Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Monte Carlo Sampling”

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 991 records · Page 55

Bayesian Inference with Latent Hamiltonian Neural Networks (L-HNNs)

When sampling for Bayesian inference, one popular approach is to use Hamiltonian Monte Carlo (HMC) and the No-U-Turn Sampler (NUTS). However, HMC and NUTS can require numerous numerical gradients of the target density and can prove slow in practice. We propose Hamiltonian neural networks (HNNs) with HMC and NUTS for solving Bayesian inference problems [1, 2]. Once trained, HNNs do not require gradients of the target density while sampling. Moreover, they satisfy important properties such as perfect time reversibility and Hamiltonian conservation, making them well suited for use within HMC and NUTS because stationarity can be shown. We also propose an HNN extension called latent HNNs (L-HNNs), which predict latent variable outputs. Compared to HNNs, L-HNNs offer improved expressivity and a reduction in integration errors. Finally, we propose employing L-HNNs in NUTS with an online error monitoring scheme to prevent degeneracy of the sampling in regions of low probability density. We demonstrate L-HNNs in NUTS with online error monitoring by using several example cases involving complex, heavy-tailed, and high local curvature probability densities. Overall, L-HNNs in NUTS with online error monitoring satisfactorily inferred these probability densities. Compared to traditional NUTS, L-HNNs in NUTS with online error monitoring improved the effective sample size (ESS) per gradient by an order of magnitude.

97 MATHEMATICS AND COMPUTING↗

Advanced manufacturing of 3D custom boron-carbide collimators designed for complex environments for neutron scattering

Scattered-beam collimation is a very useful method to reduce unwanted backgrounds and to boost the desired sample signal instead. This approach is of particular interest for samples contained within a complex environment that gives rise to much unwanted parasitic scatter. As neutron scattering instrument and techniques advances, small samples are becoming of more and more interest, which necessitates optimized collimation. Here, in this work, we describe a concept for the design and fabrication of advanced scattered-beam collimation 3D printed from B 4 C specifically tailored for samples contained within a complex environment. This concept is demonstrated through the use of a diamond anvil cell for high pressure experimentation, a technique that very typically requires small samples. The collimators here are designed through a modeling procedure via Monte Carlo neutron ray tracing that encompasses the entire system: the instrument, the complex environment and the collimator. Since the first approach of simply scaling up of the print-size was not successful, a novel concept of a multi-part alternate-blade collimator was developed. This approach addresses printing constraints but gives greater flexibility in design. Its performance is computationally compared against an unprintable progressively tighter blade collimator to assess the effect of alternating blades. No strong difference was observed. Its performance was validated through experimentation at the Spallation Neutron Source. The results emphasize the critical importance of ultra-high precision alignment while showing good overall agreement between simulation and experiment and underscore the feasibility of the method and its real-world application.

46 INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AND ↗

Improving the sampling efficiency of the Grand Canonical Simulated Quenching approach

Most common atomistic simulation techniques, like molecular dynamics or Metropolis Monte Carlo, operate under a constant interatomic Hamiltonian with a fixed number of atoms. Internal (atom positions or velocities) or external (simulation cell size or geometry) variables are then evolved dynamically or stochastically to yield sampling in different ensembles, such as microcanonical (NVE), canonical (NVT), isothermal-isobaric (NPT), etc. Averages are then taken to compute relevant physical properties. At least two limitations of these standard approaches can seriously hamper their application to many important systems: (1) they do not allow for the exchange of particles with a reservoir, and (2) the sampling efficiency is insufficient to allow the obtention of converged results because of the very long intrinsic timescales associated with these quantities. To fix ideas, one might want to identify low (free) energy configurations of grain boundaries (GB). In reality, grain boundaries are in contact the grains which act as reservoirs of defects (e.g., vacancies and interstitials). Since the GB can exchange particles with its environment, the most stable configuration cannot provably be found by sampling from NVE or NVT ensembles alone: one needs to allow the number of atoms in the sample to fluctuate. The first limitation can be circumvented by working in the grand canonical ensemble (TV ) or its derivatives (such as the semi-grand-canonical ensemble useful for the study of substitutional alloys). Monte Carlo methods have been the first to adapt to this kind of system where the number of atoms is allowed to fluctuate. Many of these methods are based on the Widom insertion method [Widom63] where the chemical potential of a given chemical species can be inferred from the potential energy changes upon random insertion of a new particle within the simulation cell. Other techniques, such as the Gibbs ensemble Monte Carlo [Panagiotopoulos87] where exchanges of particles are attempted to equilibrate the chemical potential between two cells and hence allow for the calculation of coexistence curves, exploit the same idea: particle insertion (or exchange) is attempted and accepted with a Metropolis-like rule that depends exponentially on the energy change upon insertion. A well known limitation of this kind of approach is that the probability of accepting such a move decreases extremely rapidly with increasing density, due to the extremely large short-range repulsion between atoms. In response to these difficulties it became apparent that a solution to the problem might be to avoid abrupt insertions but instead to proceed gradually, so as to allow the system to react and make way for the incoming particle. In this view of things, the 'occupation' of a certain atomic site can be viewed as a continuous variable, ranging between 0 and 1, representing 'how much' of the particle is present at any given time. These ideas proved ideal in Molecular Dynamics (MD) settings because equations of motions for these occupation variables can sometimes be obtained. For example, in the case of Grand Canonical Molecular Dynamics [Cagin91], one special particle is allowed to have a fractional occupation. This can lead to either its destruction (occupation = 0) or its complex creation (occupation = 1) so as to enforce a given chemical potential. These approaches proved useful, but mostly in the liquid state where the probability of successfully inserting a new particle is sufficiently high. At higher densities, convergence proved to be hampered by very inefficient sampling. In this work, we explore the use of a related MD-based grand canonical technique, the Grand Canonical Simulated Quenching (GCSQ) of Phillpot and Rickman [Phillpot92,Phillpot94], and explore its application to the grand canonical sampling of solid state systems. We show that, in conjunction with advanced sampling techniques, GCSQ can be a useful tool to sample conformations of complex systems, such as GBs, and assist in the identification of their most stable states and/or most likely defective states.

36 MATERIALS SCIENCE↗

Constraining the evolution of the CMB temperature with SZ measurements from Planck data

The CMB temperature-redshift relation, T{sub CMB}(z)=T{sub 0}(1+z), is a key prediction of the standard cosmology but is violated in many non-standard models. Constraining possible deviations from this law is an effective way to test the ΛCDM paradigm and to search for hints of new physics. We have determined T{sub CMB}(z), with a precision up to 3%, for a subsample (103 clusters) of the Planck SZ cluster catalog, at redshifts in the range 0.01–0.94, using measurements of the spectrum of the Sunyaev-Zel'dovich (SZ) effect obtained from Planck temperature maps at frequencies from 70 to 353 GHz. The method adopted to provide individual determinations of T{sub CMB}(z) at cluster redshift relies on the use of SZ intensity change, Δ I{sub SZ}(ν) at different frequencies and on a Monte Carlo Markov chain approach. By applying this method to the sample of 103 clusters, we limit possible deviations of the form T{sub CMB}(z)=T{sub 0}(1+z){sup 1−β} to be β= 0.012 ± 0.016, at 1σ uncertainty, consistent with the prediction of the standard model. Combining these measurements with previously published results, we get β=0.013±0.011.

79 ASTRONOMY AND ASTROPHYSICS↗

Accelerating Hamiltonian Monte Carlo for Bayesian inference in neural networks and neural operators

Hamiltonian Monte Carlo (HMC) is a powerful and accurate method to sample from the posterior distribution in Bayesian inference. However, HMC techniques are computationally demanding for Bayesian neural networks due to the high dimensionality of the network’s parameter space and the non-convexity of their posterior distributions. Therefore, various approximation techniques, such as variational inference (VI) or stochastic gradient MCMC, are often employed to infer the posterior distribution of the network parameters. Such approximations introduce inaccuracies in the inferred distributions, resulting in unreliable uncertainty estimates. In this work, we propose a hybrid approach that combines inexpensive VI and accurate HMC methods to efficiently and accurately quantify uncertainties in neural networks and neural operators. The proposed approach leverages an initial VI training on the full network. We examine the influence of individual parameters on the prediction uncertainty, which shows that a large proportion of the parameters do not contribute substantially to uncertainty in the network predictions. This information is then used to significantly reduce the dimension of the parameter space, and HMC is performed only for the subset of network parameters that strongly influence prediction uncertainties. This yields a framework for accelerating the full batch HMC for posterior inference in neural networks. We demonstrate the efficiency and accuracy of the proposed framework on deep neural networks and operator networks, showing that inference can be performed for large networks with tens to hundreds of thousands of parameters. Finally, we show that this method can effectively learn surrogates for complex physical systems by modeling the operator that maps from upstream conditions to wall-pressure data on a cone in hypersonic flow.

Bayesian inference↗

Accelerating Ab Initio Simulation via Nested Monte Carlo and Machine Learned Reference Potentials

As a corollary of the rapid advances in computing, ab initio simulation is playing an increasingly important role in modeling materials at the atomic scale. Two strategies are possible, ab initio Monte Carlo (AIMC) and molecular dynamics (AIMD) simulation. The former benefits from exact sampling from the correct thermodynamic distribution, while the latter is typically more efficient with its collective all-atom coordinate updates. In this study, using a relatively simple test model comprised of helium and argon, we show that AIMC can be brought up to, and even above, the performance levels of AIMD via a hybrid nested sampling/machine learning (ML) strategy. Here, ML provides an accurate classical reference potential (up to three-body explicit interactions) that can pilot long collective Monte Carlo moves that are accepted or rejected in toto à la nested Monte Carlo (NMC); this is in contrast to the single move nature of a naive implementation. Our proposed method only requires a small up front expense from evaluating the ab initio energies and forces of (100) random configurations for training. Importantly, our method does not totally rely on the trained, assuredly imperfect, interaction. We show that high performance and exact sampling at the desired level of theory can be realized even when the trained interaction has appreciable differences from the ab initio potential. Remarkably, at the highest levels of performance realized via our approach, a pair of statistically uncorrelated atomic configurations can be generated with (1) ab initio calculations.

74 ATOMIC AND MOLECULAR PHYSICS↗

Height control of self-assembled quantum dots by strain engineering during capping

Strain engineering during the capping of III-V quantum dots has been explored as a means to control the height of strained self-assembled quantum dots. Results of Kinetic Monte Carlo simulations are confronted with cross-sectional Scanning Tunnel Microscopy (STM) measurements performed on InAs quantum dots grown by molecular beam epitaxy. We studied InAs quantum dots that are capped by In{sub x}Ga{sub (1−x)}As layers of different indium compositions. Both from our realistic 3D kinetic Monte Carlo simulations and the X-STM measurements on real samples, a trend in the height of the capped quantum dot is found as a function of the lattice mismatch between the quantum dot material and the capping layer. Results obtained on additional material combinations show a generic role of the elastic energy in the control of the quantum dot morphology by strain engineering during capping.

75 CONDENSED MATTER PHYSICS, SUPERCONDUCTIVITY AND↗

A whole-path importance-sampling scheme for Feynman path integral calculations of absolute partition functions and free energies

Using Feynman path integrals, a molecular partition function can be written as a double integral with the inner integral involving all closed paths centered at a given molecular configuration, and the outer integral involving all possible molecular configurations. In previous work employing Monte Carlo methods to evaluate such partition functions, we presented schemes for importance sampling and stratification in the molecular configurations that constitute the path centroids, but we relied on free-particle paths for sampling the path integrals. At low temperatures, the path sampling is expensive because the paths can travel far from the centroid configuration. In this paper, we now present a scheme for importance sampling of whole Feynman paths based on harmonic information from an instantaneous normal mode calculation at the centroid configuration, which we refer to as harmonically guided whole-path importance sampling (WPIS). We obtain paths conforming to our chosen importance function by rejection sampling from a distribution of free-particle paths. Sample calculations on CH 4 demonstrate that at a temperature of 200 K, about 99.9% of the free-particle paths can be rejected without integration, and at 300 K, about 98% can be rejected. Finally, we also show that it is typically possible to reduce the overhead associated with the WPIS scheme by sampling the paths using a significantly lower-order path discretization than that which is needed to converge the partition function.

74 ATOMIC AND MOLECULAR PHYSICS↗

Solid-propellant rocket motor ballistic performance variation analyses

Results are presented of research aimed at improving the assessment of off-nominal internal ballistic performance including tailoff and thrust imbalance of two large solid-rocket motors (SRMs) firing in parallel. Previous analyses using the Monte Carlo technique were refined to permit evaluation of the effects of radial and circumferential propellant temperature gradients. Sample evaluations of the effect of the temperature gradients are presented. A separate theoretical investigation of the effect of strain rate on the burning rate of propellant indicates that the thermoelastic coupling may cause substantial variations in burning rate during highly transient operating conditions. The Monte Carlo approach was also modified to permit the effects on performance of variation in the characteristics between lots of propellants and other materials to be evaluated. This permits the variabilities for the total SRM population to be determined. A sample case shows, however, that the effect of these between-lot variations on thrust imbalances within pairs of SRMs is minor in compariosn to the effect of the within-lot variations. The revised Monte Carlo and design analysis computer programs along with instructions including format requirements for preparation of input data and illustrative examples are presented.

Sforzini, R. H.↗

Reconstruction and Selection of Neutrino Interactions in MicroBooNE using Deep Convolutional Neural Networks

In this document, we describe a new reconstruction workflow developed for the MicroBooNE experiment. It features the use of Deep Convolutional Neural Networks trained to recognize key structures within the data sufficient for the 3D reconstruction of neutrino interactions within the detector. As a test of the reconstruction utility, the products of the reconstruction workflow are used to select inclusive charged-current (CC) $\nu_e$ and $\nu_\mu$ interactions in both simulated and real MicroBooNE data. In simulation, our $\nu_e$ and $\nu_\mu$ selections achieve an efficiency of 57% and 68\%, respectively, with a purity of 91% and 96%, respectively. We find that these selections are competitive with the inclusive selections used for the most recent MicroBooNE LEE searches. In particular, the CC-$\nu_e$ inclusive selection efficiency improves by over 20% while also improving sample purity. As a first step in quantifying potential bias, the data and Monte Carlo expectati ons are compared for both selections using the MicroBooNE open data. Within statistical and systematic uncertainties, both the electron and muon CC-inclusive event samples agree. A comparison of the real data events chosen by our work and another reconstruction framework shows that the two analyses each identify a sizeable fraction of events the other does not. This suggests that future analyses integrating the strengths of each could lead to combined gains. This work demonstrates, for the first time on real LArTPC data, state-of-the-art neutrino interaction reconstruction centered around deep learning algorithms.

43 PARTICLE ACCELERATORS↗

Analyzing indirect secondary electron contrast of unstained bacteriophage T4 based on SEM images and Monte Carlo simulations

The indirect secondary electron contrast (ISEC) condition of the scanning electron microscopy (SEM) produces high contrast detection with minimal damage of unstained biological samples mounted under a thin carbon film. The high contrast image is created by a secondary electron signal produced under the carbon film by a low acceleration voltage. Here, we show that ISEC condition is clearly able to detect unstained bacteriophage T4 under a thin carbon film (10-15 nm) by using high-resolution field emission (FE) SEM. The results show that FE-SEM provides higher resolution than thermionic emission SEM. Furthermore, we investigated the scattered electron area within the carbon film under ISEC conditions using Monte Carlo simulation. The simulations indicated that the image resolution difference is related to the scattering width in the carbon film and the electron beam spot size. Using ISEC conditions on unstained virus samples would produce low electronic damage, because the electron beam does not directly irradiate the sample. In addition to the routine analysis, this method can be utilized for structural analysis of various biological samples like viruses, bacteria, and protein complexes.

60 APPLIED LIFE SCIENCES↗

Monte Carlo simulation of launchsite winds at Kennedy Space Center

This paper develops and validates an easily implemented model for simulating random horizontal wind profiles over the Kennedy Space Center (KSC) at Cape Canaveral, Florida. The model is intended for use in Monte Carlo launch vehicle simulations of the type employed in mission planning, where the large number of profiles needed for statistical fidelity of such simulation experiments makes the use of actual wind measurements impractical. The model is based on measurements made at KSC and represents vertical correlations by a decaying exponential model which is parameterized via least-squares parameter optimization against the sample data. The validity of the model is evaluated by comparing two Monte Carlo simulations of an asymmetric, heavy-lift launch vehicle. In the first simulation, the measured wind profiles are used, while in the second, the wind profiles are generated using the stochastic model. The simulations indicate that the use of either the measured or simulated wind field results in similar launch vehicle performance.

Queen, Eric M.↗

Monte Carlo Shielding Analysis Capabilities with MAVRIC

Monte Carlo shielding analysis capabilities in SCALE 6 are centered on the CADIS methodology Consistent Adjoint Driven Importance Sampling. CADIS is used to create an importance map for space/energy weight windows as well as a biased source distribution. New to SCALE 6 are the Monaco functional module, a multi-group fixed-source Monte Carlo transport code, and the MAVRIC sequence (Monaco with Automated Variance Reduction Using Importance Calculations). MAVRIC uses the Denovo code (also new to SCALE 6) to compute coarse-mesh discrete ordinates solutions which are used by CADIS to form an importance map and biased source distribution for the Monaco Monte Carlo code. MAVRIC allows the user to optimize the Monaco calculation for a specify tally using the CADIS method with little extra input compared to a standard Monte Carlo calculation. When computing several tallies at once or a mesh tally over a large volume of space, an extension of the CADIS method called FW-CADIS can be used to help the Monte Carlo simulation spread particles over phase space to get more uniform relative uncertainties.

99 GENERAL AND MISCELLANEOUS↗

Implementation of the direct S ( α , β ) method in the KENO Monte Carlo code

The Monte Carlo code KENO contains thermal scattering data for a wide variety of thermal moderators. These data are processed from Evaluated Nuclear Data Files (ENDF) by AMPX and stored as double differential probability distribution functions. The method examined in this study uses S(α,β) probability distribution functions derived from the ENDF data files directly instead of being converted to double differential cross sections. This allows the size of the cross section data on the disk to be reduced substantially amount. KENO has also been updated to allow interpolation in temperature on these data so that problems can be run at any temperature. Results are shown for several simplified problems for a variety of moderators. In addition, benchmark models based on the KRITZ reactor in Sweden were run, and the results are compared with the previous versions of KENO without the direct S(α,β) method. Results from the direct S(α,β) method compare favorably with the original results obtained using the double differential cross sections. Finally, sampling the data increases the run-time of the Monte Carlo calculation, but memory usage is decreased substantially.

73 NUCLEAR PHYSICS AND RADIATION PHYSICS↗

Efficiency in nonequilibrium molecular dynamics Monte Carlo simulations

Hybrid algorithms combining nonequilibrium molecular dynamics and Monte Carlo (neMD/MC) offer a powerful avenue for improving the sampling efficiency of computer simulations of complex systems. These neMD/MC algorithms are also increasingly finding use in applications where conventional approaches are impractical, such as constant-pH simulations with explicit solvent. However, selecting an optimal nonequilibrium protocol for maximum efficiency often represents a non-trivial challenge. This work evaluates the efficiency of a broad class of neMD/MC algorithms and protocols within the theoretical framework of linear response theory. The approximations are validated against constant pH-MD simulations and shown to provide accurate predictions of neMD/MC performance. An assessment of a large set of protocols confirms (both theoretically and empirically) that a linear work protocol gives the best neMD/MC performance. Lastly, a well-defined criterion for optimizing the time parameters of the protocol is proposed and demonstrated with an adaptive algorithm that improves the performance on-the-fly with minimal cost.

97 MATHEMATICS AND COMPUTING↗

Bayesian Treed Multivariate Gaussian Process with Adaptive Design: Application to a Carbon Capture Unit

Computer experiments (numerical simulations) are widely used in scientific research to study and predict the behavior of complex systems, which usually have responses consisting of a set of distinct outputs. The computational cost of the simulations at high resolution are often expensive and become impractical for parametric studies at different input values. To overcome these difficulties we develop a Bayesian treed multivariate Gaussian process (BTMGP) as an extension of the Bayesian treed Gaussian process (BTGP) in order to model and evaluate a multivariate process. A suitable choice of covariance function and the prior distributions facilitates the different Markov chain Monte Carlo (MCMC) movements. We utilize this model to sequentially sample the input space for the most informative values, taking into account model uncertainty and expertise gained. A simulation study demonstrates the use of the proposed method and compares it with alternative approaches. We apply the sequential sampling technique and BTMGP to model the multiphase flow in a full scale regenerator of a carbon capture unit. The application presented in this paper is an important tool for research into carbon dioxide emissions from thermal power plants.

Konomi, Bledar A.↗

Basic parameter estimation of binary neutron star systems by the advanced LIGO/Vigro network

Within the next five years, it is expected that the Advanced LIGO/Virgo network will have reached a sensitivity sufficient to enable the routine detection of gravitational waves. Beyond the initial detection, the scientific promise of these instruments relies on the effectiveness of our physical parameter estimation capabilities. A major part of this effort has been toward the detection and characterization of gravitational waves from compact binary coalescence, e.g., the coalescence of binary neutron stars. While several previous studies have investigated the accuracy of parameter estimation with advanced detectors, the majority have relied on approximation techniques such as the Fisher Matrix which are insensitive to the non-Gaussian nature of the gravitational wave posterior distribution function. Here we report average statistical uncertainties that will be achievable for strong detection candidates (S/N = 20) over a comprehensive sample of source parameters. We use the Markov Chain Monte Carlo based parameter estimation software developed by the LIGO/Virgo Collaboration with the goal of updating the previously quoted Fisher Matrix bounds. We find the recovery of the individual masses to be fractionally within 9% (15%) at the 68% (95%) credible intervals for equal-mass systems, and within 1.9% (3.7%) for unequal-mass systems. We also find that the Advanced LIGO/Virgo network will constrain the locations of binary neutron star mergers to a median uncertainty of 5.1 deg{sup 2} (13.5 deg{sup 2}) on the sky. This region is improved to 2.3 deg{sup 2} (6 deg{sup 2}) with the addition of the proposed LIGO India detector to the network. We also report the average uncertainties on the luminosity distances and orbital inclinations of strong detections that can be achieved by different network configurations.

79 ASTRONOMY AND ASTROPHYSICS↗