SEARCH · Search NASA

Results for “stochastic dynamics of structures”

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.

SourceResearch connectionFrom yearThrough year

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

At least 127 records · Page 7

All sources NASA NTRS DOE OSTI

DOE OSTI2020

Adaptive Recovery Model: Designing Systems for Testing Tracing and Vaccination to Support COVID-19 Recovery Planning.

This report documents a new approach to designing disease control policies that allocate scarce testing, contact tracing, and vaccination resources to better control community transmission of COVID19 or similar diseases. The Adaptive Recovery Model (ARM) combines a deterministic compartmental disease model with a stochastic network disease propagation model to enable us to simulate COVID-19 community spread through the lens of two complementary modeling motifs. ARM contact networks are derived from cell-phone location data that have been anonymized and interpreted as individual arrivals to specic public locations. Modeling disease spread over these networks allows us to identify locations within communities conducive to rapid disease spread. ARM applies this model- and data-derived abstractions of community transmission to evaluate the effectiveness of disease control measures including targeted social distancing, contact tracing, testing and vaccination. The architecture of ARM provides a unique capacity to help decision makers understand how best to deploy scarce testing, tracing and vaccination resources to minimize disease-spread potential in a community. This document details the novel mathematical formulations underlying ARM, presents a dynamical stability analysis of the deterministic model components, a sensitivity analysis of control parameters and network structure, and summarizes a process for deriving contact networks from cell-phone location data. An example use case steps through applying ARM to evaluate three targeted social distancing policies using Bernalillo County, New Mexico as an exemplar test locale. This step-by-step analysis demonstrates how ARM can be used to measure the relative performance of competing public health policies. Initial scenario tests of ARM shows that ARMs design focus on resource utilization rather than simple incidence prediction can provide decision makers with additional quantitative guidance for managing ongoing public health emergencies and planning future responses.

29 ENERGY PLANNING, POLICY, AND ECONOMY↗

DOE OSTI2021

Combined selection of the dynamic model and modeling error in nonlinear aeroelastic systems using Bayesian Inference

Here, we report a Bayesian framework for concurrent selection of physics-based models and (modeling) error models. We investigate the use of colored noise to capture the mismatch between the predictions of calibrated models and observational data that cannot be explained by measurement error alone within the context of Bayesian estimation for stochastic ordinary differential equations. Proposed models are characterized by the average data-fit, a measure of how well a model fits the measurements, and the model complexity measured using the Kullback–Leibler divergence. The use of a more complex error models increases the average data-fit but also increases the complexity of the combined model, possibly over-fitting the data. Bayesian model selection is used to find the optimal physical model as well as the optimal error model. The optimal model is defined using the evidence, where the average data-fit is balanced by the complexity of the model. The effect of colored noise process is illustrated using a nonlinear aeroelastic oscillator representing a rigid NACA0012 airfoil undergoing limit cycle oscillations due to complex fluid–structure interactions. Several quasi-steady and unsteady aerodynamic models are proposed with colored noise or white noise for the model error. The use of colored noise improves the predictive capabilities of simpler models.

42 ENGINEERING↗

DOE OSTI2024

Resilience of the slow component in timescale separated synchronized oscillators

Physiological networks are usually made of a large number of biological oscillators evolving on a multitude of different timescales. Phase oscillators are particularly useful in the modelling of the synchronization dynamics of such systems. If the coupling is strong enough compared to the heterogeneity of the internal parameters, synchronized states might emerge where phase oscillators start to behave coherently. Here, we focus on the case where synchronized oscillators are divided into a fast and a slow component so that the two subsets evolve on separated timescales. We assess the resilience of the slow component by, first, reducing the dynamics of the fast one using Mori-Zwanzig formalism. Second, we evaluate the variance of the phase deviations when the oscillators in the two components are subject to noise with possibly distinct correlation times. From the general expression for the variance, we consider specific network structures and show how the noise transmission between the fast and slow components is affected. Interestingly, we find that oscillators that are among the most robust when there is only a single timescale, might become the most vulnerable when the system undergoes a timescale separation. We also find that layered networks seem to be insensitive to such timescale separations.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2024

Resilience of the slow component in timescale-separated synchronized oscillators

59 BASIC BIOLOGICAL SCIENCES↗

DOE OSTI2023

An Efficient Three-Dimensional CFD-Based Numerical Wave Tank for a Wave Energy Converter in Extreme Irregular Waves: Preprint

A numerical wave tank approach for computational fluid dynamics (CFD) modelling of an extreme irregular seastate is presented. The technique couples a potential flow solution with a CFD solver for more efficient numerical predictions. This method has recently become attractive both for the research community and the industry working with offshore structures. The model is used to determine the response of a submerged pressure differential wave energy converter (WEC) in a fully nonlinear irregular wave condition using the high fidelity CFD code, STAR-CCM+. Potential flow based numerical models are commonly used to predict motions and performance of wave energy converters. Wave kinematics can deviate from potential flow predictions for extreme wave conditions; the excitation loads on an absorber can also be increasingly influenced by viscous effects, not captured in engineering level models. In these extreme conditions, a Reynolds-averaged Navier-Stokes CFD model can better predict motions and loads for a WEC. Long time series with varying random seed numbers can be used to identify singular extreme wave events from a stochastic irregular sea state. This approach simulates a more realistic wave series for a given sea state than a regular wave or a focused wave. However, it is computationally infeasible to run these long time series for three-dimensional (3D) CFD simulations. In this work, two-dimensional (2D) CFD simulations with a long domain allow the full development of an extreme nonlinear wave condition. The results are used to identify extreme events from a 50-year storm condition for the PacWave site off the coast of Oregon. A relatively short time window including this extreme event is then mapped to a 3D simulation using a user defined wave methodology. Convergence studies for domain length, wave forcing lengths, and time before the extreme event were conducted.

cfd↗

DOE OSTI2021

Discrete stochastic model of point defect-dislocation interaction for simulating dislocation climb

Dislocation climb is an important high temperature process in metals plasticity, responsible for the phenomena such as creep, swelling, or hardening. Climb is defined by the ability of dislocations to leave their original glide plane by interacting with point defects. As such, dislocation climb is controlled by point defect diffusion/absorption/emission, all of which involve thermal activation. The existing thermodynamically consistent models for climb are generally formulated in a continuum framework, through the definition of effective defect fluxes and climb propensities in response to thermodynamic driving forces. However, the point-wise discrete nature of vacancies (and/or self-interstitials) confers a highly discrete nature to the climb dynamics, which is also strongly affected by elastic forces. The combination of discreteness, thermal activation, and elasticity makes this process too challenging for direct atomistic methods such as molecular dynamics. Here we develop a kinetic Monte Carlo model that captures vacancy generation and transport kinetics acting in conjuction with the evolving elastic fields provided by discrete dislocation dynamics simulations. The two models are coupled via the applied stresses and stress gradients generated by dislocation structures at vacancy locations. Our simulations reveal two surprising results. First that climb is dominated by vacancy emission even when the background vacancy concentration is much higher than the equilibrium one. And, second, that climb velocities might be much faster than otherwise believed when one uses the classical theories of climb. These effects are due to the locality of vacancy-dislocation processes, which are not captured in classical treatments that assume smooth vacancy fluxes and homogeneous concentrations. We apply the method to study elementary climb processes in body-centered cubic iron and furnish climb mobility functions to be used in parametric dislocation dynamics and/or crystal plasticity simulations. Here, we apply the technique to study non-conservative plastic bypass of spherical precipitates by edge dislocations and point out the differences between our discrete approach and existing continuum formulations.

36 MATERIALS SCIENCE↗

DOE OSTI2025

Deep reinforcement learning control for co-optimizing energy consumption, thermal comfort, and indoor air quality in an office building

With the recent demand for decarbonization and energy efficiency, advanced HVAC control using Deep Reinforcement Learning (DRL) becomes a promising solution. Due to its flexible structures, DRL has been successful in energy reduction for many HVAC systems. However, only a few researches applied DRL agents to manage the entire central HVAC system and control multiple components in both the water loop and the air loop, owing to its complex system structures. Moreover, those researches have not extended their applications by incorporating the indoor air quality, especially both CO2 and PM2.5concentrations, on top of energy saving and thermal comfort, as achieving those objectives simultaneously can cause multiple control conflicts. What's more, DRL agents are usually trained on the simulation environment before deployment, so another challenge is to develop an accurate but relatively simple simulator. Therefore, we propose a DRL algorithm for a central HVAC system to co-optimize energy consumption, thermal comfort, indoor CO2 level, and indoor PM2.5 level in an office building. To train the controller, we also developed a hybrid simulator that decoupled the complex system into multiple simulation models, which are calibrated separately using laboratory test data. The hybrid simulator combined the dynamics of the HVAC system, the building envelope, as well as moisture, CO2, and particulate matter transfer. Three control algorithms (rule-based, MPC, and DRL) are developed, and their performances are evaluated on the hybrid simulator environment with a realistic scenario (i.e., with stochastic noises). The test results showed that, the DRL controller can save 21.4 % of energy compared to a rule-based controller, and has improved thermal comfort, reduced indoor CO2 concentration. The MPC controller showed an 18.6 % energy saving compared to the DRL controller, mainly due to savings from comfort and indoor air quality boundary violations caused by unmeasured disturbances, and it also highlights computational challenges in real-time control due to non-linear optimization. Finally, we provide the practical considerations for designing and implementing the DRL and MPC controllers based on their respective pros and cons.

Guo, Fangzhou↗

DOE OSTI2023

Projection pursuit adaptation on polynomial chaos expansions

Here, the present work addresses the issue of accurate stochastic approximations in high-dimensional parametric space using tools from uncertainty quantification (UQ). The basis adaptation method and its accelerated algorithm in polynomial chaos expansions (PCE) were recently proposed to construct low-dimensional approximations adapted to specific quantities of interest (QoI). The present paper addresses one difficulty with these adaptations, namely their reliance on quadrature point sampling, which limits the reusability of potentially expensive samples. Projection pursuit (PP) is a statistical tool to find the “interesting” projections in high-dimensional data and thus bypass the curse-of-dimensionality. In the present work, we combine the fundamental ideas of basis adaptation and projection pursuit regression (PPR) to propose a novel method to simultaneously learn the optimal low-dimensional spaces and PCE representation from given data. While this projection pursuit adaptation (PPA) can be entirely data-driven, the constructed approximation exhibits mean-square convergence to the solution of an underlying governing equation and thus captures the supports and probability distributions associated with the physics constraints. The proposed approach is demonstrated on a borehole problem and a structural dynamics problem, demonstrating the versatility of the method and its ability to discover low-dimensional manifolds with high accuracy with limited data. In addition, the method can learn surrogate models for different quantities of interest while reusing the same data set.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2025

Evolution of intermittent filaments in the scrape-off layer of NSTX

Filamentary structures naturally arise from background turbulence in the scrape-off layer (SOL) of plasmas, leading to significant particle and heat transport that can degrade overall plasma confinement. This enhanced transport can contribute to unacceptably high heat loads on plasma-facing components. As such, understanding the physics of SOL plasma filaments is critical for predicting and mitigating their effects in future fusion devices. On the National Spherical Torus Experiment (NSTX), plasma filaments—commonly referred to as “blobs”—were investigated using the gas-puff imaging (GPI) diagnostic in the edge and SOL regions. The analysis involved identifying, segmenting, and tracking the characteristic contours of the blobs in each frame of the GPI video sequences. Their evolution was characterized through shape descriptors, velocity, and angular velocity derived from their contour coordinates. The results indicate that as the blob area increases, their shapes become more concave and less circular, suggesting reduced structural stability in larger blobs. This result aligns with previous theoretical results where it was shown that larger blobs are more susceptible to instabilities [Krasheninnikov et al., J. Plasma Phys. 74, 679–717 (2008) and D'Ippolito et al., Phys. Plasmas 18, 060501 (2011)]. A positive correlation was observed between radial velocity and radial position, suggesting radially outward acceleration of the filaments, potentially driven by decreasing viscous drag toward the far SOL. Interestingly, blobs in background SOL turbulence exhibited minimal spinning in contrast to filaments originating from edge localized modes, which show substantial rotation during their paths [Lampert et al., Phys. Plasmas 29, 102502 (2022)]. Statistical analysis of the solidity and total curvature shape descriptors, along with their temporal evolution, revealed relatively broad, near-Gaussian distributions. This suggests that blob morphology is strongly influenced by stochastic turbulent processes in the surrounding plasma environment. Blob parameters were also compared with bulk plasma and radial profile measurements. Notable trends were found between blob rotation and poloidal velocity with collisionality and line-integrated density. These findings contribute to a deeper understanding of blob dynamics and provide valuable insights for refining SOL turbulence models.

Covariance and correlation↗

DOE OSTI2020

From pools to flow: The PROMISE framework for new insights on soil carbon cycling in a changing world

Soils represent the largest terrestrial reservoir of organic carbon, and the balance between soil organic carbon (SOC) formation and loss will drive powerful carbon-climate feedbacks over the coming century. To date, efforts to predict SOC dynamics have rested on pool-based models, which assume classes of SOC with internally homogenous physicochemical properties. However, emerging evidence suggests that soil carbon turnover is not dominantly controlled by the chemistry of carbon inputs, but rather by restrictions on microbial access to organic matter in the spatially heterogeneous soil environment. The dynamic processes that control the physicochemical protection of carbon translate poorly to pool-based SOC models; as a result, we are challenged to mechanistically predict how environmental change will impact movement of carbon between soils and the atmosphere. In this study, we propose a novel conceptual framework to explore controls on belowground carbon cycling: Probabilistic Representation of Organic Matter Interactions within the Soil Environment (PROMISE). In contrast to traditional model frameworks, PROMISE does not attempt to define carbon pools united by common thermodynamic or functional attributes. Rather, the PROMISE concept considers how SOC cycling rates are governed by the stochastic processes that influence the proximity between microbial decomposers and organic matter, with emphasis on their physical location in the soil matrix. We illustrate the applications of this framework with a new biogeochemical simulation model that traces the fate of individual carbon atoms as they interact with their environment, undergoing biochemical transformations and moving through the soil pore space. We also discuss how the PROMISE framework reshapes dialogue around issues related to SOC management in a changing world. We intend the PROMISE framework to spur the development of new hypotheses, analytical tools, and model structures across disciplines that will illuminate mechanistic controls on the flow of carbon between plant, soil, and atmospheric pools.

54 ENVIRONMENTAL SCIENCES↗

DOE OSTI2025

Nonlinear Scaling of Water–Ion Interactions and Dynamics in Alkaline Solutions

Water-ion interactions govern many solvent properties critical to solution-phase chemistry and the behavior of liquid water. The water-ion interactions in alkaline conditions were probed using two-dimensional infrared spectroscopy (2D IR), small-angle x-ray scattering (SAXS), and nuclear magnetic resonance spectroscopy (NMR). Energy transfer between the donor molecule KSeCN, used as a 2D IR probe, and the acceptor molecule NaOD was used to track the average separation distance of ions in a D 2 O solution, while SAXS measurements showed effects in the bulk D 2 O solvent. Here, we observe consistent nonlinear scaling in the SeCN - and OD - average separation distance as a function of NaOD concentration while bulk solution D 2 O-to-D 2 O average separation distance remained highly linear. Ultrafast measurements of solution dynamics via 2D IR and polarization-selective pump-probe spectroscopy show consistent scaling in correlation times as a function of concentration. These results suggests that the SeCN - and OD - anions participate in a water-ion network that significantly reduces the degrees of freedom for the distribution of ions in solution below the standard stochastic for ion distribution.

2D IR Spectroscopy↗

DOE OSTI2021

Manifold learning for coarse-graining atomistic simulations: Application to amorphous solids

In this work, we introduce a generalized machine learning framework to probabilistically parameterize upper-scale models in the form of nonlinear PDEs consistent with a continuum theory, based on coarse-grained atomistic simulation data of mechanical deformation and flow processes. The proposed framework utilizes a hypothesized coarse-graining methodology with manifold learning and surrogate-based optimization techniques. Coarse-grained high-dimensional data describing quantities of interest of the multiscale models are projected onto a nonlinear manifold whose geometric and topological structure is exploited for measuring behavioral discrepancies in the form of manifold distances. A surrogate model is constructed using Gaussian process regression to identify a mapping between stochastic parameters and distances. Derivative-free optimization is employed to adaptively identify a unique set of parameters of the upper-scale model capable of rapidly reproducing the system's behavior while maintaining consistency with coarse-grained atomic-level simulations. The proposed method is applied to learn the parameters of the shear transformation zone (STZ) theory of plasticity that describes plastic deformation in amorphous solids as well as coarse-graining parameters needed to translate between atomistic and continuum representations. We show that the methodology is able to successfully link coarse-grained microscale simulations to macroscale observables and achieve a high-level of parity between the models across scales.

36 MATERIALS SCIENCE↗

DOE OSTI2023

The active CGCG 077-102 NED02 galaxy within the Abell 2063 galaxy cluster

Context.Within the framework of investigating the link between the central super massive black holes in the cores of galaxies and the galaxies themselves, we detected a variable X-ray source in the center of CGCG 077-102 NED02, which is a member of the CGCG 077-102 galaxy pair within the Abell 2063 cluster of galaxies. Aims.Our goal is to combine X-ray and optical data to demonstrate that this object harbors an active super massive black hole in its core, and to relate this to the dynamical status of the galaxy pair within the Abell 2063 cluster. Methods.We usedChandraandXMM-Newtonarchival data to derive the X-ray spectral shape and variability. We also obtained optical spectroscopy to detect the expected emission lines that are typically found in active galactic nuclei. Finally, we used public ZTF imaging data to investigate the optical variability. Results.There is no evidence of multiple X-ray sources or extended components within CGCG 077-102 NED02. Single X-ray spectral models fit the source well. We detect significant, nonrandom inter-observation 0.5–10 keV X-ray flux variabilities, for observations separated by ∼4 days for short-term variations and by up to ∼700 days for long-term variations. Optical spectroscopy points toward a passive galaxy for CGCG 077-102 NED01 and a Seyfert for CGCG 077-102 NED02. The classification of CGCG 077-102 NED02 is also consistent with its X-ray luminosity of over 10 42 erg s −1 . We do not detect short-term variability in the optical ZTF light curves. However, we find a significant long-term stochastic variability in theg-band that can be well described by the damped random walk model with a best-fit characteristic damping timescale ofτ DRW = 30 −12 +28 days. Finally, the CGCG 077-102 galaxy pair is deeply embedded within the Abell 2063 potential, with a long enough history within this massive structure to have been affected by the influence of this cluster for a long time. Conclusions.Our observations point toward a moderately massive black hole in the center of CGCG 077-102 NED02 of ∼10 6 M ⊙ . As compared to another similar pair in the literature, CGCG 077-102 NED02 is not heavily obscured, perhaps because of the surrounding intracluster medium ram-pressure stripping.

Astronomy & Astrophysics↗

DOE OSTI2024

Stochastic modal velocity field in rough-wall turbulence

Stochastically generated instantaneous velocity profiles are used to reproduce the outer region of rough-wall turbulent boundary layers in a range of Reynolds numbers extending from the wind tunnel to field conditions. Each profile consists in a sequence of steps, defined by the modal velocities and representing uniform momentum zones (UMZs), separated by velocity jumps representing the internal shear layers. Height-dependent UMZ is described by a minimal set of attributes: thickness, mid-height elevation, and streamwise (modal) and vertical velocities. These are informed by experimental observations and reproducing the statistical behaviour of rough-wall turbulence and attached eddy scaling, consistent with the corresponding experimental datasets. Sets of independently generated profiles are reorganized in the streamwise direction to form a spatially consistent modal velocity field, starting from any randomly selected profile. The operation allows one to stretch or compress the velocity field in space, increases the size of the domain and adjusts the size of the largest emerging structures to the Reynolds number of the simulated flow. By imposing the autocorrelation function of the modal velocity field to be anchored on the experimental measurements, we obtain a physically based spatial resolution, which is employed in the computation of the velocity spectrum, and second-order structure functions. The results reproduce the Kolmogorov inertial range extending from the UMZ and their attached-eddy vertical organization to the very-large-scale motions (VLSMs) introduced with the reordering process. The dynamic role of VLSM is confirmed in the –u'w' co-spectra and in their vertical derivative, representing a scale-dependent pressure gradient contribution.

42 ENGINEERING↗

DOE OSTI2021

Mechanisms of Dynamic Near-Wake Modulation of a Utility-Scale Wind Turbine

The current study uses large eddy simulations to investigate the transient response of a utility-scale wind turbine wake to dynamic changes in atmospheric and operational conditions, as observed in previous field-scale measurements. Most wind turbine wake investigations assume quasi-steady conditions, but real wind turbines operate in a highly stochastic atmosphere, and their operation (e.g. blade pitch, yaw angle) changes constantly in response. Furthermore, dynamic control strategies have been recently proposed to optimize wind farm power generation and longevity. Therefore, improved understanding of dynamic wake behaviours is essential. First, changes in blade pitch are investigated and the wake expansion response is found to display hysteresis as a result of flow inertia. The time scales of the wake response to different pitch rates are quantified. Next, changes in wind direction with different time scales are explored. Under short time scales, the wake deflection is in the opposite direction of that observed under quasi-steady conditions. Finally, yaw changes are implemented at different rates, and the maximum inverse wake deflection and time scale are quantified, showing a clear dependence on yaw rate. To gain further physical understanding of the mechanism behind the inverse wake deflection, the streamwise vorticity in different parts of the wake is quantified. The results of this study provide guidance for the design of advanced wake flow control algorithms. The lag in wake response observed for both blade pitch and yaw changes shows that proposed dynamic control strategies must implement turbine operational changes with a time scale of the order of the rotor time scale or slower.

flow-structure interactions↗

DOE OSTI2021

Quantifying Turbine-Level Risk to Golden Eagles Using a High-Fidelity Updraft Model and a Stochastic Behavioral Model

To minimize the effects of wind farms on Golden Eagle (Aquila chrysaetos) populations while enabling sustainable development of renewable energy resources, it is important to understand how eagles interact with atmospheric flows, terrain features, and anthropogenic structures. Models that predict migratory flight paths provide one tool that helps us grasp how the location of wind farms may influence interactions and impacts on migrating Golden Eagles. The current state-of-the-art in predicting migratory flight paths uses a deterministic fluid-flow analogy to predict eagle trajectory using only an orographic updraft potential computed from topographical features. This model does not take into account variables, such as thermal updrafts and time varying atmospheric conditions that are known to influence migratory behavior. In this work, we improve on the model with the objective of developing tools that advance our understanding of how atmospheric flows and terrain features affect migratory eagle behavior and their interactions with wind farms. Specifically, we 1) incorporate both orographic and thermal updraft information in simulating eagle flight paths; 2) incorporate stochasticity into eagle travel patterns to better capture the influence of exogenous factors on, and the inherent stochasticity of eagle behavior; 3) consider spatio-temporal atmospheric data at wind-farm-scale when computing updraft potential; and 4) account for how atmospheric conditions and the direction of migration change seasonally and how these changes affect eagle migratory flight behavior. We tested the model using a 50km by 50km region with 50 m resolution in the western United States. We simulated 900 independent, probabilistic eagle tracks during southerly and northerly migration, assuming eagles solely rely on orographic updrafts. The preliminary results indicate that the inclusion of finer resolution atmospheric data allows for the inclusion of realistic conditions that an eagle experiences. The stochasticity in eagle tracks provides a platform to include uncertainty in eagle decision making and help produce robust eagle presence maps. We will deploy updraft and downdraft velocities computed using a high-fidelity, wind farm scale, computational fluid dynamics solver under development at National Renewable Energy Laboratory. This work is a first step in the development of a predictive and generalizable eagle behavior model at the wind farm scale that does not rely on empirical data collection. Although the current model is intended for migratory eagles, we will extend and refine this model to inform the development of additional behavioral modes, including resident eagle behavior. This modeling approach improves our ability to understand eagle use of the landscape at a fine scale, and it is our hope that this work will ultimately help advance strategies that minimize the impact of wind development on Golden Eagle populations.

49 EE - Wind and Water Power Program - Wind (EE-4W↗

DOE OSTI2023

Tuning nonequilibrium phase transitions with inertia

In striking contrast to equilibrium systems, inertia can profoundly alter the structure of active systems. Here, we demonstrate that driven systems can exhibit effective equilibrium-like states with increasing particle inertia, despite rigorously violating the fluctuation–dissipation theorem. Increasing inertia progressively eliminates motility-induced phase separation and restores equilibrium crystallization for active Brownian spheres. This effect appears to be general for a wide class of active systems, including those driven by deterministic time-dependent external fields, whose nonequilibrium patterns ultimately disappear with increasing inertia. The path to this effective equilibrium limit can be complex, with finite inertia sometimes acting to accentuate nonequilibrium transitions. The restoration of near equilibrium statistics can be understood through the conversion of active momentum sources to passive-like stresses. Unlike truly equilibrium systems, the effective temperature is now density dependent, the only remnant of the nonequilibrium dynamics. Further, this density-dependent temperature can in principle introduce departures from equilibrium expectations, particularly in response to strong gradients. Our results provide additional insight into the effective temperature ansatz while revealing a mechanism to tune nonequilibrium phase transitions.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2022

Scalable training of graph convolutional neural networks for fast and accurate predictions of HOMO-LUMO gap in molecules

Abstract Graph Convolutional Neural Network (GCNN) is a popular class of deep learning (DL) models in material science to predict material properties from the graph representation of molecular structures. Training an accurate and comprehensive GCNN surrogate for molecular design requires large-scale graph datasets and is usually a time-consuming process. Recent advances in GPUs and distributed computing open a path to reduce the computational cost for GCNN training effectively. However, efficient utilization of high performance computing (HPC) resources for training requires simultaneously optimizing large-scale data management and scalable stochastic batched optimization techniques. In this work, we focus on building GCNN models on HPC systems to predict material properties of millions of molecules. We use HydraGNN, our in-house library for large-scale GCNN training, leveraging distributed data parallelism in PyTorch. We use ADIOS, a high-performance data management framework for efficient storage and reading of large molecular graph data. We perform parallel training on two open-source large-scale graph datasets to build a GCNN predictor for an important quantum property known as the HOMO-LUMO gap. We measure the scalability, accuracy, and convergence of our approach on two DOE supercomputers: the Summit supercomputer at the Oak Ridge Leadership Computing Facility (OLCF) and the Perlmutter system at the National Energy Research Scientific Computing Center (NERSC). We present our experimental results with HydraGNN showing (i) reduction of data loading time up to 4.2 times compared with a conventional method and (ii) linear scaling performance for training up to 1024 GPUs on both Summit and Perlmutter.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗