ADVANTG 3.0.3: AutomateD VAriaNce reducTion Generator

**AUXILIARY
PROGRAMS**

DENOVO – A 3-D block parallel, multigroup discrete ordinates transport solver

**AUXILIARY LIBRARIES**

BUGLE-96 (DLC-185)

DABL69 (DLC-130)

SCALE-6.1 27n19g and 200n47g ENDF/B-VII libraries (CCC-785)

Oak Ridge National Laboratory, Oak Ridge, Tennessee.

C, C++, Fortran 90, and Python source code and executables for Linux, Mac OS X, and Windows Operating Systems. (C831MNYCP03)

ADVANTG is an automated tool for generating variance reduction parameters for fixed-source continuous-energy Monte Carlo simulations with MCNP5 V1.60 (CCC-810, not included in this distribution) based on approximate 3-D multigroup discrete ordinates adjoint transport solutions generated by Denovo (included in this distribution). The variance reduction parameters generated by ADVANTG consist of space and energy-dependent weight-window bounds and biased source distributions, which are output in formats that can be directly used with unmodified versions of MCNP5. ADVANTG has been applied to neutron, photon, and coupled neutron-photon simulations of real-world radiation detection and shielding scenarios. ADVANTG is compatible with all MCNP5 geometry features and can be used to accelerate cell tallies (F4, F6, F8), surface tallies (F1 and F2), point-detector tallies (F5), and Cartesian mesh tallies (FMESH).

ADVANTG implements the Consistent Adjoint Driven Importance Sampling (CADIS) method and the Forward-Weighted CADIS (FW-CADIS) method for generating variance reduction parameters. The CADIS and FW-CADIS methods provide a prescription for generating space- and energy-dependent weight-window targets and a consistent biased source distribution. The CADIS method was developed for accelerating individual tallies, whereas FW-CADIS can be applied to multiple tallies and mesh tallies. The CADIS method has been demonstrated to provide speed-ups in the tally FOM of O(101-104) across a broad range of radiation detection and shielding problems. The FW-CADIS method has been shown to produce relatively uniform statistical uncertainties across multiple cell tallies and large space- and energy-dependent mesh tallies in real-world applications.

Denovo implements a structured, Cartesian-grid discrete ordinates solver based on the Koch-Baker-Alcouffe algorithm for parallel sweeps across x-y domain blocks. Multiple discretization schemes are available: step characteristics, linear-discontinuous, tri-linear discontinuous and diamond difference (optionally theta-weighted or with negative-flux fixup). Multiple quadrature sets are available: QR product, QR triangular, Gauss-Legendre product, linear-discontinuous finite element, level-symmetric, as well as user-defined quadratures. Denovo contains two embedded first-collision source treatments: an analytic kernel for point sources and a Monte Carlo treatment for distributed sources. The Trilinos parallel solvers package is used to apply GMRES to accelerate the within-group iterations, resulting in a computationally efficient and robust transport solver.

The references provide a detailed description of the CADIS and FW-CADIS methods, as well as the methods and algorithms implemented in the Denovo discrete ordinates package.

**6. RESTRICTIONS
OR LIMITATIONS**

The implementations of the CADIS and FW-CADIS methods in ADVANTG are based on the use of scalar flux estimates from Denovo calculations. As a result, no directional biasing, in either the weight-window parameters or the biased source distributions, is currently implemented.

The run time consumed by ADVANTG is problem-dependent. The majority of computational time is consumed by the discrete ordinates solver, so the primary factors that affect the run time are: the size of the deterministic spatial grid, the physics of the problem (e.g., photon-only versus coupled neutron-photon, presence of upscatter, etc.), the number of quadrature directions, and the number of energy groups.

ADVANTG can drive either serial (i.e., single-processor) or parallel Denovo calculations. For serial calculations on modern desktops, typical run times vary from several minutes to several hours. The run time can be significantly reduced by executing the Denovo calculations in parallel.

ADVANTG and Denovo are operable on modern x86-64 platforms.

The included ADVANTG and Denovo executables run on 64-bit Linux, Mac OS X, and Windows operating systems.

On all platforms, ADVANTG requires an MCNP5-1.60 executable from package CCC-810 and the associated data libraries (neither are included in this distribution). Any of the pre-built i386 or x86-64 MCNP5-1.60 executables from CCC-831 will work. Executables from other versions of MCNP (e.g., MCNP6-1.0) and all MCNPX executables will not work.

On Mac OS X machines, the included ADVANTG binary distribution requires a system installation of Python, which can be obtained by downloading XCode for free from the App Store, then running xcode-select --install from the terminal.

For Windows machines, a 64-bit installation of Python-2.7.x is required (Python-2.7.10 is recommended). Microsoft MPI v5 or later is recommended to enable running Denovo calculations in parallel.

The included executables were created using the following compilers and open-source packages:

**Linux:** GCC-4.8.4 with HDF5-1.8.15-patch1, LAPACK-3.5.0, OpenMPI-1.8.6, Python-2.7.10, Silo 4.10.2-BSD,
SWIG-2.0.12, and Trilinos-12.0.1

**Mac OS X:** GCC-4.9.1 with HDF5-1.8.15, OpenMPI-1.8.4, Silo 4.10.2-BSD, SWIG-3.0.2, and Trilinos-casl-dev

**Windows:** GCC-4.9.2 with HDF5-1.8.15-patch1, LAPACK-3.5.0, Microsoft MPI 5.0.12435.6, Python 2.7.10,
Silo 4.10.2-BSD, SWIG-2.0.12, and Trilinos-12.0.1

ADVANTG and Denovo binaries can be built from source code on Linux, Mac OS X and Windows operating systems using the open-source GNU Compiler Collection (GCC). Denovo additionally supports the Intel and PGI C++ and Fortran compilers.

The open-source packages listed below are required to build
ADVANTG and Denovo source code. Note that the version number specifies the minimum
version required. An exception is that only Python-2.x (with x ≥ 7) is supported.
Python-3.x is not currently supported.

CMake 2.8.11 | http://www.cmake.org/ |

SWIG 2.0.12 | http://www.swig.org/ |

Python 2.7 | https://www.python.org/ |

LAPACK 3.0 | http://www.netlib.org/lapack/ |

Trilinos 12.0 | https://trilinos.org/ |

Silo 4.10 | https://wci.llnl.gov/simulation/computer-codes/silo/ |

OpenMPI | http://www.open-mpi.org/ |

Microsoft MPI | https://msdn.microsoft.com/en-us/library/bb524831.aspx |

Building a parallel version of Denovo requires an MPI (Message Passing Interface) library. For Linux and Mac OS X, OpenMPI is recommended. For Windows, Microsoft MPI is recommended. Both ADVANTG and Denovo output Silo-format files that can be read by the open-source VisIt 3-D, parallel visualization tool (https://wci.llnl.gov/simulation/computer-codes/visit/). Use of VisIt to inspect the quality of the deterministic solutions before starting Monte Carlo runs is highly recommended.

**Documentation
included in package:**

S.W. Mosher et al., “ADVANTG--An Automated Variance Reduction Parameter Generator,” ORNL/TM 2013/416 Rev. 1, Oak Ridge National Laboratory (2015).

J. C. Wagner and A. Haghighat, “Automated Variance Reduction of Monte Carlo Shielding Calculations Using the Discrete Ordinates Adjoint Function,” Nuclear Science and Engineering, 128, 186 (1998).

T. M. Evans, A. S. Stafford, R. N. Slaybaugh, and K. T. Clarno, “Denovo: A New Three-Dimensional Parallel Discrete Ordinates Code in SCALE,” Nuclear Technology, 171, 171–200 (2010).

J. C. Wagner, D. E. Peplow, and S. W. Mosher, “FW-CADIS Method for Global and Semi-Global Variance Reduction of Monte Carlo Radiation Transport Calculations,” Nuclear Science and Engineering, 176, 37–57 (2014).

The CD contains ADVANTG and Denovo executables and source code for Linux, Mac OS X, and Windows systems, documentation, and BUGLE-96, DABL69, SCALE 27n19g, and SCALE 200n47g multigroup cross section libraries in ANISN format.

August 2015

**KEYWORDS:** VARIANCE REDUCTION; DISCRETE ORDINATES; HYBRID TRANSPORT; CADIS; FW-CADIS; MONTE CARLO; MCNP