I am trying to setup some cutoff in the python version but somehow I fail to find docs about

&cutoff
- 8.0 4.0
H-C 8.0 4.0
H-O 8.0 4.0
H-Zn 8.0 4.0
C-O 8.0 4.0
C-Zn 8.0 4.0
O-Zn 8.0 4.0
/

any help appreciated.

Dear Dr. Tadano, and Dr. Togo

Not sure if this the right place for this.

I have been trying to use phonopy-alm interface to fit force-constants within ALM and trasnlate them back to phonopy. I am using phonopy ALM python library for this as API.

I did two test
(1) Fit the force constants within alamode (alm part of alamode) using the force and displacement and generated dispersion within anphonon i got good expected results. This is shown as Alamode (blue lines)

(2) When i use phonopy-ALM interface the dispersion is bit different... These are magneta line

This is shown in the attached plot.

Note that the same fitting method is used for these two method with identical cutoff and other parameters. Similar issue is also occur for AlN example, where small negative frequency appear near gamma point which disappear within alamode code)..

Also it is possible to invoke phonopy-alm interface with command line?

This seems a fit strange to be and I am not able to identify the source of this difference... Any pointers will be helpful.

Thank for your time/

Thanks,
Alex

The preparation stage of AiiDA CalcJob does not allow direct use of file paths, which makes it difficult to create AiiDA scripts using ALM.
Python code is

            reg = LinearRegression(fit_intercept=False).fit(X, y)
            alm.set_fc(reg.coef_)
            alm.save_fc(filename=fname_fcs, format="alamode")

and fname_fcs is used in anphon input as

 FCSXML = {fname_fcs}

But, AiiDA can't specify fname_fcs as far as I understand. AiiDA script must be something like

 with folder.open(self.options.input_filename, 'w', encoding='utf8') as handle:
            make_alm_in(alm_param, handle=handle)

Is it possible to use the file hander instead of fname_fcs?

Probably
fc2 = alm.get_fc(1, mode='all')
can obtain all the harmonic force costants.

Though https://alm.readthedocs.io/en/develop/python-module.html?highlight=create_dataset intoduces

    with h5py.File('fc.hdf5', 'w') as w:
        w.create_dataset('fc2', data=fc2, compression='gzip')
        w.create_dataset('fc3', data=fc3, compression='gzip')

how can I make the FCSXML file from fc2?

Currently, ALM stores calculated force constants in a single XML file when executed via the command line. This approach often results in a huge XML file especially when the symmetry of the system is low and/or the higher-order (>3) force constants are included. Besides, reading a data subset from a huge XML by the boost XML parser, which is done in the anphon code of ALAMODE, is very inefficient.

I expect HDF5 library can solve these performance issues as it can store data in binary format (if I understand correctly). Many major languages, such as python, have their own interface to the HDF5 data format. This is an advantage of using HDF5.

I need to work on the following points:

1. Decide the structure of the HDF5 force constant file
2. Implement the parser for C++
3. Write a document describing the data structure

Any other points?

I notice that new version 1.1.0 of alamode can support applying rotational symmetry to force constant. However, in python API , say ALM, this function is not implimented. The responsible module is ALM.set_constraint(). Could you give me some advice to modify the code?

This isn't an error, but the warning message

libtinfo.so.6: no version information available (required by bash)

is shown when $CONDA_PREFIX/lib is defined as

export LD_LIBRARY_PATH=$CONDA_PREFIX/lib:$LD_LIBRARY_PATH

In my case, libtinfo.so.6 is in /usr/lib/x86_64-linux-gnu/. So, if /usr/lib/x86_64-linux-gnu/ is placed before $CONDA_PREFIX/lib, the warning message doesn't appear.
Ref.
rstudio/rstudio#8278

Writing tests using API is not the issue here. Some test of command line interface is needed to ease the check. The last one can be shared with the test for API. Any idea?

• Input files: alamode/example/Si is OK for the first example
• A short document how to run and test: can be extracted from ALAMODE tutorial with small modification
• A shell script to run the test
• Parsers of .fcs and .xml files
• A (python?) script to assert data in .fcs and .xml files HDF5.

I found a strange behavior of the alamode API when the alm and numpy modules were loaded in the same project.

For example, the script https://github.com/ttadano/ALM/blob/develop/example/Si_fitting_external.py worked as expected in a Linux environment (tested on ubuntu), but it stopped in MacOS at the following line:

#Perform fitting
fc = np.linalg.lstsq(amat, bvec, rcond=1.0e-15)

Strangely, in another MacOS environment, the above numpy function returned some values (fc), but the returned value changes every time I run the script.

In both cases, I used an anaconda version of python3 installed via pyenv (anaconda3-5.0.1), and the alamode API was built by using GCC installed via homebrew (gcc: stable 7.3.0). Probably, this strange behavior results from the incompatibility between GCC and anaconda (MKL?).

I found the issue can be avoided by switching off the OpenMP option in GCC when building the alm API or by setting MKL_NUM_THREADS=1 before running the script.

I've been trying to understand the irreducible representation of force constants better and was wondering if there is some way to list which IFC indices should be identical?

Say we took the setup from the fcsxml example where we calculate the force constants in what ALAMODE calls the origin format which as noted by the docstrings is reducible. What I'm after is basically:

• (1x, 2y), (1y, 3z) and (2y, 5z) are symmetrically identical
• (2z, 3y), (1z, 2x),... are symmetrically identical
• etc.

EDIT: Maybe this is an ill-posed question since the symmetry operations just tell you how a pair of IFC are related (e.g. $\phi_{12} = a\phi_{13} + b \phi_{23}$) not which values in the supercell representation are identical.

The linear combinations that form the irreps would also be useful information.

Thanks!