PENTRAN™
Our deterministic 3-D radiation transport package is the PENTRAN code system (Parallel Environment Neutral-particle TRANsport), which operates on parallel computers using MPI to automatically distribute a radiation transport problem using angular, energy, and/or spatial decomposition onto a collective of processors. 

All-Inclusive Processing Capabilities. The PENTRAN system includes a suite of pre- and post-processing codes for automatic 3-D mesh generation (PENMSH-XP), including automatic input deck generation, and data collection tools for 3-D data analysis (PENDATA & PENPRL). 

Bottom-line Benefits.  With PENTRAN, an entire 3-D multigroup system calculation can be modeled in as little a few hours with extremely high fidelity and accuracy.

Other benefits include:

• Parallel Memory. All dimensions for memory intensive arrays are partitioned locally 
• Minimal Memory Useage. Tuning of memory parameters also occurs automatically in a two-level memory model
• Space, Angle, & Energy Decomposition. Full phase space decomposition is available: (space, angle, and energy), with fully automatic scaling of the problem to n process
• Communicators. To further maximize parallel execution efficiency, communication among processors is, where practical, carried out only between processors specifically involved in a given task
• Adaptive Differencing strategy. Different differencing schemes can be automatically assigned to different coarse meshes; therefore, differencing can be adaptive based on each coarse mesh
• Covers All Grids – Coarse, Medium and Fine. Variable 3-D meshing is available between different coarse meshes, with “medium” and “fine” multigrid meshing within each coarse cell
• Incorporates Taylor Projection Mesh Coupling (TPMC). Mesh interpolations for angular fluxes between adjacent coarse cell surfaces are accomplished using TPMC. This is performed in all transport sweeps to increase accuracy and minimize the information loss
• Rebalance & ADS.  Each processor independently performs rebalance using a direct solution over a zoned subset of coarse meshes to obtain group rebalancing factors following transport sweeps.  Angular sweeps are ordered/decomposed in a sequence for Alternating Direction Sweeping (ADS)
• Flux Moment Preconditioning. PENTRAN now incorporates options that allow the user to precondition the 0th and 1st flux moments in the initial SN source iteration sweep using an effective initial guess to the polar angle flux moments to provide acceleration to the SN source iteration.
• Subdomains. The spatial coarse mesh structure in PENTRAN fundamentally defines rebalance subdomains, parallel spatial decomposition subdomains, and adaptive SN differencing subdomains
• Quadratures and Legendre Scattering Expansion. Full support for level-symmetric angular quadratures through S20, and arbitrary SN order for Pn-Tn angular quadratures, fully conserving even and odd moment conditions.  Legendre scattering order (PN) is now completely arbitrary
• FIDO & Execution. Industry standard, free field format “FIDO” input is used, with standardized order for cross sections.
• Data management utility – PENDATA. Seamlessly gathers data automatically following a parallel PENTRAN run, and provides several options for the user in stripping results from parallel output files, including data extractions from binary file storage
• PENMSH-XP Mesh or Results Generation. Planar images (z-levels) or 3-D TECPLOT renderings are available using the PENMSH-XP utility. This utility generates a 3-D Cartesian mesh and can be used to automatically generate a PENTRAN input deck
• Platform Independent Code. The code is written in ANSI FORTRAN-90; it has been implemented on a single-processor PC, an IBM RISC-Workstations, SUN Workstation, and in parallel on the IBM Scalable PowerParallel Systems (IBM-SPs and BLUEGENE/P) and CRAY/SGI supercomputers, and on a variety of Linux based PC Clusters, andon the Apple G5 cluster

Benchmark testing has demonstrated that PENTRAN is greater than 97% parallelizeable, resulting in excellent parallel speedups. Actual performance depends on the problem being solved, differencing and acceleration methods used, problem load balancing, red-black coloring, and applied decomposition strategy. It has recently been executed with very high speedup on over 4000 cores; release of this data is pending.

Due to the inherent scalable parallel memory structure used, increasingly large 3-D problems that could not be solved on single processor platforms and/or within a reasonable time period can be solved in parallel using PENTRAN by adding processors.

PENTRAN has been vital to detection problems related to threat reduction and homeland security.  Recently, PENTRAN has been used to model whole core PWR reactor problems in a joint research investigation between EDF and IBM.  Other applications of PENTRAN are in modeling deterministic simulations in medical physics (high energy photons and proton therapy); these investigations are ongoing.