean or median, and closer collaboration within the modeling community. There are numerous open questions about the best way to create and analyze such ensembles. Much can be learned from the field of climate modeling, given its much longer experience with ensembles. We draw on that experience to identify questions and make propositions that should help make ensemble modeling with crop models more rigorous and informative. The propositions include defining criteria for acceptance of models in a crop MME, exploring criteria for evaluating the degree of relatedness of models in a MME, studying the effect of number of models in the ensemble, development of a statistical model of model sampling, creation of a repository for MME results, studies of possible differential weighting of models in an ensemble, creation of single model ensembles based on sampling from the uncertainty distribution of parameter values or inputs specifically oriented toward uncertainty estimation, the creation of super ensembles that sample more than one source of uncertainty, the analysis of super ensemble results to obtain information on total uncertainty and the separate contributions of different sources of uncertainty and finally further investigation of the use of the multi-model mean or median as a predictor.

Model ensembles↗
NASA NTRS2014

Conservation of Mass and Preservation of Positivity with Ensemble-Type Kalman Filter Algorithms

This paper considers the incorporation of constraints to enforce physically based conservation laws in the ensemble Kalman filter. In particular, constraints are used to ensure that the ensemble members and the ensemble mean conserve mass and remain nonnegative through measurement updates. In certain situations filtering algorithms such as the ensemble Kalman filter (EnKF) and ensemble transform Kalman filter (ETKF) yield updated ensembles that conserve mass but are negative, even though the actual states must be nonnegative. In such situations if negative values are set to zero, or a log transform is introduced, the total mass will not be conserved. In this study, mass and positivity are both preserved by formulating the filter update as a set of quadratic programming problems that incorporate non-negativity constraints. Simple numerical experiments indicate that this approach can have a significant positive impact on the posterior ensemble distribution, giving results that are more physically plausible both for individual ensemble members and for the ensemble mean. In two examples, an update that includes a non-negativity constraint is able to properly describe the transport of a sharp feature (e.g., a triangle or cone). A number of implementation questions still need to be addressed, particularly the need to develop a computationally efficient quadratic programming update for large ensemble.

EnKF↗
NASA NTRS2013

Mass Conservation and Positivity Preservation with Ensemble-type Kalman Filter Algorithms

Maintaining conservative physical laws numerically has long been recognized as being important in the development of numerical weather prediction (NWP) models. In the broader context of data assimilation, concerted efforts to maintain conservation laws numerically and to understand the significance of doing so have begun only recently. In order to enforce physically based conservation laws of total mass and positivity in the ensemble Kalman filter, we incorporate constraints to ensure that the filter ensemble members and the ensemble mean conserve mass and remain nonnegative through measurement updates. We show that the analysis steps of ensemble transform Kalman filter (ETKF) algorithm and ensemble Kalman filter algorithm (EnKF) can conserve the mass integral, but do not preserve positivity. Further, if localization is applied or if negative values are simply set to zero, then the total mass is not conserved either. In order to ensure mass conservation, a projection matrix that corrects for localization effects is constructed. In order to maintain both mass conservation and positivity preservation through the analysis step, we construct a data assimilation algorithms based on quadratic programming and ensemble Kalman filtering. Mass and positivity are both preserved by formulating the filter update as a set of quadratic programming problems that incorporate constraints. Some simple numerical experiments indicate that this approach can have a significant positive impact on the posterior ensemble distribution, giving results that are more physically plausible both for individual ensemble members and for the ensemble mean. The results show clear improvements in both analyses and forecasts, particularly in the presence of localized features. Behavior of the algorithm is also tested in presence of model error.

EnKR↗
NASA NTRS2023

Seasonal Soil Freeze/Thaw Variability Across North America via Ensemble Land Surface Modeling

Land surface modeling provides the opportunity to investigate winter soil processes such as soil freezing and thawing over large domains. However, the variability in simulated winter soil characteristics among land surface models and forcing datasets is not well understood. In this study, a nine-member ensemble was employed to characterize the spatial and inter-annual variability of three winter soil characteristics, annual number of frozen days, annual minimum temperature and annual number of freeze-thaw cycles over North America during the water years 2010–2016. The ensemble, including three land surface model (JULES, Noah2.7.1 and Noah-MP) and three forcing datasets (ECMWF, GDAS and MERRA2), was developed through the Snow Ensemble Uncertainty Project (SEUP). In many regions, there was remarkably good agreement across the ensemble for the winter soil temperatures. However, the variability among the ensemble's annual number of frozen days, as quantified with standard deviation, exceeded 150 days at the northern Pacific coastline. While the differences among the ensemble members' annual minimum temperature were typically <3 °C, the differences exceeded 6 °C north of 50°N. The ensemble members generally agreed within one to three freeze-thaw cycles in mid latitude regions, Alaska, and the south and west coasts of the USA. High variability among the ensemble, more than six freeze-thaw cycles, occurred in the Great Plains, northern Pacific coastline, and along the Appalachian Mountains. Differences in winter soil temperature characteristics were more apparent among the LSMs rather than the meteorological forcing datasets. Except for maritime regions, the Noah2.7.1 members had the fewest annual number of frozen days and the least number of freeze-thaw cycles. Noah2.7.1 also had the coldest annual minimum temperatures except for ephemeral regions. Comparisons between in-situ observations and the SEUP estimates of winter soil characteristics revealed that the modeled frozen period was much longer than observed, that the modeled annual minimum temperatures were much colder than observed, and the modeled freeze-thaw cycles occurred more frequently than observed. Excluding the high latitude sites, the observed frozen period is less than two months with the minimum temperature above − 5 °C at most of the studied sites, while the ensemble members simulated, on average, a four month frozen period with − 10 °C minimum temperature. Errors in capturing snow during the accumulation period appear to impact differences in modeled versus observed soil temperature throughout the entire winter.

Soil freeze/thaw↗
DOE OSTI2020

Ensembles of Metastable States Govern Heterogeneous Catalysis on Dynamic Interfaces

Heterogeneous catalysis is at the heart of chemical industry. Being able to tune and design efficient catalysts for processes of interest is of an utmost importance, and for this, the molecular-level understanding of heterogeneous catalysts is the first step, and indeed a prime focus of the modern catalysis research. For a long time, a single most thermodynamically stable structure of the catalytic interface attained in reaction conditions had been envisioned as the reactive phase. However, some catalytic interfaces continue to undergo structural dynamics in the steady state, triggered by high temperatures, pressures, and binding and changing reagents. Among particularly dynamic interfaces are such widely-used catalysts as crystalline and amorphous surfaced supporting (sub-)nano metallic clusters. Recently, it became clear that this dynamic fluxionality causes the supported clusters to populate many distinct structural and stoichiometric states in catalytic conditions. Hence, the catalytic interface should be viewed as an evolving statistical ensemble of many (not one) structures. As such, every member in the ensemble contributes to the properties of the catalyst differently, and in proportion to its probability to be populated. This new notion flips the established paradigm and calls for new theory, modeling approaches, operando measurements, and updated design strategies. The statistical ensemble nature of surface-supported sub-nano cluster catalysts can be exemplified by oxide-supported and adsorbate-covered Pt, Pd, Cu, CuPd clusters, catalytic toward oxidative and non-oxidative dehydrogenation. They have access to a variety of 3D and quasi-2D shapes. The compositions of their thermal ensembles are dependent on the cluster size, leading to size-specific catalytic activities and the famous “every atom counts” phenomenon. The support and adsorbates affect catalyst structures, and state of the reacting species causes the ensemble to change in every reaction intermediate. The most stable member of the ensemble dominates the thermodynamic properties of the corresponding intermediate, whereas the kinetics can be determined by more active but less populated metastable catalyst states, and that suggests that many earlier studies might have overlooked the actual active sites. Both effects depend on the relative time-scales of catalyst restructuring and reaction dynamics. The catalyst may routinely operate off-equilibrium. Ensemble phenomena lead to surprising exceptions from established rules of catalysis, such as scaling relations, and Arrhenius behavior. Catalyst deactivation is also an ensemble property, and its extent of mitigation can be predicted through the new paradigm. These findings were enabled by advances in theory, such as global optimization and subsequent utilization of multiple local minima, and pathways sampling, as well as operando catalyst characterization. The fact that the per-site and per-species resolution is needed for the description and predicting of catalyst properties gives theory the central role in catalysis research, as most experiments provide ensemble-average information and cannot detect the crucial minority species that may be responsible for the catalytic activity.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗
DOE OSTI2022

Running Ensemble Workflows at Extreme Scale: Lessons Learned and Path Forward

The ever-increasing volumes of scientific data combined with sophisticated techniques for extracting information from them have led to the increasing popularity of ensemble workflows which are a collection of runs of individual workflows. A traditional approach followed by scientists to run ensembles is to rely on simple scripts to execute different runs and manage resources. This approach is not scalable and is error-prone, thereby motivating the development of workflow management systems that specialize in executing ensembles on HPC clusters. However, when the size of both the ensemble and the target system reach extreme scales, existing workflow management systems face new challenges that hamper their efficient execution. In this paper, we describe our experience scaling an ensemble workflow from the computational biology domain from the early design stages to the execution at extreme scale on Summit, a leadership class supercomputer at the Oak Ridge National Laboratory. We discuss challenges that arise when scaling ensembles to several million runs on thousands of HPC nodes. We identify challenges with composition of the ensemble itself, its execution at large scale, post-processing of the generated data, and scalability of the file system. Based on the experience acquired, we develop a generic vision of the capabilities and abstractions to add to existing workflow management systems to enable the execution of ensemble workflows at extreme scales. We believe that the understanding of these fundamental challenges will help application teams along with workflow system developers with designing the next generation of infrastructure for composing and executing extreme-scale ensemble workflows.

Mehta, Kshitij↗
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↗