judftteam / aiida-fleur

AiiDA plugin of the high-performance density functional theory code FLEUR (www.judft.de) for high-throughput electronic structure calculations.

FLEUR with AiiDA

This software contains a plugin that enables the usage of the all-electron DFT FLEUR code with the AiiDA framework.

Developed at Forschungszentrum Jülich GmbH

Compatibility matrix

FLEUR Plugin	AiiDA CORE	Python	FLEUR/XML file version
v2.0.0 < v3.0.0			/<=0.37
v1.2.0 < v2.0.0			/<=0.35
v1.0.0 < v1.2.0			/<=0.33
< v0.6.3			/<=0.31

Documentation and User Support

Hosted at http://aiida-fleur.readthedocs.io/en/develop/index.html. For other information see the AiiDA-core docs or http://www.flapw.de.

Users can post any questions in the Fleur user forum

For bugs, feature requests and further issues please use the issue tracker on github of the aiida-fleur repository.

License:

MIT license. See the license file.

How to cite:

If you use this package please consider citing:

J. Broeder, D. Wortmann, and S. Blügel,
Using the AiiDA-FLEUR package for all-electron ab initio electronic structure
data generation and processing in materials science,
In Extreme Data Workshop 2018 Proceedings, 2019, vol 40, p 43-48

Comments/Disclaimer:

The plug-in and the workflows will only work with a Fleur version using xml files as I/O, i.e >v0.27.

Contents

1. Introduction
2. Installation Instructions
3. Code Dependencies
4. Further Information

Introduction

This is a python package (AiiDA plugin, workflows and utility) allowing to use the FLEUR-code in the AiiDA Framework. The FLEUR-code is an all-electron DFT code using the FLAPW method, that is widely applied in the material science and physics community.

The plugin :

The Fleur plugin consists of:

1. A data-structure representing input files and called FleurinpData.
2. inpgen calculation
3. FLEUR calculation
4. Workchains
5. utility

Workchains in this package:

workflow entry point name	Description
fleur.scf	SCF-cycle of Fleur. Converge the charge density and the Total energy with multiple FLEUR runs
fleur.eos	Calculate and Equation of States with FLEUR (currently cubic systems only)
fleur.dos	Calculate a Density of States (DOS) with FLEUR
fleur.band	Calculate a Band structure with FLEUR
fleur.relax	Relaxation of the atomic positions of a crystal structure with FLEUR
fleur.init_cls	Calculate initial corelevel shifts and formation energies with FLEUR
fleur.corehole	Workflow for corehole calculations, calculation of Binding energies with FLEUR
fleur.dmi	Calculates Dzyaloshinskii–Moriya Interaction energy dispersion of a spin spiral
fleur.ssdisp	Calculates exchange interaction energy dispersion of a spin spiral
fleur.mae	Calculates Magnetic Anisotropy Energy

See the AiiDA documentation for general info about the AiiDA workflow system or how to write workflows.

Utility/tools:

filename	Description
Structure_util.py	Constains some methods to handle AiiDA structures (some of them might now be methods of the AiiDA structureData, if so use them from there!)
merge_parameter.py	Methods to handle parameterData nodes, i.e merge them. Which is very useful for all-electron codes, because instead of pseudo potentialsfamilies you can create now families of parameter nodes for the periodic table.
read_cif.py	This can be used as stand-alone to create StructureData nodes from .cif files from an directory tree.

Utility and tools, which are independend of AiiDA are moved to the masci-tools (material science tools) repository, which is a dependency of aiida-fleur.

Command line interface (CLI)

Besides the python API, aiida-fleur comes with a builtin CLI: aiida-fleur. This interface is built using the click library and supports tab-completion.

To enable tab-completion, add the following to your shell loading script, e.g. the .bashrc or virtual environment activate script:

eval "$(_AIIDA_FLEUR_COMPLETE=source aiida-fleur)"

the main subcommands include:

data: Commands to create and inspect data nodes
    fleurinp   Commands to handle `FleurinpData` nodes.
    parameter  Commands to create and inspect `Dict` nodes containing FLAPW parameters
    structure  Commands to create and inspect `StructureData` nodes.
launch: Commands to launch workflows and calcjobs of aiida-fleur

    banddos          Launch a banddos workchain
    corehole         Launch a corehole workchain
    create_magnetic  Launch a create_magnetic workchain
    dmi              Launch a dmi workchain
    eos              Launch a eos workchain
    fleur            Launch a base_fleur workchain.
    init_cls         Launch an init_cls workchain
    inpgen           Launch an inpgen calcjob on given input If no    code is...
    mae              Launch a mae workchain
    relax            Launch a base relax workchain # TODO final scf    input
    scf              Launch a scf workchain
    ssdisp           Launch a ssdisp workchain

plot: Invoke the plot_fleur command on given nodes

workflow: Commands to inspect aiida-fleur workchains and prepare inputs

for example to launch an scf workchain on a given structure execute:

$ aiida-fleur launch scf -i <inpgenpk> -f <fleurpk> -S <structurepk>

the command can also process structures in any format ase can handle, this includes Cif, xsf and poscar files. In such a case simply parse the path to the file:

$ aiida-fleur launch scf -i <inpgenpk> -f <fleurpk> -S ./structure/Cu.cif

Installation Instructions

From the aiida-fleur folder (after downloading the code, recommended) use:

$ pip install .
# or which is very useful to keep track of the changes (developers)
$ pip install -e .

To uninstall use:

$ pip uninstall aiida-fleur

Or install latest release version from pypi:

$ pip install aiida-fleur

Test Installation

To test rather the installation was successful use:

$ verdi plugins list aiida.calculations

   # example output:

   ## Pass as a further parameter one (or more) plugin names
   ## to get more details on a given plugin.
   ...
   * fleur.fleur
   * fleur.inpgen

You should see 'fleur.*' in the list

The other entry points can be checked with the AiiDA Factories (Data, Workflow, Calculation, Parser). (this is done in test_entry_points.py)

We suggest to run all the (unit)tests in the aiida-fleur/aiida_fleur/tests/ folder.

$ bash run_all_cov.sh

---

Code Dependencies

Requirements are listed in pyproject.toml

most important are:

• aiida_core >= 2.0
• lxml
• ase
• masci-tools

Mainly AiiDA:

1. Download from www.aiida.net -> Download
2. install and setup -> aiida's documentation

Easy plotting and other useful routines that do not depend on aiida_core are part of the masci-tools (material science tools) repository.

For easy plotting we recommend using 'plot_methods' from masci-tools, which are also deployed by the 'plot_fleur(<node(s)>)' function.

Further Information

The plug-in source code documentation is here. also some documentation of the plug-in, further things can be found at www.flapw.de. Usage examples are shown in 'examples'.

Acknowledgements

Besides the Forschungszentrum Juelich, this work is supported by the European MaX Centre of Excellence 'Materials design at the Exascale' MaX funded by the Horizon 2020 EINFRA-5 program, Grant No. 676598 and under grant agreement No. 824143. This work is further supported by the Joint Lab Virtual Materials Design (JLVMD) of the Forschungszentrum Jülich.

For this work essential is AiiDA, which itself is supported by the MARVEL National Centre for Competency in Research funded by the Swiss National Science Foundation.

aiida-fleur's Issues

Changing floats in inp,xml

Fleur is not capable of dealing with floats like 1e-5. (Change on the fleur side?)

Therefore, fleurinpmodifier has to prevent to write the floats in that form.

A workaround so far is to give the modifier methods a string instead of a float.

queue limits are a problem try to deal on workflow level

as long AiiDA does not have this feature.
Idea: write workflows in a style that you can count/know the numbers of subworkflows currently running. Then maybe do not submit any sub workflow as long as the number did not decrease.

EOS workflow should do a Birch–Murnaghan Fit instead of parabola

output node of fleur_scf_wc content

currently the content is:
{
"distance_charge": 7.994e-07,
"distance_charge_all": [
1.5084338969,
1.4243665042,
0.1836076641,
0.0446805799,
0.0346745327,
0.0057244928,
0.0024850833,
0.0009573742,
3.56938e-05,
2.84684e-05,
7.994e-07
],
"distance_charge_units": "me/bohr^3",
"iterations_total": 11,
"last_calc_uuid": "80ba2838-1607-40c8-b74c-7ce6b29c8577",
"loop_count": 1,
"material": "Kr",
"successful": true,
"total_cpu_time": "",
"total_cpu_time_units": "hours",
"total_energy": -2790.4780331553,
"total_energy_all": [
-2790.4763685438,
-2790.4764415277,
-2790.4777323388,
-2790.4779329712,
-2790.4779668154,
-2790.478013726,
-2790.4780279666,
-2790.4780279116,
-2790.4780325126,
-2790.4780331659,
-2790.4780331553
],
"total_energy_units": "Htr",
"warnings": [],
"workflow_name": "fleur_scf_wc",
"workflow_version": "0.1.0"
}

What is missing?

prop_dir has to be set automatically based on q_vectors in DMI and SSDisp workchains

Now a user needs to specify prop_dir parameter in workchain parameters to explicitly show what kind of q vectors will be used in a further forceTheorem step.
However, a user also must specify the full set of q-vectors to calculate. Thus, prop_dir can be calculated basing on the given q-vectors mesh.

Fleur_convergence workflow should always terminate and store the error if calculation fails

currently if a calculation inpgen or fleur fails in the scf workflow, the workflow will fail at a later point if it tries to access these files. Therefore make checks if calculation finished else terminate in a controlled way.

Implement scf workchain failure handling

Scf workchian should probably use RestartBaseWorkChain to handle some failures.

Examples:

1. Retry: try to converge restarting from this calculation, i.e. if scf is ended and did not lead to convergence, resubmit it.
2. ggf handle bad convergence behavior, delete broyd (do not copy), i.e. try to change mixing scheme, factors...

PYTHONPATH not there

if PYTHONPATH is not defined in the os.env the plugin does not work.

this should not be
-> make a try and catch error

submit calculation in parallel for eos workflow

Return node producing workfunctions should produce the result node

the right way would be to produce the output node in the workfunction not duplicate the output nodes to get the links in the aiida graph right. (and without extra additional helper nodes) The problem is that currently the output node is produced from the context of the workchains. and that a workfunction can only have aiida objects as input, meaning we cannot make it simply a method of workchain class.

An other solution (maybe better) might be to make this function a method of the workchain classes and produce the connecting links and 'workfunction node' by hand. This might be less work and more robust of the workchain.

maybe it is also worth to wait a bit and see if the aiida workfunction policy changes.

Kpoints data usage of inpgen

Inpgen should be able to use an AiiDa kpoints node. This should be done with care.

There are several cases.
Kpoints given as a list - > override kpoint list in inp.xml at the end. at parse time before creating new fleurinp.
kpoints given as mesh (3,3,3) -> set div1,2,3
kpoints given as path -> ?

fleur parser fails in certain magnetic cases with soc

this might only happen if calculation was failed. i.e there are some checks missing.

the parser fails with

'''
charge_density1 = get_xml_attribute(charge_densitys[0], distance_name)
| IndexError: list index out of range
'''
So charge density is prob empty, because it was not in the out.xml file....

Output node of Fleur calculations

What is missing in the output node of a Fleur calculations.
the output node should be standalone enough to provide the follow information:

1. What was calculated and how?
2. Basic run information, was everything fine, or what went wrong
3. Important output (results, but also sanity check values)

Things to add:

1. More units fields should be added.
2. Warnings, Hints and Errors should be parsed.
3. Other physical quantities?

• total magnetic moment cell last iteration
• total spinDependentCharges last iteration
• cell_volume because one often likes to give something per volume
• reciprocal cell_volume ?
• structure information, add Formula, Elements, spacegroup, dimensions, .... which is redundant but helpful information....
• parse atom types information
• fleur mode since in case of other fleur modes, output is different,
• 'number_valence_electrons' : X, 'number_core_electrons' : X

Current content:

{
    "bandgap": 0.0005412425,
    "bandgap_units": "eV",
    "charge_den_xc_den_integral": -111.8771323535,
    "charge_density1": 1.871e-07,
    "charge_density2": 1.803e-07,
    "creator_name": "fleur 31",
    "creator_target_architecture": "GEN",
    "creator_target_structure": " ",
    "density_convergence_units": "me/bohr^3",
    "end_date": {
        "date": "2020/11/30",
        "time": "12:19:18"
    },
    "energy": -69269.59001367,
    "energy_core_electrons": -1444.2206331035,
    "energy_hartree": -2545.6104803879,
    "energy_hartree_units": "Htr",
    "energy_units": "eV",
    "energy_valence_electrons": -25.3255485869,
    "fermi_energy": 0.3702091804,
    "fermi_energy_units": "Htr",
    "force_largest": 0.0,
    "kmax": 5.0,
    "magnetic_moment_units": "muBohr",
    "magnetic_moments": [
        2.3339400372,
        2.3339362223
    ],
    "magnetic_spin_down_charges": [
        6.2618479986,
        6.2618511289
    ],
    "magnetic_spin_up_charges": [
        8.5957880358,
        8.5957873512
    ],
    "number_of_atom_types": 2,
    "number_of_atoms": 2,
    "number_of_iterations": 20,
    "number_of_iterations_total": 20,
    "number_of_kpoints": 286,
    "number_of_species": 1,
    "number_of_spin_components": 2,
    "number_of_symmetries": 48,
    "orbital_magnetic_moment_units": "muBohr",
    "orbital_magnetic_moments": [],
    "orbital_magnetic_spin_down_charges": [],
    "orbital_magnetic_spin_up_charges": [],
    "output_file_version": "0.27",
    "overall_charge_density": 7.73e-08,
    "parser_info": "AiiDA Fleur Parser v0.3.0",
    "parser_warnings": [],
    "spin_density": 3.593e-07,
    "start_date": {
        "date": "2020/11/30",
        "time": "12:10:16"
    },
    "sum_of_eigenvalues": -1469.5461816904,
    "title": "A Fleur input generator calculation with aiida",
    "unparsed": [],
    "walltime": 542,
    "walltime_units": "seconds",
    "warnings": {
        "debug": {},
        "error": {},
        "info": {},
        "warning": {}
    }

fleur_scf_wc resource choice does not always respects maximum resources given... especially for large systems...

fleur_scf_wc resource choice does not always respects maximum resources given... especially for systems larger 20 atoms...

system with 31 atoms and 50 kpoints ends up with 24 nodes, 2 mpi 12 openmp as resources while maximum resources of two nodes was given...

which makes no sense...

Fleurinp get_structure data

There is a bug in the get_structure data methods from fleurinpdata. for compounds...
s.get_formula() gives wrong result. Be2W4 -W4 therefore Be is missing...

General finding the schema file

So far we look in the current dir, and walk the PYTHONPATH, or hack it.

One should still think about a more general solution. Especially after the plugin is made pip install able, because then it should be clear where the schemas are.

Fleur plugin Walltime calculation has a bug

Differences of dates and times have to be done differently. currently over the new year we got negative numbers.

fleur plugin/ parser does not mark certain Fleur calculation failures as failed.

This might cause certain problems down the line, if routines deal with calculations they assume FINSIHED.

If outfile is incomplete. FLEUR calculations should be marked as FAILED.

inpgen returns data type instead of fleurinp data

There seems to be a bug sometimes in creating fleurinp data sometimes. Through it might or might not have todo with the small changes of me in aiida_core, escape_for_bash and so on. Investigate.

Implement Fleur failure handling in FleurBaseWorkChain

All the text below is extracted from scf workchain:

• handle out of walltime (not one interation run) abort, tell user to specifi mor resources, or different cutoffs

• handle fleur error fermi level convergence
  restart fleur with more tempertature broad or other smearing type (gauss = T)

• qfix needed, restart fleur with Qfix

• differ errors, solving dirac equation
  abort, or restart with different parameters

• muffin tin radii overlap, restart with smaller MT

• wrong command line switches
  hdf, -h , ... restart with the right one, or abort

• something about kpt grid
  abort, or restart with increased number /different number of kpoints

• wrong parallelisation
  abort or restart with right parallelisation

ALL FLEUR current (07.2017) ERROR hints:

1. • nsh greater than dimensioned | increase nmax in jcoff2'
2. • Fermi-level determination did not converge| change temperature or set input = F
3. • Too low eigenvalue detected | If the lowest eigenvalue is more than 1Htr below the lowest energy parameter, you probably have picked up a ghoststate
4. • e >= vz0 | Vacuum energy-parameter too high
5. • kidx /= stars | something wrong with kmxxc_fft or nxc3_fft
6. • Wrong size of matrix | Your basis might be too large or the parallelization fail or ??
     If no LO's are set skiplo must be 0 in enpara
7. • You compiled without support for HDF5. Please use another mode
8. • Your HDF5 library does not support parallel IO
9. • Use -mpi or -hdf instead
10. • Consider setting 'autocomp = false' in apwefl.
11. • Film setup not centered" | The z = 0 plane is the center of the film
      setcore_bystr |
      expnadconfig |
12. • File not readable:"//filename | FLEUR wants to read a file that is not present
13. • Both inp & inp.xml given | Please delete one of the input files or specify -xml to use inp.xml
      inp.xml not found | You gave the -xml option but provided no inp.xml file
14. • No input file found | To use FLEUR, you have to provide either an 'inp' or an 'inp.xml' file in the working directory
15. • Use a supercell or a different functional
16. • MPI parallelization failed | ou have to either compile FLEUR with a parallel diagonalization library (ELPA,SCALAPACK...) or you have to run such that the No of kpoints can be distributed on the PEs

AiiDA related:

1. • Submissionfailed:
     SSHException: SSH session not active -> wait and relaunch

Scheduler/HPC related:

1. • scf scheduler needs other things than torque, try to provide both as default
     that they both work
2. • ran out of walltime with out a single interation done. (killed by scheduler)
     -> terminate, because the user should specify more recources or in the
     calculation something is fishy
3. • bumped by scheduler joblimit. reached
     -> think about something here and retry
4. • no space/files left on remote machine.
     -> also terminate, tell user to delete stuff
5. • not eneough memory, (killed by scheduler, or some fleur error)
     -> can we do something here

NOTE:

• If some wc parameters were adjusted, they have to be passed to higher-level workchain somehow. Otherwise we end up in a situation when SCF workchain submits not correct parameters for the second iteration again.

fleur_scf_wc output node total wall time units right?

Wall time units right?
In the fleur_scf_wc output node it says the wall time unites are hours, but it looks from the test more like seconds....

find schema

the plugin finds the schema file, but does not reconize that the version is right.

set_kpointsdata has to be moved to FleurinpModifier

FleurinpData.set_kpointsdata() is a CalcFunction which stores a new FleurinpData in the database.
It does not actually change FleurinpData and should be moved to the FleurinpModifier

Connect Fleur Bandstructure files to AiiDA Bandstructure Node

fleur_initial_state workflow behavior if not all references are given...

Currently the worklfow will fail late in collect results, if not all references are (wrongly) given... this should be checked early on. Or work also with partial references...

reduce size of pypi install

As of today, after installing aiida-fleur 0.6.2 via pip, the package takes up 38MB on disk space, 36MB of which is in tests/files (e.g. local_exe (??) ).

Can this be improved?

A few other packages for comparison

1.4M	paramiko
1.5M	setuptools
3.5M	tornado
5.8M	IPython
8.2M	sqlalchemy
38M  	aiida-fleur

please fix dependencies

• aiida-core >= 0.12.0[atomic_tools] results in always selecting aiida-core 0.12.2 (even though 0.12.3 is out). I believe the correct syntax should be: aiida-core[atomic_tools] >=0.12.0,<1.0
• please don't fix the numpy version as it is fixed by aiida-core

Maintenance of xpaths (and therefore whole plugin) should be easier

So far all xpaths are hardcodes somewhere in the plugin, therefore the plugin needs a bit of maintenance.

A solution to extracted all these paths from the schema file, which should not take so much time (or be done only once and installation and stored in a file or so) should replace this.

ParameterData output node for inpgen calculation

Connect Fleur DOS files to AiiDA DOS nodes

Add switch to workchains to run them in serial (no calcualtion submited in parallel)

This might be helpful, if one runs fleur on his computer and not an hpc resource. I am not sure how several openpi fleur versions started on the same machine influence each other.

fleur_corehole to many unnecessary nodes

In the current design the workflow creates a list of pairs with atom types and coresetup stuff and then creates calculations out of that list. Which results into the moved of the supercell x times for an atom type, where x is the number of core levels to calculate.

A better design would avoid these double nodes. But since the workfunction for the parameter creating has to be called anyway the nodes saved is only x-1. per atom type.

Work Flow for relaxation

Dear Jens,

I am wondering if anyone is working on the structural relaxation workflow, if not I would like to make one. There seems to be a WF with old workflows but none in the new ones. Let me know so that I can start to work on it.

Regards
Vamshi

FleurinpData should have some verdi commands

u.a to view the inp.xml file

FLEUR parser wrong values for certain keys...

Check the keys:

• nspecies...
• Walltime (should not be negative. probable because a day was in between and date handling is wrong)

fleurinp methods buggy

The methods get_structure_data and get_kpoints_data_nwf currently fail with

XMLSyntaxError: Element 'fleurInput', attribute 'fleurInputVersion': [facet 'enumeration'] The value '' is not an element of the set {'0.28', '0.27'}.

this is after the change to the new xml schema management. And should be fixed. Also the work functions of these should be working again.

pypi package broken

pip install aiida-fleur
   Downloading https://files.pythonhosted.org/packages/f7/c7/4824f9f3a391f59658a258a0ba01265d81f54bca574f8af645e956bfe9c4/aiida-fleur-0.6.0.tar.gz (155kB)
        Complete output from command python setup.py egg_info:
        Traceback (most recent call last):
          File "<string>", line 1, in <module>
          File "/tmp/pip-install-FIzeJI/aiida-fleur/setup.py", line 9, in <module>
            with open('setup.json', 'r') as info:
        IOError: [Errno 2] No such file or directory: 'setup.json'

In order to fix this, please add a MANIFEST.in with content

include setup.json

see e.g. https://github.com/cp2k/aiida-cp2k/blob/master/MANIFEST.in

test for all workchains, options interface change

Since the interface how options are provided to the base scf workchain was changed, all workchains where adapted, but there are still some issues left. u.a there are errors if the 'custom_scheduler_commands' option is not provided....

corehole wc output node

Currently a borehole wc output node alone does not tell you much. Post ideas what should be added to the node here.

fleur_scf_wc finished stays in None if inpgen failed..

A clear termination is needed...

fleur_eos_wc bulkmodulus has a factor in it...

How the plugin uses the structure needs to be overworked

So far the plugin assumes atoms at sites and does not care about weights or vacancies. There should be checks for the whole functionality of structureData and warnings/ignores if Fleur is unable to use certain structureDatas. Otherwise the Prov between different codes is at risk...

Host documentation online (read the docs)

Error in parsing the inp.xml

Hi,

I am trying to run the Cu_bulk example and I get the following parsing error

XMLSyntaxError: Element 'relPos', attribute 'label': The attribute 'label' is not allowed. (line 0)

I am using MaX-R2 version of fleur and the latest aiida-fleur. Am I missing something here?

Documentation of Fleurinp data (code and real)

Fleurinp Data should also initialize with relative paths for files

Fleurinp modifier functions

The functions of Fleurinp modifier have to be overworked.

In detail for noco, lda+U, los and other features.
Also the standard xml methods

Bug SCF for magnetic systems

currently for magnetic systems we store the overall_charge and charge_density1 and charge_density2.

maybe this should become a list under one key charge_density.
currently scf fails because it accesses charge_density. also the numbers do not seem right, check the parsing.

FleurCalculation, verdi calculation inputcat should show inp.xml

this was not possible before, but there was an adjustment in AiiDA. Maybe know it is.

(You have to set some variable in the plugin, but in the case of a fleurcalculation this can only be set AFTER the fleurinp data node was connected to the calculation.

Workflows ouput node is not directly connected to the last calculations, subworklfow outputs...