libnegf / libnegf

libNEGF

libNEGF is a general library for Non Equilibrium Green's Functions.

With libNEGF you can calculate Equilibrium and Non Equilibrium Green's Function in open systems and related quantities, within an efficient sparse iterative scheme.

libNEGF is available under LGPL license.

For an example of application, see the use of libNEGF in dftb+

*A. Pecchia, L. Salvucci, G. Penazzi and A. Di Carlo, New Journal of Physics 10, 065022 (2008)

libnegf's Issues

Downstream issue from call destroy(negf%cont(i)%HC)

Segfault for transport only calculation.

Release 1.1.0 is broken: it doesn't install any headers

segfault when calling destroy_negf if contacts are not initialized

This minimal test causes a segfault:

program hello
  use libnegf
  use lib_param
  implicit none
  Type(Tnegf), pointer :: pnegf

  call init_negf(pnegf)
  call destroy_negf(pnegf)

end program hello

The segfault happens during destroy_matrices called in destroy_negf, because negf%cont has not been allocated. Fix: add an allocation check or allocate a default 0 size array during init.

Linking error with MPI on OSX

Compiling libnegf (release v1.0.2) on OSX with MPI support fails when linking the dylib with

Undefined symbols for architecture x86_64:
  "_mpi_cart_create_", referenced from:
      ___mpi_globals_MOD_negf_cart_init in mpi_globals.f90.o
  "_mpi_cart_sub_", referenced from:
      ___mpi_globals_MOD_negf_cart_init in mpi_globals.f90.o
ld: symbol(s) not found for architecture x86_64

See conda-forge/libnegf-feedstock#4

Fails when WITH_INELASTIC=ON: Error: Can't open included file 'fftw3.f03'

Version: 1.0.3
OS: FreeBSD 13

Conda package

I would like to package libnegf for conda-forge and make it useable with the DFTB+ package there. Let me know if there is interest or if you want to join as feedstock maintainers on this.

One thing that would be incredibly helpful for this purpose would be an official release of libnegf. Would you consider creating a versioned release of libnegf and export this version via CMake/pkg-config? I could help setting up the CMake to do the latter if it is not already in place.

Let me know if you are interested in distributing libnegf on conda-forge.

Binary files in repository

Hi,

not sure, whether on purpose or not, but you have two (non-pdf) binary files in the repository:

• ext_sparskit/INOUT/test

• ext_metis/metis-4.0/Graphs/mtest

You may consider to remove them. I think, at least the first one is really superfluous, as the source file for the test program is available. I am not sure about the second one...

Remove Metis from repo ?

Connected to Balint's issue, I think we should remove Metis-4.0, as it is not even used. In case we will decide we need it we can set an automatic download.

Electron entropy (required for finite temperature free energy)

This is currently not available.

PEXSI evaluates this for contour integrals in closed systems using the AAA algorithm.

CMake config fails without explicit BUILD_SHARED_LIBS setting

Note, the current official release does not build with CMake as a standalone, if one does not set BUILD_SHARED_LIBS explicitely. Nothing tragic, but could be maybe fixed when sometimes later a new release is issued...

cmake ..

yields

...
CMake Error at ext_system/CMakeLists.txt:10 (set_target_properties):
  set_target_properties called with incorrect number of arguments.

Using

cmake -DBUILD_SHARED_LIBS=False

works, though.

Error message from sparsekit_drv.F90 halt

Issue discussed downstream in the dftb issues list

Support for MPI in cmake, testing and C api

• When compiled with MPI support, we need to get and compile mpifx as well. This is not supported in cmake yet
• For the C api is a bit more tricky. We should pass a bare communicator and wrap as mpifx_comm in libnegf as mpifx has no C bindings, but is not trivial (see e.g. https://stackoverflow.com/questions/42530620/how-to-pass-mpi-communicator-handle-from-fortran-to-c-using-iso-c-binding). As intermediate step we could have a function which initialize MPI directly within libnegf. It would be enough to compile and add testing.

Free energy under bias is not available

The usual mu * N_elec in the device definition for the Gibbs electronic free energy is only available for equilibrium calculations at the moment. In principle steady state free energy and more general non-equilibrium should be possible.

Clash of LICENSE and license/ on case-insensitive file systems

Would it be possible to rename license/ to, perhaps, licenses/? It is common in MacOS to have a case-insensitive file system, so it's impossible to have both LICENSE and license/ in the same directory.

CMake export files

It would be cool, if libNEGF would export a CMake file, so that it can be invoked in other CMake projects as an external project. See my cmake branch of scalapackfx for a possible minimal example.

Fails to compile with Intel compilers (version 2021.1)

I encountered this problem while building DFTB+ with libnegf and the Intel oneAPI classic compilers (ifort, icc).

Tested version of libnegf: 83d7588

Intel Fortran version:

> ifort -V
Intel(R) Fortran Intel(R) 64 Compiler Classic for applications running on Intel(R) 64, Version 2021.1 Build 20201112_000000
Copyright (C) 1985-2020 Intel Corporation.  All rights reserved.

CMake version:

> cmake --version
cmake version 3.19.4

CMake suite maintained and supported by Kitware (kitware.com/cmake)

Ninja version:

> ninja --version
1.10.2

Build configuration is done with

> export FC=ifort CC=icc FFLAGS="-g -warn all -stand f08 -check -diag-error-limit 1 -traceback"
> cmake -B _build_debug -G Ninja -DCMAKE_BUILD_TYPE=Debug

Build output:

> ninja -C _build_debug
ninja: Entering directory `_build_debug'
...
[134/143] Building Fortran object src/CMakeFiles/negf.dir/complexbands.f90.o
FAILED: src/CMakeFiles/negf.dir/complexbands.f90.o src/include/complexbands.mod 
/opt/intel/oneapi/compiler/2021.1.1/linux/bin/intel64/ifort  -Isrc -Iext_system/include -Isrc/include -g -warn all -stand f08 -check -diag-error-limit 1 -traceback -g -module src/include -c src/CMakeFiles/negf.dir/complexbands.f90-pp.f90 -o src/CMakeFiles/negf.dir/complexbands.f90.o
src/complexbands.f90(278): warning #8889: Explicit declaration of the EXTERNAL attribute is required.   [ZGGEV]
    call zggev(JOBVL,JOBVR,Sdim, TA, Sdim, TB, Sdim, Ad, Bd, &
---------^
src/complexbands.f90(286): warning #8889: Explicit declaration of the EXTERNAL attribute is required.   [ZGGEV]
    call zggev(JOBVL,JOBVR,Sdim, TA, Sdim, TB, Sdim, Ad, Bd, &
---------^
src/complexbands.f90(439): warning #8889: Explicit declaration of the EXTERNAL attribute is required.   [ZHEGV]
    call ZHEGV(1,JOBV,'L',PLdim,HH,PLdim,SS,PLdim,&
---------^
src/complexbands.f90(209): remark #7712: This variable has not been used.   [E]
  subroutine complex_k(E,PLdim,Z11,Z12,kz,Z21,Cl,Cr,vf)
-----------------------^
src/complexbands.f90(389): catastrophic error: **Internal compiler error: internal abort** Please report this error along with the circumstances in which it occurred in a Software Problem Report.  Note: File and line given may not be explicit cause of this error.
compilation aborted for src/CMakeFiles/negf.dir/complexbands.f90-pp.f90 (code 1)

Seems to be an issue with the Intel compiler, which is not necessarily a problem in libnegf, therefore I just opening this issue here to document the compiler failure.

Bulid fails if "python" is not available

When built as standalone, the fypp delivered with the project uses /usr/bin/env python, which fails on Ubuntu derivates, where the standard Python3 interpreter is called python3.