Search NASA⌕ Search

SEARCH · Search NASA

Results for “Ensemble computing.”

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

SAChES: Scalable Adaptive Chain-Ensemble Sampling.

We present the development of a parallel Markov Chain Monte Carlo (MCMC) method called SAChES, Scalable Adaptive Chain-Ensemble Sampling. This capability is targed to Bayesian calibration of com- putationally expensive simulation models. SAChES involves a hybrid of two methods: Differential Evo- lution Monte Carlo followed by Adaptive Metropolis. Both methods involve parallel chains. Differential evolution allows one to explore high-dimensional parameter spaces using loosely coupled (i.e., largely asynchronous) chains. Loose coupling allows the use of large chain ensembles, with far more chains than the number of parameters to explore. This reduces per-chain sampling burden, enables high-dimensional inversions and the use of computationally expensive forward models. The large number of chains can also ameliorate the impact of silent-errors, which may affect only a few chains. The chain ensemble can also be sampled to provide an initial condition when an aberrant chain is re-spawned. Adaptive Metropolis takes the best points from the differential evolution and efficiently hones in on the poste- rior density. The multitude of chains in SAChES is leveraged to (1) enable efficient exploration of the parameter space; and (2) ensure robustness to silent errors which may be unavoidable in extreme-scale computational platforms of the future. This report outlines SAChES, describes four papers that are the result of the project, and discusses some additional results.

97 MATHEMATICS AND COMPUTING↗

MPI jobs within MPI jobs: A practical way of enabling task-level fault-tolerance in HPC workflows

While the use of workflows for HPC is growing, MPI interoperability remains a challenge for workflow management systems. The MPI standard and/or its implementations provide a number of ways to build multiple-programs-multiple-data (MPMD) applications. These methods present limitations related to fault tolerance, and are not easy to use. In this paper, we advocate for a novel MPI_Comm_launch function acting as the parallel counterpart of a system(3) call. MPI_Comm_launch allows a child MPI application to be launched inside the resources originally held by processes of a parent MPI application. Two important aspects of MPI_Comm_launch is that it pauses the calling process, and runs the child processes on the parent’s CPU cores, but in an isolated manner with respect to memory. This function makes it easier to build MPMD applications with well-decoupled subtasks. We show how this feature can provide better flexibility and better fault tolerance in ensemble simulations and HPC workflows. Here, we report results showing 2x throughput improvement for application workflows with faults, and scaling results for challenging workloads up to 256 nodes.

97 MATHEMATICS AND COMPUTING↗

Quantum formalism for classical statistics

In static classical statistical systems the problem of information transport from a boundary to the bulk finds a simple description in terms of wave functions or density matrices. While the transfer matrix formalism is a type of Heisenberg picture for this problem, we develop here the associated Schrödinger picture that keeps track of the local probabilistic information. The transport of the probabilistic information between neighboring hypersurfaces obeys a linear evolution equation, and therefore the superposition principle for the possible solutions. Operators are associated to local observables, with rules for the computation of expectation values similar to quantum mechanics. We discuss how non-commutativity naturally arises in this setting. Also other features characteristic of quantum mechanics, such as complex structure, change of basis or symmetry transformations, can be found in classical statistics once formulated in terms of wave functions or density matrices. We construct for every quantum system an equivalent classical statistical system, such that time in quantum mechanics corresponds to the location of hypersurfaces in the classical probabilistic ensemble. For suitable choices of local observables in the classical statistical system one can, in principle, compute all expectation values and correlations of observables in the quantum system from the local probabilistic information of the associated classical statistical system. Realizing a static memory material as a quantum simulator for a given quantum system is not a matter of principle, but rather of practical simplicity.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Rapid Monte Carlo simulation of detector DQE(f)

Purpose: Performance optimization of indirect x-ray detectors requires proper characterization of both ionizing (gamma) and optical photon transport in a heterogeneous medium. As the tool of choice for modeling detector physics, Monte Carlo methods have failed to gain traction as a design utility, due mostly to excessive simulation times and a lack of convenient simulation packages. The most important figure-of-merit in assessing detector performance is the detective quantum efficiency (DQE), for which most of the computational burden has traditionally been associated with the determination of the noise power spectrum (NPS) from an ensemble of flood images, each conventionally having 10{sup 7} − 10{sup 9} detected gamma photons. In this work, the authors show that the idealized conditions inherent in a numerical simulation allow for a dramatic reduction in the number of gamma and optical photons required to accurately predict the NPS. Methods: The authors derived an expression for the mean squared error (MSE) of a simulated NPS when computed using the International Electrotechnical Commission-recommended technique based on taking the 2D Fourier transform of flood images. It is shown that the MSE is inversely proportional to the number of flood images, and is independent of the input fluence provided that the input fluence is above a minimal value that avoids biasing the estimate. The authors then propose to further lower the input fluence so that each event creates a point-spread function rather than a flood field. The authors use this finding as the foundation for a novel algorithm in which the characteristic MTF(f), NPS(f), and DQE(f) curves are simultaneously generated from the results of a single run. The authors also investigate lowering the number of optical photons used in a scintillator simulation to further increase efficiency. Simulation results are compared with measurements performed on a Varian AS1000 portal imager, and with a previously published simulation performed using clinical fluence levels. Results: On the order of only 10–100 gamma photons per flood image were required to be detected to avoid biasing the NPS estimate. This allowed for a factor of 10{sup 7} reduction in fluence compared to clinical levels with no loss of accuracy. An optimal signal-to-noise ratio (SNR) was achieved by increasing the number of flood images from a typical value of 100 up to 500, thereby illustrating the importance of flood image quantity over the number of gammas per flood. For the point-spread ensemble technique, an additional 2× reduction in the number of incident gammas was realized. As a result, when modeling gamma transport in a thick pixelated array, the simulation time was reduced from 2.5 × 10{sup 6} CPU min if using clinical fluence levels to 3.1 CPU min if using optimized fluence levels while also producing a higher SNR. The AS1000 DQE(f) simulation entailing both optical and radiative transport matched experimental results to within 11%, and required 14.5 min to complete on a single CPU. Conclusions: The authors demonstrate the feasibility of accurately modeling x-ray detector DQE(f) with completion times on the order of several minutes using a single CPU. Convenience of simulation can be achieved using GEANT4 which offers both gamma and optical photon transport capabilities.

61 RADIATION PROTECTION AND DOSIMETRY↗

Assimilation of MODIS Snow Cover Through the Data Assimilation Research Testbed and the Community Land Model Version 4

To improve snowpack estimates in Community Land Model version 4 (CLM4), the Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover fraction (SCF) was assimilated into the Community Land Model version 4 (CLM4) via the Data Assimilation Research Testbed (DART). The interface between CLM4 and DART is a flexible, extensible approach to land surface data assimilation. This data assimilation system has a large ensemble (80-member) atmospheric forcing that facilitates ensemble-based land data assimilation. We use 40 randomly chosen forcing members to drive 40 CLM members as a compromise between computational cost and the data assimilation performance. The localization distance, a parameter in DART, was tuned to optimize the data assimilation performance at the global scale. Snow water equivalent (SWE) and snow depth are adjusted via the ensemble adjustment Kalman filter, particularly in regions with large SCF variability. The root-mean-square error of the forecast SCF against MODIS SCF is largely reduced. In DJF (December-January-February), the discrepancy between MODIS and CLM4 is broadly ameliorated in the lower-middle latitudes (2345N). Only minimal modifications are made in the higher-middle (4566N) and high latitudes, part of which is due to the agreement between model and observation when snow cover is nearly 100. In some regions it also reveals that CLM4-modeled snow cover lacks heterogeneous features compared to MODIS. In MAM (March-April-May), adjustments to snowmove poleward mainly due to the northward movement of the snowline (i.e., where largest SCF uncertainty is and SCF assimilation has the greatest impact). The effectiveness of data assimilation also varies with vegetation types, with mixed performance over forest regions and consistently good performance over grass, which can partly be explained by the linearity of the relationship between SCF and SWE in the model ensembles. The updated snow depth was compared to the Canadian Meteorological Center (CMC) data. Differences between CMC and CLM4 are generally reduced in densely monitored regions.

data assimilation↗

A new ensemble-based consistency test for the Community Earth System Model (pyCECT v1.0)

Abstract. Climate simulation codes, such as the Community Earth System Model (CESM), are especially complex and continually evolving. Their ongoing state of development requires frequent software verification in the form of quality assurance to both preserve the quality of the code and instill model confidence. To formalize and simplify this previously subjective and computationally expensive aspect of the verification process, we have developed a new tool for evaluating climate consistency. Because an ensemble of simulations allows us to gauge the natural variability of the model's climate, our new tool uses an ensemble approach for consistency testing. In particular, an ensemble of CESM climate runs is created, from which we obtain a statistical distribution that can be used to determine whether a new climate run is statistically distinguishable from the original ensemble. The CESM ensemble consistency test, referred to as CESM-ECT, is objective in nature and accessible to CESM developers and users. The tool has proven its utility in detecting errors in software and hardware environments and providing rapid feedback to model developers.

58 GEOSCIENCES↗

Modeling laser-driven ion acceleration with deep learning

Developments in machine learning promise to ameliorate some of the challenges of modeling complex physical systems through neural-network-based surrogate models. High-intensity, short-pulse lasers can be used to accelerate ions to mega-electronvolt energies, but to model such interactions requires computationally expensive techniques such as particle-in-cell simulations. Multilayer neural networks allow one to take a relatively sparse ensemble of simulations and generate a surrogate model that can be used to rapidly search the parameter space of interest. In this study, we created an ensemble of over 1,000 simulations modeling laser-driven ion acceleration and developed a surrogate to study the resulting parameter space. A neural-network-based approach allows for rapid feature discovery not possible for traditional parameter scans given the computational cost. A notable observation made during this study was the dependence of ion energy on the pre-plasma gradient length scale. While this methodology harbors great promise for ion acceleration, it has ready application to all topics in which large-scale parameter scans are restricted by significant computational cost or relatively large, but sparse, domains.

70 PLASMA PHYSICS AND FUSION TECHNOLOGY↗

Ensembles of Networks Produced from Neural Architecture Search

Neural architecture search (NAS) is a popular topic at the intersection of deep learning and high performance computing. NAS focuses on optimizing the architecture of neural networks along with their hyperparameters in order to produce networks with superior performance. Much of the focus has been on how to produce a single best network to solve a machine learning problem, but as NAS methods produce many networks that work very well, this affords the opportunity to ensemble these networks to produce an improved result. Additionally, the diversity of network structures produced by NAS drives a natural bias towards diversity of predictions produced by the individual networks. This results in an improved ensemble over simply creating an ensemble that contains duplicates of the best network architecture retrained to have unique weights.

Herron, Emily↗

Automated detection of pulmonary nodules in PET/CT images: Ensemble false-positive reduction using a convolutional neural network technique

Purpose: Automated detection of solitary pulmonary nodules using positron emission tomography (PET) and computed tomography (CT) images shows good sensitivity; however, it is difficult to detect nodules in contact with normal organs, and additional efforts are needed so that the number of false positives (FPs) can be further reduced. In this paper, the authors propose an improved FP-reduction method for the detection of pulmonary nodules in PET/CT images by means of convolutional neural networks (CNNs). Methods: The overall scheme detects pulmonary nodules using both CT and PET images. In the CT images, a massive region is first detected using an active contour filter, which is a type of contrast enhancement filter that has a deformable kernel shape. Subsequently, high-uptake regions detected by the PET images are merged with the regions detected by the CT images. FP candidates are eliminated using an ensemble method; it consists of two feature extractions, one by shape/metabolic feature analysis and the other by a CNN, followed by a two-step classifier, one step being rule based and the other being based on support vector machines. Results: The authors evaluated the detection performance using 104 PET/CT images collected by a cancer-screening program. The sensitivity in detecting candidates at an initial stage was 97.2%, with 72.8 FPs/case. After performing the proposed FP-reduction method, the sensitivity of detection was 90.1%, with 4.9 FPs/case; the proposed method eliminated approximately half the FPs existing in the previous study. Conclusions: An improved FP-reduction scheme using CNN technique has been developed for the detection of pulmonary nodules in PET/CT images. The authors’ ensemble FP-reduction method eliminated 93% of the FPs; their proposed method using CNN technique eliminates approximately half the FPs existing in the previous study. These results indicate that their method may be useful in the computer-aided detection of pulmonary nodules using PET/CT images.

60 APPLIED LIFE SCIENCES↗

Sensitivity and Uncertainty Workflow of Full System SIERRA Models Supporting High Consequence Applications (Executive Summary)

The overall goal of this work was to advance the integrated workflow capabilities to provide the weapon system analysts the tools to construct, communicate and robustly execute end-to-end computational simulation models and analysis. Specifically, this includes developing and running automated, CompSim workflows that map model inputs to responses from large-scale studies executed on a distributed, heterogenous computing environment (e.g., CAD applications on Windows laptop to Sierra FEM codes on Trinity within a single workflow). The workflows developed enable ensemble calculations in support of ND program decisions, including sensitivity analysis, optimization and uncertainty quantification.

97 MATHEMATICS AND COMPUTING↗

Bridging Hydrological Ensemble Simulation and Learning Using Deep Neural Operators

Ensemble-based simulation and learning (ESnL) has long been used in hydrology for parameter inference, but computational demands of process-based ESnL can be quite high. To address this issue, we propose a deep neural operator learning approach. Neural operators are generic machine learning algorithms that can learn functional mappings between infinite-dimensional spaces, providing a highly flexible tool for scientific machine learning. Our approach is built upon DeepONet, a specific deep neural operator, and is designed to address several common problems in hydrology, namely, model parameter estimation, prediction at ungaged locations, and uncertainty quantification. Here we demonstrate the effectiveness of our DeepONet-based workflow using an existing large model ensemble created for an eastern U.S. watershed that is instrumented with 10 streamflow gages. Results suggest DeepONet achieves high efficiency in learning an ML surrogate model from the model ensemble, with the modified Kling-Gupta Efficiency exceeding 0.9 on holdout test sets. Parameter inference, carried out using the trained DeepONet surrogate model and genetic algorithm, also yields robust results. Additionally, we formulate and train a separate DeepONet model for physics-informed, seq-to-seq streamflow forecasting, which further reduces biases in the pre-trained DeepONet surrogate model. While this study focuses primarily on a single watershed, our approach is general and may be extended to enable learning from model ensembles across multiple basins or models. Thus, this research represents a significant contribution to the application of hybrid machine learning in hydrology.

54 ENVIRONMENTAL SCIENCES↗

Coupling Visualization, Simulation, and Deep Learning for Ensemble Steering of Complex Energy Models

We describe a new framework that allows users to explore and steer ensembles of energy systems simulations by coupling multiple energy models and interactive visualization through a dataflow API. Through the visual interface, users can interactively explore complex parameter spaces populated by hundreds, or thousands, of simulation runs and interactively spawn new simulations to 'fill in' regions of interest in the parameter space. The computational and visualization capabilities reside within a general-purpose dataflow architecture for connecting producers of multidimensional timeseries data, such as energy simulations, with consumers of that data, whether they be visualizations, statistical analyses, or datastores. Fast computation and agile dataflow can enhance the engagement with energy simulations, allowing users to populate the parameter space in real time. However, many energy simulations are far too slow to provide an interactive response. To support interactive feedback, we are creating reduced-form simulations developed through machine learning techniques, which provide statistically sound estimates of the results of the full simulations at a fraction of the computational cost. These reduced-form simulations have response times on the order of seconds, suitable for real-time human-in-the-loop design and analysis. The approximation methods apply to a wide range of computational models, including supply-chain models, electric power grid simulations, and building models. Such reduced-form representations do not replace or re-implement existing simulations, but instead supplement them by enabling rapid scenario design and exploration for large ensembles of simulations. The improved understanding, facilitated by the reduced-form models, dataflow API, and visualization tools, allows researchers to better allocate computational resources to capture informative relationships within the system as well as provide a low-cost method for validating and quality-checking large-scale modeling efforts.

97 MATHEMATICS AND COMPUTING↗

Selecting representative geological realizations to model subsurface CO 2 storage under uncertainty

Carbon capture and storage (CCS) is one of the quickest and most effective solutions for reducing carbon emissions. The majority of subsurface storage occurs in saline aquifers, for which geological information is lacking which in turn results in geological uncertainty. To evaluate uncertainty in CO 2 injection projections, the use of multiple geological realizations (GRs) has been practiced very commonly. In this approach, hundreds or thousands of high-resolution GRs is used that quickly becomes computationally expensive. This issue can be addressed with representative geological realizations (RGRs) that preserve the uncertainty domain of the ensemble GRs. Here, in this study, we propose the use of unsupervised machine learning (UML) frameworks, including dissimilarity measurement, dimensionality reduction, clustering and sampling algorithms ta select a predetermined number of RGRs. We compare the simulation outputs of the RGR sets and the ensemble using the Kolmogorov–Smirnov (KS) test to select the best UML. The UML frameworks and their associated selection processes are evaluated using a saline aquifer with a single CO 2 injection well and 200 GRs with varying uncertain petrophysical characteristics. The best UML framework is selected to use only 5% of the GRs while maintaining the uncertainty domain of the ensemble GRs. In addition, the best UML framework is tested using a saline aquifer with three CO 2 injection wells and varied GRs. The results show that our proposed UML framework can be used to choose RGRs, capturing the whole uncertainty domain. Our approach leads to a significant reduction in the computational cost associated with scenario testing, decision-making, and development planning for CO 2 storage sites under geological uncertainty.

58 GEOSCIENCES↗

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↗

Catalytic Activity of an Ensemble of Sites for CO 2 Hydrogenation to Methanol on a ZrO 2 -on-Cu Inverse Catalyst

The significant increase in CO 2 emissions from heavy fossil fuel utilization has raised serious concerns, highlighting the need for effective methods to convert CO 2 into value-added chemicals. Here, in this work, we report a computational investigation on the catalytic activity of ZrO 2 -on-Cu inverse catalysts for CO 2 hydrogenation to methanol, considering highly dispersed ZrO 2 trimers on Cu (111). Such clusters present a large ensemble of formate-containing configurations, Zr 3 O n (OH) m (OCHO) l , making the evaluation of the catalytic activity very challenging. We found that the sites on the various catalyst configurations exhibit markedly different activities for formate hydrogenation, despite their similar free energy and composition. To understand these differences in reactivity, we examined the structural and electronic nature of the low free-energy catalyst configurations and identified that the energy of the lowest unoccupied orbital of the reacting formate, modified by its binding with the catalytic site, is a descriptor for the reaction energy of the formate hydrogenation step. From there, we screened an ensemble of catalyst structures using this descriptor to predict highly active metastable catalyst configurations and computed the reaction pathways and transition states for formate hydrogenation. From this investigation, we distinguished reactive from nonreactive sites and formate species on the ZrO 2 /Cu inverse catalyst based on structural and electronic features. We showed that rare metastable configurations control the activity. Additionally, an efficient method for examining the reactivity of a large number of coexisting catalyst structures was developed.

catalysts↗

Evaluating skill in predicting the Interdecadal Pacific Oscillation in initialized decadal climate prediction hindcasts in E3SMv1 and CESM1 using two different initialization methods and a small set of start years

Abstract It is a daunting challenge to conduct initialized hindcasts with enough ensemble members and associated start years to form a drifted climatology from which to compute the anomalies necessary to quantify the skill of the hindcasts when compared to observations. This limits the ability to experiment with case studies and other applications where only a few initial years are needed. Here we run a set of hindcasts with CESM1 and E3SMv1 using two different initialization methods for a limited set of start years and use the respective uninitialized free-running historical simulations to form the model climatologies. Since the drifts from the observed initial states in the hindcasts toward the uninitialized model state are large and rapid, after a few years the drifted initialized models approach the uninitialized model climatological errors. Therefore, hindcasts from the limited start years can use the uninitialized climatology to represent the drifted model states after about lead year 3, providing a means to compute forecast anomalies in the absence of a large hindcast sample. There is comparable skill for predicting spatial patterns of multi-year Pacific sea surface temperature anomalies in the domain of the Interdecadal Pacific Oscillation using this method compared to the conventional methodology with a large hindcast data set, though there is a model dependence to the drifts in the two initialization methods.

54 ENVIRONMENTAL SCIENCES↗

Approximate solutions of the filtered radiative transfer equation in large eddy simulations of turbulent reactive flows

An analysis of the relevance of turbulence-radiation interaction in the numerical simulation of turbulent reactive flows is presented. A semi-causal stochastic model was used to generate a time-series of turbulent scalar fluctuations along optical paths of Sandia flame D, a widely studied piloted turbulent jet nonpremixed flame. The radiative transfer equation was integrated along these paths for every realization using a grid resolution typical of a direct numerical simulation. The correlated k-distribution method was employed to compute the radiative properties of the medium. The results were used to determine the ensemble average, as well as the extreme values, of quantities that indicate the importance of the turbulence-radiation interaction. Several approximate methods are then proposed to solve the filtered radiative transfer equation in the framework of large eddy simulations. The proposed methods are applicable along with combustion models that either assume the filtered probability density function of a conserved scalar or solve a transport equation for a joint scalar or joint scalar/velocity filtered density function. It is concluded that the errors resulting from neglecting the turbulence-radiation interaction in large eddy simulations are much lower than those found in Reynolds-averaged Navier-Stokes calculations. The optically thin fluctuation approximation may be extended to large eddy simulations yielding predictions in excellent agreement with the reference solution. If the turbulence-radiation interaction is accounted for using this approximation, the average relative error of the filtered total radiation intensity is generally below 0.3% for the studied flame. (author)

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗