tjsmart/zfs-code

Zero Field Splitting Code Repository

zfs-code's Issues

Adding non-root option to FFTW_install.sh

Need to better handle 'make install' for fftw library.

• Add check to see if user has root permission.
• Add a prefix option (for non-root installation) so that FFTW_install.sh will configure as:
  ./configure --prefix=
• Perhaps default for non-root installation is to install fftw to ~/.fftw-. (If this is done, also update configure to add this directory to list of directories to check for fftw under.)

Improving input files

Switch to using namelists similar to QE. A minimum working example is demonstrated below.

Here is a minimal fortran code

cat > min.f90 << EOF
program min
    implicit none
    integer     :: a
    logical     :: flag
    real        :: gamma
    namelist /input/ a, gamma, flag

    open(10, file="file.in")
    read(10, nml=input)
    close(10)
    print *, "a = ", a
    print *, "flag = ", flag
    print *, "gamma = ", gamma

end program min
EOF

Compile the program

gfortran min.f90

Create an input file

cat > "file.in" << EOF
&input
a = 2,
flag = .true., gamma = 3.141520
/
EOF

Run the calculation

./a.out

I/O needs to be improved

faster i/o (binary formatting) and improved interfacing with QE

Change main_mpi

Instead of main_mpi calling, it would be better if main_mpi returned to main myloop and such and then called main_inner?

Configure does not include fftdir in make.inc

Configure needs to link fftdir in make.inc such as:
flags_fft = \
-lfftw3 \
-I$fftdir \
-L$fftdir/../lib
Or something of this sorts

Changing output

output should be changed in a few ways:

• print mpi details
• print details of what machine it was run on?

Perhaps other things will come to mind

Bash scripts error need to be directed to stderr (&2)

Need to redirect error messages to stderr (&2).
Currently: echo "Error: ..."
Instead: echo "Error: ..." >&2

Improving efficiency of sum_G

Currently G_ab - d3_ab is computed for each set of bands despite the fact it is identical for each set of bands -- it should be computed only once

perhaps called by main and read into zfs_calc

MPI error, likely in mpi_reduce

Error when running the code currently with mpi on kairay:

Fatal error in PMPI_Reduce: Invalid root, error stack:
PMPI_Reduce(2334): MPI_Reduce(sbuf=0x7ffe6f51eab0, rbuf=0x7ffe6f51f0e8, count=1, MPI_DOUBLE_COMPLEX, MPI_SUM, root=32686, MPI_COMM_WORLD) failed
PMPI_Reduce(2235): Invalid root (value given was 32686)

mpi_reduce is called at the end of the 'mpi_routine' subroutine in the 'main_mpi' module.

1. check different mpi compilers and machines for consistency of issue.
2. double check that mpi_reduce was used correctly here (this issue is strange as earlier versions of the code did not experience this issue).

Debugging ZFS code

After meeting today with Prof. Ping & Junqing, it is clear that the code needs more testing on two fronts:

1. Testing more sine-wave examples, including higher-frequency elements (up-to and above 6) and more complex forms.
2. For realistic QE example, take a step back and do a simple H2 molecule and then perhaps simple bulk semiconductors (Si) and metals (Al). Compare results of psi(r) against QE output with QE's postprocessing.
3. How does energy cutoff effect fft/direct convolution results.
4. How does volume of the cell effect results?