Computational design-based optimization is a well-used tool in science and engineering. Our report documents the successful use of a particle sensitivity analysis for design-based optimization within Monte Carlo sampling-based particle simulation—a currently unavailable capability. Such a capability enables the particle simulation communities to go beyond forward simulation and promises to reduce the burden on overworked analysts by getting more done with less computation.
Neuromorphic computing, which aims to replicate the computational structure and architecture of the brain in synthetic hardware, has typically focused on artificial intelligence applications. What is less explored is whether such brain-inspired hardware can provide value beyond cognitive tasks. Here we show that the high degree of parallelism and configurability of spiking neuromorphic architectures makes them well suited to implement random walks via discrete-time Markov chains. These random walks are useful in Monte Carlo methods, which represent a fundamental computational tool for solving a wide range of numerical computing tasks. Using IBM’s TrueNorth and Intel’s Loihi neuromorphic computing platforms, we show that our neuromorphic computing algorithm for generating random walk approximations of diffusion offers advantages in energy-efficient computation compared with conventional approaches. We also show that our neuromorphic computing algorithm can be extended to more sophisticated jump-diffusion processes that are useful in a range of applications, including financial economics, particle physics and machine learning.
The widely parallel, spiking neural networks of neuromorphic processors can enable computationally powerful formulations. While recent interest has focused on primarily machine learning tasks, the space of appropriate applications is wide and continually expanding. Here, we leverage the parallel and event-driven structure to solve a steady state heat equation using a random walk method. The random walk can be executed fully within a spiking neural network using stochastic neuron behavior, and we provide results from both IBM TrueNorth and Intel Loihi implementations. Additionally, we position this algorithm as a potential scalable benchmark for neuromorphic systems.
Neutral particle transport in media exhibiting large and complex material property spatial variation is modeled by representing cross sections as lognormal random functions of space and generated through a nonlinear memory-less transformation of a Gaussian process with covariance uniquely determined by the covariance of the cross section. A Karhunen-Loève decomposition of the Gaussian process is implemented to effciently generate realizations of the random cross sections and Woodcock Monte Carlo used to transport particles on each realization and generate benchmark solutions for the mean and variance of the particle flux as well as probability densities of the particle reflectance and transmittance. A computationally effcient stochastic collocation method is implemented to directly compute the statistical moments such as the mean and variance, while a polynomial chaos expansion in conjunction with stochastic collocation provides a convenient surrogate model that also produces probability densities of output quantities of interest. Extensive numerical testing demonstrates that use of stochastic reduced-order modeling provides an accurate and cost-effective alternative to random sampling for particle transport in random media.
Stochastic media transport problems have long posed challenges for accurate modeling. Brute force Monte Carlo or deterministic sampling of realizations can be expensive in order to achieve the desired accuracy. The well-known Levermore-Pomraning (LP) closure is very simple and inexpensive, but is inaccurate in many circumstances. We propose a generalization to the LP closure that may help bridge the gap between the two approaches. Our model consists of local calculations to approximately determine the relationship between ensemble-averaged angular fluxes and the corresponding averages at material interfaces. The expense and accuracy of the method are related to how "local" the model is and how much local detail it contains. We show through numerical results that our approach is more accurate than LP for benchmark problems, provided that we capture enough local detail. Thus we identify two approaches to using ensemble calculations for stochastic media calculations: direct averaging of ensemble results for transport quantities of interest, or indirect use via a generalized LP equation to determine those same quantities; in some cases the latter method is more efficient. However, the method is subject to creating ill-posed problems if insufficient local detail is included in the model.
We present an improved deterministic method for analyzing transport problems in random media. In the original method realizations were generated by means of a product quadrature rule; transport calculations were performed on each realization and the results combined to produce ensemble averages. In the present work we recognize that many of these realizations yield identical transport problems. We describe a method to generate only unique transport problems with the proper weighting to produce identical ensemble-averaged results at reduced computational cost. We also describe a method to ignore relatively unimportant realizations in order to obtain nearly identical results with further reduction in costs. Our results demonstrate that these changes allow for the analysis of problems of greater complexity than was practical for the original algorithm.
A Monte Carlo solution method for the system of deterministic equations arising in the application of stochastic collocation (SCM) and stochastic Galerkin (SGM) methods in radiation transport computations with uncertainty is presented for an arbitrary number of materials each containing two uncertain random cross sections. Moments of the resulting random flux are calculated using an intrusive and a non-intrusive Monte Carlo based SCM and two different SGM implementations each with two different truncation methods and compared to the brute force Monte Carlo sampling approach. For the intrusive SCM and SGM, a single set of particle histories is solved and weight adjustments are used to produce flux moments for the stochastic problem. Memory and runtime scaling of each method is compared for increased complexity in stochastic dimensionality and moment truncation. Results are also compared for efficiency in terms of a statistical figure-of-merit. The memory savings for the total-order truncation method prove significant over the full-tensor-product truncation. Scaling shows relatively constant cost per moment calculated of SCM and tensor-product SGM. Total-order truncation may be worthwhile despite poorer runtime scaling by achieving better accuracy at lower cost. The figure-of-merit results show that all of the intrusive methods can improve efficiency for calculating low-order moments, but the intrusive SCM approach is the most efficient for calculating high-order moments.
This is the final report on the LDRD, though the interested reader is referred to the ANS Transactions paper which more thoroughly documents the technical work of this project.
This report describes efforts to automate the biasing of coupled electron-photon Monte Carlo particle transport calculations. The approach was based on weight-windows biasing. Weight-window settings were determined using adjoint-flux Monte Carlo calculations. A variety of algorithms were investigated for adaptivity of the Monte Carlo tallies. Tree data structures were used to investigate spatial partitioning. Functional-expansion tallies were used to investigate higher-order spatial representations.
ITS is a powerful and user-friendly software package permitting state-of-the-art Monte Carlo solution of lineartime-independent coupled electron/photon radiation transport problems, with or without the presence of macroscopic electric and magnetic fields of arbitrary spatial dependence. Our goal has been to simultaneously maximize operational simplicity and physical accuracy. Through a set of preprocessor directives, the user selects one of the many ITS codes. The ease with which the makefile system is applied combines with an input scheme based on order-independent descriptive keywords that makes maximum use of defaults and internal error checking to provide experimentalists and theorists alike with a method for the routine but rigorous solution of sophisticated radiation transport problems. Physical rigor is provided by employing accurate cross sections, sampling distributions, and physical models for describing the production and transport of the electron/photon cascade from 1.0 GeV down to 1.0 keV. The availability of source code permits the more sophisticated user to tailor the codes to specific applications and to extend the capabilities of the codes to more complex applications. Version 6, the latest version of ITS, contains (1) improvements to the ITS 5.0 codes, and (2) conversion to Fortran 90. The general user friendliness of the software has been enhanced through memory allocation to reduce the need for users to modify and recompile the code.
ITS is a powerful and user-friendly software package permitting state-of-the-art Monte Carlo solution of linear time-independent coupled electron/photon radiation transport problems, with or without the presence of macroscopic electric and magnetic fields of arbitrary spatial dependence. Our goal has been to simultaneously maximize operational simplicity and physical accuracy. Through a set of preprocessor directives, the user selects one of the many ITS codes. The ease with which the makefile system is applied combines with an input scheme based on order-independent descriptive keywords that makes maximum use of defaults and internal error checking to provide experimentalists and theorists alike with a method for the routine but rigorous solution of sophisticated radiation transport problems. Physical rigor is provided by employing accurate cross sections, sampling distributions, and physical models for describing the production and transport of the electron/photon cascade from 1.0 GeV down to 1.0 keV. The availability of source code permits the more sophisticated user to tailor the codes to specific applications and to extend the capabilities of the codes to more complex applications. Version 5.0, the latest version of ITS, contains (1) improvements to the ITS 3.0 continuous-energy codes, (2) multigroup codes with adjoint transport capabilities, (3) parallel implementations of all ITS codes, (4) a general purpose geometry engine for linking with CAD or other geometry formats, and (5) the Cholla facet geometry library. Moreover, the general user friendliness of the software has been enhanced through increased internal error checking and improved code portability.
Extremely short collision mean free paths and near-singular elastic and inelastic differential cross sections (DCS) make analog Monte Carlo simulation an impractical tool for charged particle transport. The widely used alternative, the condensed history method, while efficient, also suffers from several limitations arising from the use of precomputed smooth distributions for sampling. There is much interest in developing computationally efficient algorithms that implement the correct transport mechanics. Here we present a nonanalog transport-based method that incorporates the correct transport mechanics and is computationally efficient for implementation in single event Monte Carlo codes. Our method systematically preserves important physics and is mathematically rigorous. It builds on higher order Fokker-Planck and Boltzmann Fokker-Planck representations of the scattering and energy-loss process, and we accordingly refer to it as a Generalized Boltzmann Fokker-Planck (GBFP) approach. We postulate the existence of nonanalog single collision scattering and energy-loss distributions (differential cross sections) and impose the constraint that the first few momentum transfer and energy loss moments be identical to corresponding analog values. This is effected through a decomposition or hybridizing scheme wherein the singular forward peaked, small energy-transfer collisions are isolated and de-singularized using different moment-preserving strategies, while the large angle, large energy-transfer collisions are described by the exact (analog) DCS or approximated to a high degree of accuracy. The inclusion of the latter component allows the higher angle and energy-loss moments to be accurately captured. This procedure yields a regularized transport model characterized by longer mean free paths and smoother scattering and energy transfer kernels than analog. In practice, acceptable accuracy is achieved with two rigorously preserved moments, but accuracy can be systematically increased to analog level by preserving successively higher moments with almost no change to the algorithm. Details of specific moment-preserving strategies will be described and results presented for dose in heterogeneous media due to a pencil beam and a line source of monoenergetic electrons. Error and runtimes of our nonanalog formulations will be contrasted against condensed history implementations.
This document describes the modeling of the physics (and eventually features) in the Integrated TIGER Series (ITS) codes [Franke 04] which is largely pulled from various sources in the open literature (especially [Seltzer 88], [Seltzer 91], [Lorence 89], [Halbleib 92]), although those sources often describe the ETRAN Code from which the physics engine of ITS is derived, not necessarily identical. This is meant to be an evolving document, with more coverage and detail as time goes on. As such, entire sections are still incomplete. Presently, this document covers the continuous-energy ITS codes with more completeness on photon transport (though electron transport will not be completely ignored). In particular, this document does not cover the Multigroup code, MCODES (externally applied electromagnetic fields), or high-energy phenomena (photon pair-production). In this version, equations are largely left to the references though they may be pulled in over time.
This test plan describes the testing strategy for the ITS (Integrated-TIGER-Series) suite of codes. The processes and procedures for performing both verification and validation tests are described. ITS Version 5.0 was developed under the NNSA's ASC program and supports Sandia's stockpile stewardship mission.
ITS is a powerful and user-friendly software package permitting state of the art Monte Carlo solution of linear time-independent couple electron/photon radiation transport problems, with or without the presence of macroscopic electric and magnetic fields of arbitrary spatial dependence. Our goal has been to simultaneously maximize operational simplicity and physical accuracy. Through a set of preprocessor directives, the user selects one of the many ITS codes. The ease with which the makefile system is applied combines with an input scheme based on order-independent descriptive keywords that makes maximum use of defaults and internal error checking to provide experimentalists and theorists alike with a method for the routine but rigorous solution of sophisticated radiation transport problems. Physical rigor is provided by employing accurate cross sections, sampling distributions, and physical models for describing the production and transport of the electron/photon cascade from 1.0 GeV down to 1.0 keV. The availability of source code permits the more sophisticated user to tailor the codes to specific applications and to extend the capabilities of the codes to more complex applications. Version 5.0, the latest version of ITS, contains (1) improvements to the ITS 3.0 continuous-energy codes, (2)multigroup codes with adjoint transport capabilities, and (3) parallel implementations of all ITS codes. Moreover the general user friendliness of the software has been enhanced through increased internal error checking and improved code portability.
The authors consider the steady-state transport of normally incident pencil beams of radiation in slabs of material. A method has been developed for determining the exact radial moments of 3-D beams of radiation as a function of depth into the slab, by solving systems of 1-D transport equations. They implement these radial moment equations in the ONEBFP discrete ordinates code and simulate energy-dependent, coupled electron-photon beams using CEPXS-generated cross sections. Modified P{sub N} synthetic acceleration is employed to speed up the iterative convergence of the 1-D charged particle calculations. For high-energy photon beams, a hybrid Monte Carlo/discrete ordinates method is examined. They demonstrate the efficiency of the calculations and make comparisons with 3-D Monte Carlo calculations. Thus, by solving 1-D transport equations, they obtain realistic multidimensional information concerning the broadening of electron-photon beams. This information is relevant to fields such as industrial radiography, medical imaging, radiation oncology, particle accelerators, and lasers.