GPUQT

An efficient implementation of linear scaling quantum transport (LSQT) methods which supports both pure CPU and GPU+CPU computations. This code can be used to obtain charge and spin transport properties of large systems described by a real-space tight-binding Hamiltonian. This is a work in progress. We aim to complete version 1.0 in a few months.

Related code

There is an independent code from Stephan Roche's group which might be more comprehensive than GPUQT. Here is the link: https://github.com/proyectoRMP/proyectormp.github.io

References

The most original paper on this method is:

• [1] S. Roche and D. Mayou, Conductivity of Quasiperiodic Systems: A Numerical Study, Phys. Rev. Lett. 79, 2518 (1997). https://doi.org/10.1103/PhysRevLett.79.2518

The major reference for the CUDA implementation is

• [2] Z. Fan, A. Uppstu, T. Siro, and A. Harju, Efficient linear-scaling quantum transport calculations on graphics processing units and applications on electron transport in graphene, Comput. Phys. Commun. 185, 28 (2014). https://doi.org/10.1016/j.cpc.2013.08.009

This code was first published along with the following paper:

• [3] Z. Fan, V. Vierimaa, and Ari Harju, GPUQT: An efficient linear-scaling quantum transport code fully implemented on graphics processing units, Comput. Phys. Commun. 230, 113 (2018). https://doi.org/10.1016/j.cpc.2018.04.013

There is a comprehensive review article discussing the linear scaling quantum transport methods:

• [4] Zheyong Fan, Jose Hugo Garcia, Aron W. Cummings, Jose-Eduardo Barrios, Michel Panhans, Ari Harju, Frank Ortmann, and Stephan Roche, Linear Scaling Quantum Transport Methodologies, submitted to Reviews of Modern Physics, https://arxiv.org/abs/1811.07387

File organizations

• After downloading and unpacking this code, one can see three directories: src, doc, and examples.

• The directory src contains all the source files (with suffix .h or .cu) of the main code and two makefiles. The source files are:

  • main.cu - the main function
  • lsqt.h and lsqt.cu - the "driver function"
  • sigma.h and sigma.cu - functions to obtain the transport properties
  • model.h, model.cu, model_general.cu, and model_lattice.cu - class to define the simulation model
  • anderson.h and anderson.cu - class to add Anderson disorder
  • charge.h and charge.cu - class to add charged impurities
  • hamiltonian.h and hamiltonian.cu - class to perform the matrix-related operations
  • vector.h and vector.cu - class to perform the vector-related operations

• The directory examples contains some sub-directories and sub-sub-directories, corresponding to different examples.

• There is also a file named input.txt in the examples directory, which is a "driver input file" (see below).

• The doc directory contains the pdf and source files for the manual of this code.

Prerequisites

• To use the CPU version, it only requires a g++ compiler.
• To use the GPU version, it also requires a CUDA-enabled GPU with compute capability of 3.5 or higher and a CUDA toolkit.
• The code has only been tested in linux systems.

Installing

• Go to src and
  • type make -f makefile.cpu to build the CPU version. This will produce an executable called lsqt_cpu in the src folder.
  • type make -f makefile.gpu to build the GPU version. This will produce an executable called lsqt_gpu in the src folder.

Running the examples

• Edit the file examples/input.txt to include the paths (relative or absolute) of the working directories containing the examples you want to run.

• Go to the main folder where you can see the src folder and type one of the following commands:

  • src/lsqt_gpu examples/input.txt
  • src/lsqt_cpu examples/input.txt

• The results will be written into the output files (with suffix .out) in the working directories specified in examples/input.txt. If you run a simulation multiple times, new data will be appended to the existing output files.

Analyzing the results

Go to the working directories and run the MATLAB scripts we have prepared. After getting familiar with the output files, one can analyze the results using her/his favorite computer language(s).

Authors

• Zheyong Fan (Aalto University): Wrote the first working version of this code.

• Ville Vierimaa (Aalto University): Changed the code from the original C style to the current C++ style and made many other improvements.

• Ari Harju (Aalto University): The supervisor of this project.

Contact

• Zheyong Fan: brucenju(at)gmail.com; zheyong.fan(at)aalto.fi;