ShengBTE: A solver of the Boltzmann transport equation for phonons

Abstract

ShengBTE is a software package for computing the lattice thermal conductivity of crystalline bulk materials and nanowires with diffusive boundary conditions. It is based on a full iterative solution to the Boltzmann transport equation. Its main inputs are sets of second- and third-order interatomic force constants, which can be calculated using third-party ab-initio packages. Dirac delta distributions arising from conservation of energy are approximated by Gaussian functions. A locally adaptive algorithm is used to determine each process-specific broadening parameter, which renders the method fully parameter free. The code is free software, written in Fortran and parallelized using MPI. A complementary Python script to help compute third-order interatomic force constants from a minimum number of ab-initio calculations, using a real-space finite-difference approach, is also publicly available for download. Here we discuss the design and implementation of both pieces of software and present results for three example systems: Si, InAs and lonsdaleite. Program summaryProgram title: ShengBTECatalogue identifier: AESL_v1_0Program summary URL:http://cpc.cs.qub.ac.uk/summaries/AESL_v1_0.htmlProgram obtainable from: CPC Program Library, Queen’s University, Belfast, N. IrelandLicensing provisions: GNU General Public License, version 3No. of lines in distributed program, including test data, etc.: 292052No. of bytes in distributed program, including test data, etc.: 1989781Distribution format: tar.gzProgramming language: Fortran 90, MPI.Computer: Non-specific.Operating system: Unix/Linux.Has the code been vectorized or parallelized?: Yes, parallelized using MPI.RAM: Up to several GBClassification: 7.9.External routines: LAPACK, MPI, spglib (http://spglib.sourceforge.net/)Nature of problem:Calculation of thermal conductivity and related quantities, determination of scattering rates for allowed three-phonon processesSolution method:Iterative solution, locally adaptive Gaussian broadeningRunning time:Up to several hours on several tens of processors