ASE Workshop
Chalmers, November 19-22, 2019

Jens Jørgen Mortensen, CAMd, Department of Physics, Technical University of Denmark

• ASE-1.0
• ASE-2.0
• ASE-3.0
• The future

It all started with Dacapo (a Fortran USPP plane-wave DFT code):

S. R. Bahn and K. W. Jacobsen, An object-oriented scripting interface to a legacy electronic structure code, Comput. Sci. Eng., Vol. 4, 56-66, 2002, https://dx.doi.org/10.1109/5992.998641

• Use Python scripts
• Write input to netCDF file
• Communicate with Fortran code through a socket
• Read output from netCDF file

from Simulations.Dacapo import *
mysim = Simulation()
mysim.bands = ElectronicBands(9)
mysim.config = ListOfAtoms(
    atoms=[Atom(Mg_GGA, Vector([0, 0, 0]))],
    unitcell=BravaisLattice(
        [[-1.425, 1.425, 1.425],
         [1.425, -1.425, 1.425],
         [1.425, 1.425, -1.425]])
mysim.plancut = PlaneWaveCutoff(340)
mysim.Execute()

Problems with ASE-1:

• Other codes (from Dacapo to ASAP)
• Millions of atoms
• Too many objects?

After many discussions and meetings, ASE-2.0 was created from scratch.

• Simulation -> ListOfAtoms
• ListOfAtoms.GetCartesianPositions() -> numpy.ndarray

ASE-3.19: ~40 calculators

Problems with ASE-2:

• Weird unit system
• Weird naming convention (files and function names)
• ListOfAtoms -> Atoms
• ...

Rewrite from scratch or adapt current code base?

1. Rewrite
2. Ask for permission later

279 contributors so far ...

• A beta version of the new ASE-3.0 will be used for the electronic structure course at CAMd. (10 Jan 2008)
• The new Sphinx page is now up and running! (2 Apr 2008)
• CAMd/Cinf will do a "doc-sprint" from 9 to 16. (17 Apr 2008)
• ASE version 3.0.0 released (13 November 2008).
• ASE has reached revision 1000 (16 July 2009).
• The source code is now on GitLab (18 September 2015).
• Reference paper in J. Phys. Condens. Matter: The Atomic Simulation Environment | A Python library for working with atoms (7 June 2017).

• Web-page: Home-made thing -> MoinMoin wiki -> Sphinx + Python scripts.
  • https://wiki.fysik.dtu.dk/ase/ -> https://a.se/ ?
• Testing: home-made framework:
  • After every commit: GitLab-CI
  • Every night: VASP + GPAW
• Numeric -> numarray -> numpy, scipy
• Python 1.x -> 3.8
• CVS -> SVN -> Git
• spglib?, C-code?

Thank you!

Let's do it right:

• code review
• documentation
  • low level: docstrings
  • high level: how to use it
• tests
• release notes
• remove old way of doing it

Think of a contribution like a puppy: you might view it as this cute,
wonderful thing you're giving me while I'm looking at it as over a
decade of feeding, walking, and vet bills.

-- Brett Cannon

Not everything is a good fit for ASE

Alternatives:

• Code snippets in a tutorial plus a few helper functions added to ASE.
• Its own thing: PyPI + GitLab + ReadTheDocs + JOSS

• less is more
• so nothing at all could be a good solution
• if your thing has documentation and tests then it might be a good fit for ASE
• if it doesn't, we don't want it
• ... unless you are from Denmark

Merge requests: 1245 merged, 195 closed and 47 open.

https://xkcd.com/1822/

http://trac.fysik.dtu.dk/projects/ase

Keeping ASE going is more work that most people think!

• Review and merge MR's
• Fix issues
• Mail-list support
• Find and nudge the expert that knows something

There are many bigger projects necessary in order to keep ASE moving in the right direction. Example: Calculators!

"If you have found ASE to be useful in your work, research or company, please consider making a donation to the project commensurate with your resources. Any amount helps! All donations will be used strictly to fund the development of ASE’s open source software, documentation and community."

• ASE will also be in good shape two years from now
• Generate goodwill thanks to your appearance in the list of sponsors
• Influence the work of sponsored developers so that ASE continues to fit your needs
• Your favorite electronic structure code will be supported and part of the daily tests.

A simple example from ASE's code:

import re

def strip_number(s):
    """Split string to number and the rest.

    >>> strip_number('2Au')
    2
    """
    m = re.match('[0-9]*', s)
    return int(m.group() or 1), s[m.end():]

x = strip_number('12Cu') * 2.5

See also: https://gitlab.com/ase/ase-workshop-discussion/issues/2

import re
from typing import Tuple

def strip_number(s: str) -> Tuple[int, str]:
    m = re.match('[0-9]*', s)
    return int(m.group() or 1), s[m.end():]

x = strip_number('12Cu') * 2.5

$ mypy hint.py
hint.py:6: error: Item "None" of "Optional[Match[str]]"
           has no attribute "group"
hint.py:6: error: Item "None" of "Optional[Match[str]]"
           has no attribute "end"
hint.py:8: error: Unsupported operand types for * ("Tuple[int, str]"
                                                   and "float")
Found 3 errors in 1 file (checked 1 source file)

import re
from typing import Tuple

def strip_number(s: str) -> Tuple[int, str]:
    m = re.match('[0-9]*', s)
    assert m is not None
    return int(m.group() or 1), s[m.end():]

x = strip_number('12Cu')[0] * 2.5

$ mypy hint.py
Success: no issues found in 1 source file

• we can do it gradually

• it's documentation that mypy can check for us

• code is read a lot more than written

• make it easier for new contributors

• improved editor-completion

• find bugs for us

• lead to simpler code

• wait for ASE to drop Python 3.5:

  ...
  number = []  # type: List[int]
  number: List[int] = []
  

• ASE has been rewritten from scratch twice - I hope we will not need a 4.0 rewrite
• Please donate to ASE - somehow
• Type hints are great!