Search NASA⌕ Search

SEARCH · Search NASA

Results for “Kernel methods”

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 1,189 records · Page 66

Simulation and Analysis on Reactor Pressure Vessel (RPV) subjected to Pressurized Thermal Shock (PTS) under SBLOCA scenario by using Cardinal to support the fracture mechanics analyses

The structural components that comprise nuclear reactors and their supporting structures are subjected to harsh operating environments that can challenge their integrity, especially after exposure for extended durations or under accident condition. As one of the most significant components of a Reactor, the Reactor Pressure Vessel (RPV) is exposed to an aggressive environment during the operation time (e.g. more than 40 years). Ageing degradation mechanisms (e.g. thermos-fatigue) could grow initial defects up to a critical size, increasing the susceptibility to failure in the RPV. The conventional methods are mostly based on simple crack and structure geometries. Very limited studies consider the real conditions of the RPV subjected to a thermal shock due to a Loss of Coolant Accident (LOCA). During a LOCA event, the most severe conditions take place when the emergency core cooling (ECC) water is injected inside the cold legs filled initially with hotter water and/or steam. The rapid cooling of the down-comer and the internal RPV surface followed probably by re-pressurization of the RPV causes large temperature gradients and variation of pressure which induces thermal-mechanical stresses. In order to develop the model for integrity assessment of a reactor pressure vessel (RPV) subjected to pressurized thermal shock (PTS), a multi-physics simulation, which includes the thermo-hydraulic, thermo-mechanical and fracture mechanics analyses is necessary. The multi-physics simulations are performed using Cardinal, a wrapping of the GPU-oriented spectral element Computational Fluid Dynamics (CFD) code NekRS and other multi-physics sub-modules within the MOOSE framework. Cardinal now fully supports MOOSE stochastic perturbations of NekRS models with varying boundary conditions, initial conditions, material properties, and any other quantity which is defined by a kernel (such as coefficients in a momentum source model). The implementation is designed in a flexible manner so that scalar values are sent from MOOSE into a user scratch space in NekRS, which can then be applied for any purpose within the NekRS case files (both on the host and device). When modeling PTS, several factors can impact the results significantly. In this report, the impacts of the geometry of the model, Reynolds number and buoyancy effect are investigated. Two geometry, i.e., a simplified model and a realistic RPV model, with both laminar and turbulent flow condition are adopted for the PTS simulation with and without buoyancy effect. The purpose of the investigation is to understand the impact of these factors on the prediction of temperature history of RPV. The accurate prediction on the temperature evolution, which will be exported to Grizzly code for further analyses on the progression of aging mechanisms and their effects on the integrity of RPV structures, is very crucial. Based on the understanding of these factors, a more sophisticated model is built to analysis the PTS under SBLOCA scenario. A literature survey is conducted to pick the SBLOCA scenario for the multi-physics simulation. The analysis helps to explain the form and the transformation of the cold plum when the ECC is activated under SBLOCA. This model can be can be applied to study the PTS effect for different RPV configurations. The results can help to assess structural component degradation for advanced reactors.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Co-optimized machine-learned manifold models for large eddy simulation of turbulent combustion

Many modeling approaches in large eddy simulation (LES) of turbulent combustion employ a projection of the thermochemical state onto a low-dimensional manifold within state space to reduce the number of transported variables and hence computational cost. Flamelet-generated manifolds (FGM) is an example of a well-established, physics-based approach, but increasingly, principal component analysis (PCA) is being used as a data-driven method for generating manifold models. For both approaches, the nonlinear relationship between the location on the predefined manifold and the outputs of interest, such as reaction rates, can be tabulated or encoded in a neural network. This work proposes a new approach for manifold modeling that extends these existing approaches. A modified neural network structure simultaneously encodes the definition of the manifold variables, the nonlinear mapping, and the subfilter closure for LES. This allows all three of these aspects of the model to be co-optimized, generating a model from any source of combustion thermochemical state data. The manifold parameterizing variables are constrained to be linear combinations of species, as in FGM and PCA-based models, to aid in interpretability and implementation. For LES, subfilter variances of the manifold variables are also included as inputs. Two types of a priori analysis are performed to evaluate the new approach. In the first, the model is trained on data from one-dimensional premixed flames. In this case, the approach recovers the behavior of flamelet-based manifold approaches, and in fact slightly improves performance by identifying an optimized progress variable. The approach is also applied to data from direct numerical simulations of spherical ignition kernels in isotropic turbulence. For any specified manifold dimensionality, the new approach provides substantially lower prediction errors than a PCA-based model developed from the same data set. Additionally, the LES formulation of the new approach can provide accurate predictions for filtered reaction rates across a variety of filter widths.

97 MATHEMATICS AND COMPUTING↗

Dosimetric characteristics of the novel 2D ionization chamber array OCTAVIUS Detector 1500

Purpose: The dosimetric properties of the OCTAVIUS Detector 1500 (OD1500) ionization chamber array (PTW-Freiburg, Freiburg, Germany) have been investigated. A comparative study was carried out with the OCTAVIUS Detector 729 and OCTAVIUS Detector 1000 SRS arrays. Methods: The OD1500 array is an air vented ionization chamber array with 1405 detectors in a 27 × 27 cm{sup 2} measurement area arranged in a checkerboard pattern with a chamber-to-chamber distance of 10 mm in each row. A sampling step width of 5 mm can be achieved by merging two measurements shifted by 5 mm, thus fulfilling the Nyquist theorem for intensity modulated dose distributions. The stability, linearity, and dose per pulse dependence were investigated using a Semiflex 31013 chamber (PTW-Freiburg, Freiburg, Germany) as a reference detector. The effective depth of measurement was determined by measuring TPR curves with the array and a Roos chamber type 31004 (PTW-Freiburg, Freiburg, Germany). Comparative output factor measurements were performed with the array, the Semiflex 31010 ionization chamber and the Diode 60012 (both PTW-Freiburg, Freiburg, Germany). The energy dependence of the OD1500 was measured by comparing the array’s readings to those of a Semiflex 31010 ionization chamber for varying mean photon energies at the depth of measurement, applying to the Semiflex chamber readings the correction factor k{sub NR} for nonreference conditions. The Gaussian lateral dose response function of a single array detector was determined by searching the convolution kernel suitable to convert the slit beam profiles measured with a Diode 60012 into those measured with the array’s central chamber. An intensity modulated dose distribution measured with the array was verified by comparing a OD1500 measurement to TPS calculations and film measurements. Results: The stability and interchamber sensitivity variation of the OD1500 array were within ±0.2% and ±0.58%, respectively. Dose linearity was within 1% over the range from 5 to 1000 MU. The effective point of measurement of the OD1500 for dose measurements in RW3 phantoms was determined to be (8.7 ± 0.2) mm below its front surface. Output factors showed deviations below 1% for field sizes exceeding 4 × 4 cm{sup 2}. The dose per pulse dependence was smaller than 0.4% for doses per pulse from 0.2 to 1 mGy. The energy dependence of the array did not exceed ±0.9%. The parameter σ of the Gaussian lateral dose response function was determined as σ{sub 6MV} = (2.07 ± 0.02) mm for 6 MV and σ{sub 15MV} = (2.09 ± 0.02) mm for 15 MV. An IMRT verification showed passing rates well above 90% for a local 3 mm/3% criterion. Conclusions: The OD1500 array’s dosimetric properties showed the applicability of the array for clinical dosimetry with the possibility to increase the spatial sampling frequency and the coverage of a dose distribution with the sensitive areas of ionization chambers by merging two measurements.

07 ISOTOPE AND RADIATION SOURCES↗

Elastic full-waveform inversion and parameterization analysis applied to walk-away vertical seismic profile data for unconventional (heavy oil) reservoir characterization

We report seismic full-waveform inversion (FWI) methods hold strong potential to recover multiple subsurface elastic properties for hydrocarbon reservoir characterization. Simultaneously updating multiple physical parameters introduces the problem of interparameter tradeoff, arising from the covariance between different physical parameters, which increases nonlinearity and uncertainty of multiparameter FWI. The coupling effects of different physical parameters are significantly influenced by model parameterization and acquisition arrangement. An appropriate choice of model parameterization is critical to successful field data applications of multiparameter FWI. The objective of this paper is to examine the performance of various model parameterizations in isotropic-elastic FWI with walk-away vertical seismic profile (W-VSP) dataset for unconventional heavy oil reservoir characterization. Six model parameterizations are considered: velocity-density (α, β and ρ'), modulus-density (κ, μ and ρ), Lamé-density (λ, μ' and ρ'''), impedance-density (IP, IS and ρ''), velocity-impedance-I (α', β' and I' P ), and velocity-impedance-II (α'', β'' and I'S). We begin analyzing the interparameter tradeoff by making use of scattering radiation patterns, which is a common strategy for qualitative parameter resolution analysis. In this paper, we discuss the advantages and limitations of the scattering radiation patterns and recommend that interparameter tradeoffs be evaluated using interparameter contamination kernels, which provide quantitative, second-order measurements of the interparameter contaminations and can be constructed efficiently with an adjoint-state approach. Synthetic W-VSP isotropic-elastic FWI experiments in the time domain verify our conclusions about interparameter tradeoffs for various model parameterizations. Density profiles are most strongly influenced by the interparameter contaminations; depending on model parameterization, the inverted density profile can be over-estimated, under-estimated or spatially distorted. Among the six cases, only the velocity-density parameterization provides stable and informative density features not included in the starting model. Field data applications of multicomponent W-VSP isotropic-elastic FWI in the time domain were also carried out. Finally, the heavy oil reservoir target zone, characterized by low α-to-β ratios and low Poisson’s ratios, can be identified clearly with the inverted isotropic-elastic parameters.

58 GEOSCIENCES↗

Microscopic constraints for the equation of state and structure of neutron stars: A Bayesian model mixing framework

Bayesian model mixing (BMM) is a statistical technique that can combine constraints from different regions of an input space in a principled way. Here we extend our BMM framework for the equation of state (EOS) of strongly interacting matter from symmetric nuclear matter to asymmetric matter, specifically focusing on zero-temperature, charge-neutral, 𝛽-equilibrated matter. We use Gaussian processes (GPs) to infer constraints on the neutron-star matter EOS at intermediate densities from two different microscopic theories: chiral effective-field theory (𝜒⁢EFT) at baryon densities around nuclear saturation, 𝑛 𝐵 ∼ 𝑛 0 , and perturbative QCD at asymptotically high baryon densities, 𝑛 𝐵 ⩾ 20⁢𝑛 0 . The uncertainties of the 𝜒⁢EFT and pQCD EOSs are obtained using the BUQEYE truncation error model. We demonstrate the flexibility of our framework through the use of two categories of GP kernels: conventional stationary kernels and a nonstationary changepoint kernel. We use the latter to explore potential constraints on the dense matter EOS by including exogenous data representing theory predictions and heavy-ion collision measurements at densities ⩾ 2⁢𝑛 0 . We also use our EOSs to obtain neutron-star mass-radius relations and their uncertainties. Finally, our framework, whose implementation will be available through a GitHub repository, provides a prior distribution for the EOS that can be used in large-scale neutron-star inference frameworks.

Bayesian methods↗

Physics-Informed Machine Learning Models for Predicting the Progress of Reactive-Mixing

This paper presents a physics-informed machine learning (ML) framework to construct reduced-order models (ROMs) for reactive-transport quantities of interest (QoIs) based on high-fidelity numerical simu-lations. QoIs include species decay, product yield, and degree of mixing. The ROMs for QoIs are applied to quantify and understand how the chemical species evolve over time. First, high-resolution datasets for constructing ROMs are generated by solving anisotropic reaction-di?usion equations using a non-negative finite element formulation for di?erent input parameters. The reactive-mixing model input parameters are: time-scale associated with flipping of velocity, spatial-scale controlling small/large vortex structures of velocity, perturbation parameter of the vortex-based velocity, anisotropic dispersion strength/contrast, and molecular diffusion. Second, random forests, F-test, and mutual information criterion are used to evaluate the importance of model inputs/features with respect to QoIs. We observed that anisotropic dispersion strength/contrast is the most important feature and time-scale associated with flipping of velocity is the least important feature. Third, Support Vector Machines (SVM) and Support Vector Regression (SVR) are used to construct ROMs based on the model inputs. The constructed SVR-ROMs are then used to predict scaling of QoIs. We also present estimates and inequalities on the QoIs, which inform that the species decay, mix, and produce in an exponential fashion. These inequalities also inform that a radial basis function is the most suitable kernel for the SVM/SVR models for QoIs. It is observed that R2-score for SVR-ROMs on unseen data is greater than 0.9, implying that the SVR-ROMs are able to predict the reaction-diffusion system state reasonably well. Finally, in terms of the computational cost, the proposed SVM-ROMs are O(107) times faster than running a high-fidelity finite element simulation for evaluating QoIs. This makes the proposed ML-based ROMs attractive for reactive-transport sensing and real-time monitoring applications as they are significantly faster yet reasonably accurate.

Mudunuru, Maruti K.↗

Machine learning Frenkel Hamiltonian parameters to accelerate simulations of exciton dynamics

In this manuscript, we develop multiple machine learning (ML) models to accelerate a scheme for parameterizing site-based models of exciton dynamics from all-atom configurations of condensed phase sexithiophene systems. This scheme encodes the details of a system’s specific molecular morphology in the correlated distributions of model parameters through the analysis of many single-molecule excited-state electronic-structure calculations. These calculations yield excitation energies for each molecule in the system and the network of pair-wise intermolecular electronic couplings. In this work, we demonstrate that the excitation energies can be accurately predicted using a kernel ridge regression (KRR) model with Coulomb matrix featurization. We present two ML models for predicting intermolecular couplings. The first one utilizes a deep neural network and bi-molecular featurization to predict the coupling directly, which we find to perform poorly. The second one utilizes a KRR model to predict unimolecular transition densities, which can subsequently be analyzed to compute the coupling. We find that the latter approach performs excellently, indicating that an effective, generalizable strategy for predicting simple bimolecular properties is through the indirect application of ML to predict higher-order unimolecular properties. Such an approach necessitates a much smaller feature space and can incorporate the insight of well-established molecular physics.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

Ultra-nonlocality in density functional theory for photo-emission spectroscopy

We derive an exact expression for the photocurrent of photo-emission spectroscopy using time-dependent current density functional theory (TDCDFT). This expression is given as an integral over the Kohn-Sham spectral function renormalized by effective potentials that depend on the exchange-correlation kernel of current density functional theory. We analyze in detail the physical content of this expression by making a connection between the density-functional expression and the diagrammatic expansion of the photocurrent within many-body perturbation theory. We further demonstrate that the density functional expression does not provide us with information on the kinetic energy distribution of the photo-electrons. Such information can, in principle, be obtained from TDCDFT by exactly modeling the experiment in which the photocurrent is split into energy contributions by means of an external electromagnetic field outside the sample, as is done in standard detectors. We find, however, that this procedure produces very nonlocal correlations between the exchange-correlation fields in the sample and the detector.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

CEED-MS8:Initial Integration of CEED Software in ECP Applications

Key components of CEED software development involve fast finite element operator storage and evaluation, architecture optimizations, performant algorithms for all orders, global kernels for finite element operators, efficient use of the memory sub-system, optimal data locality and motion, enhanced scalability and parallelism, and fast tensor contractions. As ultimate goals of the project, CEED is exploring and identifying the best algorithms for the full range of discretizations and applying algorithmic and software development to support ECP applications’ needs. In this milestone we report on the initial integration of CEED software in the selected first-wave ECP/CEED apps and the continued exploration of other ECP applications as second-wave (milestone CEED-MS23) and third-wave (milestone CEED-MS35) ECP/CEED application candidates. An extensive evaluation of various approaches for GPU acceleration of the CEED computational motives is also a major focus of the report, both in the settings of CEED’s Nekbone and Laghos miniapps, as well as versions of CEED’s bake-off problems (introduced in milestone CEED-MS6). The first-wave ECP applications targeted for coupling with CEED software include the ExaSMR and MARBL applications. We also report on near-term collaboration with the seed-funded Urban Systems project, as well as engagements with ACME, ExaAM, GEOS and the Application Assessment and Proxy App projects. The ExaSMR project involves coupled thermal-hydraulics/neutronics, with the latter based on Monte Carlo methods and the former based on high-order discretizations of the Navier-Stokes and energy equations for accurate simulation of turbulent transport. The project requires fast and efficient turbulence simulation capabilities that will scale to exascale platforms. Work within CEED will ensure that all aspects of the Nek5000 workflow scale to the target problem space, which involves > 10 10 degrees of freedom in typical production runs. CEED’s goal is to provide implementations that perform with maximum attainable efficiency and deliver low turn-around times. A deliverable for this milestone is beta-release of the GPU-based variant of Nek5000. MARBL is targeting high-energy-density physics problems, including single fluid multi-material hydrodynamics and radiation/magnetic diffusion simulation, with applications in inertial confined fusion, pulsed power experiments, and equation of state/material strength analysis. The code uses high-order methods based arbitrary Lagrangian-Eulerian, direct Eulerian, and unstructured adaptive mesh refinement. A deliverable for this milestone is collaboration with MARBL for improved performance based on partially-assembled operators. The Urban exascale project is planning to couple real-time data with high performance simulations to study particulate transport in cities. Nek5000 will be used for large eddy and RANS (Reynolds-averaged Navier-Stokes) simulations of turbulent transport in these domains. Areas where the CEED team is aiding the Urban Canyon group include creating high-order meshes for urban geometries and developing boundary conditions appropriate for simulations close to real-life urban conditions. The software artifacts delivered as part of this milestone include GPU-enabled versions of the Nekbone and Laghos miniapps, which are provided through the CEED website, http://ceed.exascaleproject.org and the CEED GitHub organization, http://github.com/ceed. In addition to details and results from the above R&D efforts, in this document we are also reporting on other project-wide activities performed in Q4 of FY17, including: the CEED first annual meeting and participation of CEED researchers in a variety of outreach activities.

97 MATHEMATICS AND COMPUTING↗

Multiscale Analysis of Pebble Bed Reactors

– The PEBBED code was developed at the Idaho National Laboratory for design and analysis of pebble-bed high temperature reactors. The diffusion-depletion-pebble-mixing algorithm of the original PEBBED code was enhanced through coupling with the THERMIX-KONVEK code for thermal fluid analysis and by the COMBINE code for online cross section generation. The COMBINE code solves the B-1 or B-3 approximations to the transport equation for neutron slowing down and resonance interactions in a homogeneous medium with simple corrections for shadowing and thermal self-shielding. The number densities of materials within specified regions of the core are averaged and transferred to COMBINE from PEBBED for updating during the burnup iteration. The simple treatment of self-shielding in previous versions of COMBINE led to inaccurate results for cross sections and unsatisfactory core performance calculations. A new version of COMBINE has been developed that treats all levels of heterogeneity using the 1D transport code ANISN. In a 3-stage calculation, slowing down is performed in 167 groups for each homogeneous subregion (kernel, particle layers, graphite shell, control rod absorber annulus, etc.) Particles in a local average pebble are homogenized using ANISN then passed to the next (pebble) stage. A 1D transport solution is again performed over the pebble geometry and the homogenized pebble cross sections are passed to a 1-d radial model of a wedge of the pebble bed core. This wedge may also include homogeneous reflector regions and a control rod region composed of annuli of different absorbing regions. Radial leakage effects are therefore captured with discrete ordinates transport while axial and azimuthal effects are captured with a transverse buckling term. In this paper, results of various PBR models will be compared with comparable models from literature. Performance of the code will be assessed.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

Model Residuals as Shields: A Two-Level Formulation to Defend Smart Grids From Poisoning Attacks

The advancement of smart grids presents both vast opportunities and heightened cybersecurity risks. Data-driven defense mechanisms, though designed as a shield against these threats, can fall prey to poisoning attacks. We delve into regression settings, underscoring the imperative to fortify defenses against a spectrum of poison ratios, notably those above 0.5—an issue scarcely addressed in prior studies. Recognizing the susceptibilities of smart grids and their manipulable sensors, we exploit the very intent of poisoning attacks, compromising model accuracy, as our defense mechanism. Our proposed two-level optimization framework discerns between poisoned and authentic data based on model residuals, outperforming or matching existing methods in 72% to 77% of precision and 75% to 80% of recalls across various poisoning attacks, poison ratios, and datasets. Once the authentic data are identified, the trained model is adaptable for a variety of applications. Comprehensive evaluations on different smart grid datasets, pitted against myriad poisoning schemes, validate our methodology’s edge over existing methods. Here, we also shed light on the implications of model misspecification originating from temporal auto-correlation, a common feature in Internet of Things and smart grid data.

Adversarial machine learning (ML)↗

Microchannel Reactor for Ethanol to Butene: CRADA 503 [Abstract only]

A key challenge facing most bioprocessing operations is that multiple unit operations are required, thereby resulting in complex, energy-intensive, and expensive processes. Further, biomass transportation costs drive the need for smaller, distributed processing plants. To incorporate the smaller scales desirable for biomass, novel processes must be developed with reduced capital costs. With over 20 years of experience in the development and commercialization of microchannel reactor technology, Oregon State University will partner with Pacific Northwest National Laboratory to demonstrate a microchannel reactor with lower capital costs for an alcohol-to-jet (ATJ) process technology that is currently being commercialized by LanzaTech. Ethanol can be produced from biomass feedstocks such as LanzaTech’s proprietary biochemical process using carbon from a number of possible feedstocks; syngas generated from biomass resources (e.g., MSW, organic industrial waste, agriculture waste) or reformed biogas, or from other biomass feedstocks such as corn kernel fiber. Ethanol then undergoes catalytic dehydration to form ethylene followed by a two-step oligomerization, hydrogenation, and fractionation to control the hydrocarbon product slate to the jet-range. Successful process development aided by a market pull for low carbon aviation fuel has spurred scale-up and commercial demonstration. However, Sustainable Aviation Fuel is a very price sensitive market and improved economics through process intensification will make the current ATJ process even more attractive. Recent efforts at PNNL have culminated in the development of a new catalyst technology for the conversion of ethanol to n-butene-rich olefins. A greater than 90% conversion, total olefin selectivity of 80-90% (n-butene selectivity ~60%), and good stability over a 100 hour test duration has been demonstrated at the bench scale. Producing butene-rich olefins directly from ethanol with high yield is new and impactful because the higher olefins can be selectively oligomerized to distillate-range hydrocarbons, thus eliminating one process step from the current ATJ process. Further, coupling the severely endothermic ethanol dehydration with exothermic C-C bond formation results in more energy efficient processing. Additional intensification and energy savings will stem from incorporating this new ethanol to n-butene catalyst technology within the ATJ process implemented using a microchannel reactor platform. Due to recent advances in microchannel manufacturing methods and associated cost reductions we believe the time is right to adapt this technology toward new commercial bioconversion applications.

02 PETROLEUM↗

Unraveling the impact of initial choices and in-loop interventions on learning dynamics in autonomous scanning probe microscopy

The current focus in Autonomous Experimentation (AE) is on developing robust workflows to conduct the AE effectively. This entails the need for well-defined approaches to guide the AE process, including strategies for hyperparameter tuning and high-level human interventions within the workflow loop. This paper presents a comprehensive analysis of the influence of initial experimental conditions and in-loop interventions on the learning dynamics of Deep Kernel Learning (DKL) within the realm of AE in scanning probe microscopy. We explore the concept of the “seed effect,” where the initial experiment setup has a substantial impact on the subsequent learning trajectory. Additionally, we introduce an approach of the seed point interventions in AE allowing the operator to influence the exploration process. Using a dataset from Piezoresponse Force Microscopy on PbTiO 3 thin films, we illustrate the impact of the “seed effect” and in-loop seed interventions on the effectiveness of DKL in predicting material properties. The study highlights the importance of initial choices and adaptive interventions in optimizing learning rates and enhancing the efficiency of automated material characterization. This work offers valuable insights into designing more robust and effective AE workflows in microscopy with potential applications across various characterization techniques.

47 OTHER INSTRUMENTATION↗

Optimizing a Multiple Right-Hand Side Dslash Kernel for Intel Knights Corner

Lattice Quantum Chromo-dynamics (LQCD) stands unique as the only computationally tractable, non-perturbative, and model-independent quantum field theory of the strong nuclear force. LQCD achieves tractability by discretizing our familiar 4-dimensional space-time onto a hypercubic lattice. The core component of the linear solver used in LQCD is the Wilson-Dirac operator whose finite difference component is called the Dslash operator. Dslash is a nearest neighbor stencil operator summing matrix-vector multiplications over lattice points. The arithmetic intensity of a regular Dslash operator acting on a single vector in single precision is between 0.88-2.29 (depending on the tricks employed) making it memory bandwidth bound on most current architectures, including Xeon Phi. Maximizing the performance of Dslash on Xeon Phi requires deliberate exploitation of its architectural features. One key feature of Xeon Phi is its powerful vector processing unit with lengthy vector registers. There is a class of LQCD computations which aims to solve the Dirac equation for the same lattice gauge fields multiple times. In this case one can gain additional reuse of the gauge field, and potentially simplified vectorization, through applying Dslash to several vectors at the same time; often referred to as Multiple Right Hand Side (MRHS) Dslash. Previous work in this area focused on applying this approach to the Highly Improved Staggered operator, and in cases to the multi-Grid algorithm for Wilson clover fermions. Here, we present two approaches to MRHS Dslash: a vector register blocking approach and one using the software package QPhiX with a custom Code Generator for low-level intrinsics. We explore the Knights Corner instruction set, memory hierarchy, and threading capabilities. For standard lattice sizes used in testing, we observed significant speedups, with sustained performance of over 700 GFLOPS (single precision) in one instance. We achieved up to 27% of theoretical peak performance compared to a maximum of 12% obtained by the previous single RHS method in QPhiX.

Joó, Bálint↗

Machine learning-assisted early ignition prediction in a complex flow

Machine learning methods are used to improve the efficiency by which turbulence-resolved simulations predict whether a hydrogen jet in air crossflow will successfully ignite. The flush-mounted jet issues perpendicularly from the wall of a low-speed wind tunnel into a turbulent boundary layer wherein a laser-induced optical breakdown (LIB) hotspot is deposited. A detailed hydrogen chemical mechanism is used to model the radicals and any subsequent chemical reactions. A dielectric-barrier discharge actuator generates body forces and hydrogen radicals near the jet orifice. Here we focus on the success or not of the ignition based on LIB location. A challenge is that definitive determination of this requires long simulations, up to 440 µs after the LIB deposition. This is particularly expensive since multiple simulations are required to find the threshold. To reduce the computational effort, three short-time (91 µs) criteria are proposed, evaluated, and compared: a constructed criterion based on detailed observations of radicals near the stoichiometric surface and two machine learning approaches, each trained on 38 realizations. The constructed criterion provides a low-cost estimate of the ignition boundary that is unambiguous in 45 of the 50 training and test trials, so only 10% of them would need to be simulated longer. The trained neural networks correctly predict outcomes in all cases evaluated, with the more automated procedure—a convolutional neural network (CNN) trained on two-dimensional images—providing the most definitive outcome prediction. From the CNN, a sensitivity analysis is used to determine which kernel features from the two-dimensional input data, as defined by intermediate-layer network weights, are important for identifying ignition.

42 ENGINEERING↗

The Non-Gaussian Nature of Prostate Motion Based on Real-Time Intrafraction Tracking

Purpose: The objective of this work is to test the validity of the Gaussian approximation for prostate motion through characterization of its spatial distribution. Methods and Materials: Real-time intrafraction prostate motion was observed using Calypso 4-dimensional (4D) nonradioactive electromagnetic tracking system. We report the results from a total of 1024 fractions from 31 prostate cancer patients. First, the correlation of prostate motion in right/left (RL), anteroposterior (AP), and superoinferior (SI) direction were determined using Pearson's correlation of coefficient. Then the spatial distribution of prostate motion was analyzed for individual fraction, individual patient including all fractions, and all patients including all fractions. The displacement in RL, AP, SI, oblique, or total direction is fitted into a Gaussian distribution, and a Lilliefors test was used to evaluate the validity of the hypothesis that the displacement is normally distributed. Results: There is high correlation in AP/SI direction (61% of fractions with medium or strong correlation). This is consistent with the longitudinal oblique motion of the prostate, and likely the effect from respiration on an organ confined within the genitourinary diaphragm with the rectum sitting posteriorly and bladder sitting superiorly. In all directions, the non-Gaussian distribution is more common for individual fraction, individual patient including all fractions, and all patients including all fractions. The spatial distribution of prostate motion shows an elongated shape in oblique direction, indicating a higher range of motion in the AP and SI directions. Conclusions: Our results showed that the prostate motion is highly correlated in AP and SI direction, indicating an oblique motion preference. In addition, the spatial distribution of prostate motion is elongated in an oblique direction, indicating that the organ motion dosimetric modeling using Gaussian kernel may need to be modified to account for the particular organ motion character of prostate.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Mathematical nuances of Gaussian process-driven autonomous experimentation

Abstract The fields of machine learning (ML) and artificial intelligence (AI) have transformed almost every aspect of science and engineering. The excitement for AI/ML methods is in large part due to their perceived novelty, as compared to traditional methods of statistics, computation, and applied mathematics. But clearly, all methods in ML have their foundations in mathematical theories, such as function approximation, uncertainty quantification, and function optimization. Autonomous experimentation is no exception; it is often formulated as a chain of off-the-shelf tools, organized in a closed loop, without emphasis on the intricacies of each algorithm involved. The uncomfortable truth is that the success of any ML endeavor, and this includes autonomous experimentation, strongly depends on the sophistication of the underlying mathematical methods and software that have to allow for enough flexibility to consider functions that are in agreement with particular physical theories. We have observed that standard off-the-shelf tools, used by many in the applied ML community, often hide the underlying complexities and therefore perform poorly. In this paper, we want to give a perspective on the intricate connections between mathematics and ML, with a focus on Gaussian process-driven autonomous experimentation. Although the Gaussian process is a powerful mathematical concept, it has to be implemented and customized correctly for optimal performance. We present several simple toy problems to explore these nuances and highlight the importance of mathematical and statistical rigor in autonomous experimentation and ML. One key takeaway is that ML is not, as many had hoped, a set of agnostic plug-and-play solvers for everyday scientific problems, but instead needs expertise and mastery to be applied successfully. Graphical abstract

97 MATHEMATICS AND COMPUTING↗

Fast and accurate implementation of Fourier spectral approximations of nonlocal diffusion operators and its applications

This work is concerned with the Fourier spectral approximation of various integral differential equations associated with some linear nonlocal diffusion and peridynamic operators under periodic boundary conditions. For radially symmetric kernels, the nonlocal operators under consideration are diagonalizable in the Fourier space so that the main computational challenge is on the accurate and fast evaluation of their eigenvalues or Fourier symbols consisting of possibly singular and highly oscillatory integrals. For a large class of fractional power-like kernels, we propose a new approach based on reformulating the Fourier symbols both as coefficients of a series expansion and solutions of some simple ODE models. We then propose a hybrid algorithm that utilizes both truncated series expansions and high order Runge–Kutta ODE solvers to provide fast evaluation of Fourier symbols in both one and higher dimensional spaces. It is shown that this hybrid algorithm is robust, efficient and accurate. As applications, we combine this hybrid spectral discretization in the spatial variables and the fourth-order exponential time differencing Runge–Kutta for temporal discretization to offer high order approximations of some nonlocal gradient dynamics including nonlocal Allen–Cahn equations, nonlocal Cahn–Hilliard equations, and nonlocal phase-field crystal models. Numerical results show the accuracy and effectiveness of the fully discrete scheme and illustrate some interesting phenomena associated with the nonlocal models.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗