Implementation of Generalized CoarseMesh Rebalance of NEWTRNX for Acceleration of Parallel BlockJacobi Transport
Abstract
The NEWTRNX transport module solves the multigroup, discreteordinates sourcedriven or keigenvalue transport equation in parallel on a 3D unstructured tetrahedral mesh using the extended step characteristics (ESC), also known as the slicebalance approach (SBA), spatial discretization. The spatial domains are decomposed using METIS. NEWTRNX is under development for nuclear reactor analysis on computer hardware ranging from clusters to massively parallel machines, like the Cray XT4. Transport methods that rely on full sweeps across the spatial domain have been shown to display poor scaling for thousands of processors. The Parallel BlockJacobi (PBJ) algorithm allows each spatial partition to sweep over all discreteordinate directions and energies independently of all other domains, potentially allowing for much better scaling than possible with full sweeps. The PBJ algorithm has been implemented in NEWTRNX using a GaussSeidel iteration in energy and an asynchronous communication by an energy group, such that each partition utilizes the latest boundary solution available for each group before solving the withingroup scattering in a given group. For each energy group, the withingroup scattering converges with a generalized minimum residual (GMRES) solver, preconditioned with beta transport synthetic acceleration ({beta}TSA).
 Authors:
 ORNL
 Publication Date:
 Research Org.:
 Oak Ridge National Lab. (ORNL), Oak Ridge, TN (United States)
 Sponsoring Org.:
 USDOE Laboratory Directed Research and Development (LDRD) Program
 OSTI Identifier:
 958802
 DOE Contract Number:
 DEAC0500OR22725
 Resource Type:
 Conference
 Resource Relation:
 Conference: ANS2007 Winter Meeting, Washington, DC, USA, 20071111, 20071115
 Country of Publication:
 United States
 Language:
 English
 Subject:
 97 MATHEMATICAL METHODS AND COMPUTING; ACCELERATION; ALGORITHMS; COMMUNICATIONS; COMPUTERS; DISCRETE ORDINATE METHOD; IMPLEMENTATION; REACTORS; SCATTERING; TRANSPORT
Citation Formats
Clarno, Kevin T. Implementation of Generalized CoarseMesh Rebalance of NEWTRNX for Acceleration of Parallel BlockJacobi Transport. United States: N. p., 2007.
Web.
Clarno, Kevin T. Implementation of Generalized CoarseMesh Rebalance of NEWTRNX for Acceleration of Parallel BlockJacobi Transport. United States.
Clarno, Kevin T. Mon .
"Implementation of Generalized CoarseMesh Rebalance of NEWTRNX for Acceleration of Parallel BlockJacobi Transport". United States.
doi:.
@article{osti_958802,
title = {Implementation of Generalized CoarseMesh Rebalance of NEWTRNX for Acceleration of Parallel BlockJacobi Transport},
author = {Clarno, Kevin T},
abstractNote = {The NEWTRNX transport module solves the multigroup, discreteordinates sourcedriven or keigenvalue transport equation in parallel on a 3D unstructured tetrahedral mesh using the extended step characteristics (ESC), also known as the slicebalance approach (SBA), spatial discretization. The spatial domains are decomposed using METIS. NEWTRNX is under development for nuclear reactor analysis on computer hardware ranging from clusters to massively parallel machines, like the Cray XT4. Transport methods that rely on full sweeps across the spatial domain have been shown to display poor scaling for thousands of processors. The Parallel BlockJacobi (PBJ) algorithm allows each spatial partition to sweep over all discreteordinate directions and energies independently of all other domains, potentially allowing for much better scaling than possible with full sweeps. The PBJ algorithm has been implemented in NEWTRNX using a GaussSeidel iteration in energy and an asynchronous communication by an energy group, such that each partition utilizes the latest boundary solution available for each group before solving the withingroup scattering in a given group. For each energy group, the withingroup scattering converges with a generalized minimum residual (GMRES) solver, preconditioned with beta transport synthetic acceleration ({beta}TSA).},
doi = {},
journal = {},
number = ,
volume = ,
place = {United States},
year = {Mon Jan 01 00:00:00 EST 2007},
month = {Mon Jan 01 00:00:00 EST 2007}
}

This paper proposes a new acceleration method for neutron transport calculations: the generalized coarsemesh rebalance (GCMR) method. The GCMR method is a unified scheme of the traditional coarsemesh rebalance (CMR) and the coarsemesh finite difference (CMFD) acceleration methods. Namely, by using an appropriate acceleration factor, formulation of the GCMR method becomes identical to that of the CMR or CMFD method. This also indicates that the convergence property of the GCMR method can be controlled by the acceleration factor since the convergence properties of the CMR and CMFD methods are generally different. In order to evaluate the convergence property of themore »

Fourier analysis of parallel inexact BlockJacobi splitting with transport synthetic acceleration in slab geometry
A Fourier analysis is conducted for the discreteordinates (SN) approximation of the neutron transport problem solved with Richardson iteration (Source Iteration) and Richardson iteration preconditioned with Transport Synthetic Acceleration (TSA), using the Parallel BlockJacobi (PBJ) algorithm. Both 'traditional' TSA (TTSA) and a 'modified' TSA (MTSA), in which only the scattering in the low order equations is reduced by some nonnegative factor {beta} and < 1, are considered. The results for the unaccelerated algorithm show that convergence of the PBJ algorithm can degrade. The PBJ algorithm with TTSA can be effective provided the {beta} parameter is properly tuned for a givenmore » 
CoarseMesh Angular Dependent Rebalance Acceleration of the Discrete Ordinates Transport Calculations
A new nonlinear coarsemesh rebalance (CMR) method is developed and tested to accelerate the one and twodimensional discrete ordinates neutron transport calculations. The method is based on rebalance factors that are angular dependent and defined on the coarsemesh boundaries only. Unlike the conventional CMR method that is only conditionally stable, Fourier analysis and numerical tests show that this coarsemesh angular dependent rebalance (CMADR) method is unconditionally stable for any optical thickness, scattering ratio, and coarseness and that the acceleration is very effective in most cases. 
Application of coarsemesh rebalance acceleration to Monte Carlo eigenvalue problems
The coarsemesh rebalance method is adopted in Monte Carlo schemes for aiming at accelerating the convergence of a source iteration process to obtain the eigenvalue of a nuclear reactor system. At every completion of the Monte Carlo game for one batch of neutron histories, the scaling factor for the neutron flux is calculated to achieve the neutron balance in each coarsemesh zone. This rebalance factor is multiplied to the weight of each fission neutron in the coarsemesh zone for playing the next Monte Carlo game. The numerical examples show that the present rebalance method gives a new usable sampling techniquemore »