SEARCH · Search NASA
Results for “variance reduction”
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.
Taylor approximation and variance reduction for PDE-constrained optimal control under uncertainty
Not provided.
ADVANTG An Automated Variance Reduction Parameter Generator
The AutomateD VAriaNce reducTion Generator (ADVANTG) software automates the generation of variance reduction (VR) parameters for continuous-energy Monte Carlo simulations of fixed-source neutron, photon, and coupled neutron-photon transport problems using MCNP5 (X-5 Monte Carlo Team 2003). ADVANTG generates space- and energy dependent mesh-based weight-window bounds and biased source distributions from three dimensional (3-D) discrete ordinates (SN) calculations that are performed by the Denovo package (Evans et al. 2010). The deterministic calculations can be performed using multiple cores and/or processors (e.g., on multi-core desktop systems and clusters). The final variance reduction parameters are output in a format that can be used with unmodified versions of MCNP. The primary objective of the development of ADVANTG has been to reduce both the user effort and the computational time required to obtain accurate and precise tally estimates across a broad range of challenging transport application areas.
Variance Reduction Factor of Nuclear Data for Integral Neutronics Parameters
We propose a new quantity, a variance reduction factor, to identify nuclear data for which further improvements are required to reduce uncertainties of target integral neutronics parameters. Important energy ranges can be also identified with this variance reduction factor. Variance reduction factors are calculated for several integral neutronics parameters. The usefulness of the variance reduction factors is demonstrated.
A hybrid Monte Carlo-deterministic second moment method with efficient variance reduction
In this work, we present a hybrid method that combines Monte Carlo with deterministic finite element methods to solve a linear Boltzmann transport equation. Our hybrid method runs orders of magnitude faster than Monte Carlo, without sacrificing accuracy, for a proxy problem from radiative transfer that contains both optically-thick and optically-thin material. We believe that this is the first demonstration of a hybrid Second Moment Method in more than one spatial dimension, the first to consider more than one material, and the first to use variance reduction. Our variance reduction approach arises from an asymptotic analysis in which we show that the magnitude of the scattering source grows without bound. We transform the problem to compute the deviation of the radiation intensity from isotropy. The magnitude of the source in the transformed problem is bounded, and the quality of the hybrid method solution is dramatically improved by a substantial reduction in the variance.
Proof that Combining the Forced-collision and DXTRAN Monte Carlo Variance-reduction Techniques is Fair
This report provides mathematical proof that the MCNP DXTRAN (also known in other codes as forced-flight) and forced-collision variance-reduction techniques do not bias the expected value of Monte Carlo simulation estimates when combined. Proof is also provided that the techniques are unbiased when used independently. To prove that the techniques are unbiased, this report derives the first History Score Moment Equation (HSME) for non-multiplying media employing only forced collisions, only DXTRAN, and then both the variance-reduction techniques combined, as defined next. The HSMEs are found by forming the History Score Probability Density Functions (HSPDFs) and then taking the first score moment. Following its derivation, each HSME with a variance-reduction technique employed is reduced to the HSME for an analog simulation. Because the HSME represents the expected contribution to estimators in the simulation, reducing the HSME with variance reduction to the analog HSME shows that the simulation is unbiased despite the variance-reduction technique considered in the HSME. Analysis regarding higher-score moments of each technique is the subject of prior work and is not addressed herein. Throughout this work, the phase space p is defined to be the particle position x, direction-of-flight unit vector $\hat{Ω}$, energy E, and statistical weight w, $$p ≡ (x; \hat{Ω}; E; w).$$ A reduced phase-space excluding the statistical weight of the particle, $$r ≡ (x; \hat{Ω}; E),$$ is also used.
Discrete ordinates analysis of the forced-flight variance reduction technique in Monte Carlo neutral particle transport simulations
This paper presents mathematical formulations and methods to predict the effect of forced-flight variance reduction on Monte Carlo tally variance and calculation time. This includes deducing biasing operators that are then used to construct a history-score probability density function (HSPDF), which represents all possible Monte Carlo random walks and gives the probability of a Monte Carlo history scoring in a tally from a particular phase-space position. The history-score moment equations (HSMEs), the statistical moments of the HSPDF, are then derived to calculate the statistical behavior of the Monte Carlo tally when forced-flight variance reduction is applied. In addition, the future-time equation (FTE) is derived to predict the Monte Carlo computational time as a result of applying forced-flight variance reduction. The solutions of the HSMEs and FTE can be used to predict Monte Carlo computational cost. Furthermore, this work also describes a discrete ordinates method to solve the forced-flight HSMEs and FTE. Several 1-D and 2-D test problems verify that the derivations are performed and implemented correctly.
Cost-optimized Automated Variance Reduction for Highly Angle-dependent Radiation Transport Analyses [Dissertation]
Monte Carlo variance-reduction techniques that directly bias particle direction have historically received limited attention despite being useful for many radiation transport applications such as those in which radiation travels through large regions with a low probability of colliding. One such technique is known as DXTRAN (short for deterministic transport) in the MCNP Monte Carlo radiation transport code. Until now, effectively applying DXTRAN in calculations is largely based on empirical observations of computational performance. Optimal DXTRAN parameters are identified through manual iteration. This work develops new mathematical descriptions of the DXTRAN variance-reduction process and demonstrates a new automated variance-reduction method that applies these mathematical formulations to determine the optimal application of DXTRAN in a given problem.
Implementation of hybrid variance reduction methods in a multi group Monte Carlo code for deep shielding problems
In this paper, the work that has been done to implement variance reduction techniques in a three dimensional, multi group Monte Carlo code - Tortilla, that works within the frame work of the commercial deterministic code - Attila, is presented. This project is aimed to develop an integrated Hybrid code that seamlessly takes advantage of the deterministic and Monte Carlo methods for deep shielding radiation detection problems. Tortilla takes advantage of Attila's features for generating the geometric mesh, cross section library and source definitions. Tortilla can also read importance functions (like adjoint scalar flux) generated from deterministic calculations performed in Attila and use them to employ variance reduction schemes in the Monte Carlo simulation. The variance reduction techniques that are implemented in Tortilla are based on the CADIS (Consistent Adjoint Driven Importance Sampling) method and the LIFT (Local Importance Function Transform) method. These methods make use of the results from an adjoint deterministic calculation to bias the particle transport using techniques like source biasing, survival biasing, transport biasing and weight windows. The results obtained so far and the challenges faced in implementing the variance reduction techniques are reported here. (authors)
A model and variance reduction method for computing statistical outputs of stochastic elliptic partial differential equations
We present a model and variance reduction method for the fast and reliable computation of statistical outputs of stochastic elliptic partial differential equations. Our method consists of three main ingredients: (1) the hybridizable discontinuous Galerkin (HDG) discretization of elliptic partial differential equations (PDEs), which allows us to obtain high-order accurate solutions of the governing PDE; (2) the reduced basis method for a new HDG discretization of the underlying PDE to enable real-time solution of the parameterized PDE in the presence of stochastic parameters; and (3) a multilevel variance reduction method that exploits the statistical correlation among the different reduced basis approximations and the high-fidelity HDG discretization to accelerate the convergence of the Monte Carlo simulations. The multilevel variance reduction method provides efficient computation of the statistical outputs by shifting most of the computational burden from the high-fidelity HDG approximation to the reduced basis approximations. Furthermore, we develop a posteriori error estimates for our approximations of the statistical outputs. Based on these error estimates, we propose an algorithm for optimally choosing both the dimensions of the reduced basis approximations and the sizes of Monte Carlo samples to achieve a given error tolerance. We provide numerical examples to demonstrate the performance of the proposed method.
Automated Variance Reduction Applied to Nuclear Well-Logging Problems
The Monte Carlo method enables detailed, explicit geometric, energy and angular representations, and hence is considered to be the most accurate method available for solving complex radiation transport problems. Because of its associated accuracy, the Monte Carlo method is widely used in the petroleum exploration industry to design, benchmark, and simulate nuclear well-logging tools. Nuclear well-logging tools, which contain neutron and/or gamma sources and two or more detectors, are placed in boreholes that contain water (and possibly other fluids) and that are typically surrounded by a formation (e.g., limestone, sandstone, calcites, or a combination). The response of the detectors to radiation returning from the surrounding formation is used to infer information about the material porosity, density, composition, and associated characteristics. Accurate computer simulation is a key aspect of this exploratory technique. However, because this technique involves calculating highly precise responses (at two or more detectors) based on radiation that has interacted with the surrounding formation, the transport simulations are computationally intensive, requiring significant use of variance reduction techniques, parallel computing, or both. Because of the challenging nature of these problems, nuclear well-logging problems have frequently been used to evaluate the effectiveness of variance reduction techniques (e.g., Refs. 1-4). The primary focus of these works has been on improving the computational efficiency associated with calculating the response at the most challenging detector location, which is typically the detector furthest from the source. Although the objective of nuclear well-logging simulations is to calculate the response at multiple detector locations, until recently none of the numerous variance reduction methods/techniques has been well-suited to simultaneous optimization of multiple detector (tally) regions. Therefore, a separate calculation is typically performed for each desired detector response. This paper summarizes recent work to apply and evaluate the effectiveness of deterministic-adjoint-based variance reduction methods, including a recently developed method for simultaneous optimization of multiple detectors, for two simple nuclear well-logging tool problems that have been widely used in the variance reduction literature. The computational effectiveness of the method for simultaneous optimization is also compared to the use of multiple, individually optimized simulations for these simple well-logging problems.
FW-CADIS Method for Global and Regional Variance Reduction of Monte Carlo Radiation Transport Calculations
This paper presents a hybrid (Monte Carlo/deterministic) method for increasing the efficiency of Monte Carlo calculations of distributions, such as flux or dose rate distributions (e.g., mesh tallies), as well as responses at multiple localized detectors and spectra. This method, referred to as Forward-Weighted CADIS (FW-CADIS), is an extension of the Consistent Adjoint Driven Importance Sampling (CADIS) method, which has been used for more than a decade to very effectively improve the efficiency of Monte Carlo calculations of localized quantities (e.g., flux, dose, or reaction rate at a specific location). The basis of this method is the development of an importance function that represents the importance of particles to the objective of uniform Monte Carlo particle density in the desired tally regions. Implementation of this method utilizes the results from a forward deterministic calculation to develop a forward-weighted source for a deterministic adjoint calculation. The resulting adjoint function is then used to generate consistent space-and energy-dependent source biasing parameters and weight windows that are used in a forward Monte Carlo calculation to obtain more uniform statistical uncertainties in the desired tally regions. The FW-CADIS method has been implemented and demonstrated within the MAVRIC (Monaco with Automated Variance Reduction using Importance Calculations) sequence of SCALE and the ADVANTG (Automated Deterministic Variance Reduction Generator)/MCNP framework. Application of the method to representative real-world problems, including calculation of dose rate and energy-dependent flux throughout the problem space, dose rates in specific areas, and energy spectra at multiple detectors, is presented and discussed. Results of the FW-CADIS method and other recently developed global variance-reduction approaches are also compared, and the FW-CADIS method outperformed the other methods in all cases considered.
Variance reduction in lattice QCD observables via normalizing flows
Normalizing flows can be used to construct unbiased, reduced-variance estimators for lattice field theory observables that are defined by a derivative with respect to action parameters. This work implements the approach for observables involving gluonic operator insertions in the SU(3) Yang-Mills theory and two-flavor QCD in four space-time dimensions. Variance reduction by factors of 10–60 is achieved in glueball correlation functions and in gluonic matrix elements related to hadron structure, with demonstrated computational advantages. The observed variance reduction is found to be approximately independent of the lattice volume, so volume transfer can be utilized to minimize training costs.
Fringe biasing: A variance reduction technique for optically thick meshes
Fringe biasing is a stratified sampling scheme applicable to Monte Carlo thermal radiation transport codes. The thermal emission source in optically thick cells is partitioned into separate contributions from the cell interiors (where the likelihood of the particles escaping the cells is virtually zero) and the 'fringe' regions close to the cell boundaries. Thermal emission in the cell interiors can now be modelled with fewer particles, the remaining particles being concentrated in the fringes so that they are more likely to contribute to the energy exchange between cells. Unlike other techniques for improving the efficiency in optically thick regions (such as random walk and discrete diffusion treatments), fringe biasing has the benefit of simplicity, as the associated changes are restricted to the sourcing routines with the particle tracking routines being unaffected. This paper presents an analysis of the potential for variance reduction achieved from employing the fringe biasing technique. The aim of this analysis is to guide the implementation of this technique in Monte Carlo thermal radiation codes, specifically in order to aid the choice of the fringe width and the proportion of particles allocated to the fringe (which are interrelated) in multi-dimensional simulations, and to confirm that the significant levels of variance reduction achieved in simulations can be understood by studying the behaviour for simple test cases. The variance reduction properties are studied for a single cell in a slab geometry purely absorbing medium, investigating the accuracy of the scalar flux and current tallies on one of the interfaces with the surrounding medium. (authors)
The MAVRIC-Shift Sequence in SCALE for Radiation Transport and Shielding Calculations with Automated Variance Reduction and Parallel Computing
A new MAVRIC-Shift sequence has been implemented and tested within the SCALE Code System to provide parallel computing capabilities for radiation transport and shielding analysis with automated variance reduction. This new sequence provides the most widely used capabilities of the MAVRIC sequence and also features the highly scalable Shift Monte Carlo solver, thus enabling faster time-to-solution when run on multiple cores. This report provides a summary of the theory of variance reduction approaches in MAVRIC and the integration of Shift into the MAVRIC sequence of SCALE. Similarities and differences in methodologies between the traditional MAVRIC sequence and the new MAVRIC-Shift sequence are discussed. Validation cases are presented with models of radiation shielding experiments and verification is provided with a series of practical application cases of interest to the US Nuclear Regulatory Commission, the sponsors of this work, including spent nuclear fuel transportation and storage systems.
Variance reduction techniques for Monte Carlo neutron noise simulations
The small fluctuations of the neutron flux caused by small perturbations of the macroscopic cross-sections take the name of neutron noise. Advanced Monte Carlo methods have been recently proposed in order to solve the neutron noise equations in the frequency domain, which allows establishing reference solutions to validate faster but approximate deterministic solvers. Due to the presence of particles carrying two statistical weights (for the real and imaginary components of the noise field), both of which may be positive or negative, Monte-Carlo simulations of neutron noise pose distinct challenges in terms of variance reduction. In this work we investigate two variance-reduction techniques, namely branchless collisions and weight cancellation, and probe their effectiveness for a benchmark con- figuration concerning the noise field induced by a pin with oscillating cross sections in a fuel assembly.
Variance reduction for Fokker–Planck based particle Monte Carlo schemes
Recently, Fokker–Planck based particle Monte Carlo schemes have been proposed and evaluated for simulations of rarefied gas flows [1–3]. In this paper, the variance reduction for particle Monte Carlo simulations based on the Fokker–Planck model is considered. First, deviational based schemes were derived and reviewed, and it is shown that these deviational methods are not appropriate for practical Fokker–Planck based rarefied gas flow simulations. This is due to the fact that the deviational schemes considered in this study lead either to instabilities in the case of two-weight methods or to large statistical errors if the direct sampling method is applied. Motivated by this conclusion, we developed a novel scheme based on correlated stochastic processes. The main idea here is to synthesize an additional stochastic process with a known solution, which is simultaneously solved together with the main one. By correlating the two processes, the statistical errors can dramatically be reduced; especially for low Mach numbers. To assess the methods, homogeneous relaxation, planar Couette and lid-driven cavity flows were considered. For these test cases, it could be demonstrated that variance reduction based on parallel processes is very robust and effective.
Deflation as a method of variance reduction for estimating the trace of a matrix inverse
Many fields require computing the trace of the inverse of a large, sparse matrix. The typical method used for such computations is the Hutchinson method which is a Monte Carlo (MC) averaging over matrix quadratures. To improve its convergence, several variance reductions techniques have been proposed. In this paper, we study the effects of deflating the near null singular value space. We make two main contributions. First, we analyze the variance of the Hutchinson method as a function of the deflated singular values and vectors. Although this provides good intuition in general, by assuming additionally that the singular vectors are random unitary matrices, we arrive at concise formulas for the deflated variance that include only the variance and mean of the singular values. We make the remarkable observation that deflation may increase variance for Hermitian matrices but not for non-Hermitian ones. This is a rare, if not unique, property where non-Hermitian matrices outperform Hermitian ones. The theory can be used as a model for predicting the benefits of deflation. Second, we use deflation in the context of a large scale application of "disconnected diagrams" in Lattice QCD. On lattices, Hierarchical Probing (HP) has previously provided an order of magnitude of variance reduction over MC by removing "error" from neighboring nodes of increasing distance in the lattice. Although deflation used directly on MC yields a limited improvement of 30% in our problem, when combined with HP they reduce variance by a factor of over 150 compared to MC. For this, we pre-computated 1000 smallest singular values of an ill-conditioned matrix of size 25 million. Furthermore, using PRIMME and a domain-specific Algebraic Multigrid preconditioner, we perform one of the largest eigenvalue computations in Lattice QCD at a fraction of the cost of our trace computation.