Workshop discussions 2024

[ml-evs]

Here are the discussion topics suggested for this year's workshop. We will cycle through them throughout the week:

• Trajectories
• Collections + providing data for training ML potentials
• Properties
  • cheminformatics (SMILES), defects, materials discovery, magnetic, spectra, ...
• OPTIMADE jsonl
• workflows
• OPTIMADE python tools
• Future of OPTIMADE

[gjc29]

DISCUSSION OF NON-EQUILIBRIUM AND MAGNETIC PROPERTIES ON TUESDAY

Non-equilibrium structures i.e. atomic potentials (a few equilibrium) (Eric) Energy (and if it is a difference could be defined wrt a reference value stored as a database ID) Forces (on each atom, same order as list of atoms, questions: which force for which atom, same order, what if primitive vs simulation cell) Stress tensor (could be full 3x3 or Voigt notation), (virial (normalized by volume or not?)) Need parameters on how collected: e.g. DFT parameters Could also do: local energy, local electronic density, local orbitals overlap

Recommendation: Make a group of interested participants, and github repository, link to trajectories, stability

Discussion: Rickard: could store in trajectory endpoint (with just a single entry)

To do: sub-sub group to consult trajectory working group whether a “trajectory” should be a “collection” of uncorrelated snapshots?

Define a collection as a set of trajectories all used in training a given machine learning potentials One of those trajectories is a set of structures (could be time evolved, nested sampling) Trajectory name is confusing (time evolution) (and delta between steps)→ ensemble ?

Further discussion: trajectory now favored data structure, but its defnition need to be clear that can be unordered.

Magnets: (Oskar)

Properties per atom: Jij magnetic exchage Zeeman contribution, Heisenberg contribution H per atom Properties per crystal Total energy (B^2/2mu) Curie temperature Magnetic susceptibility / permeability (tensor) Magnetic coercivity Ferromagentic, magentic field, magnetic polarization Magnetic anisotropy constant Superconductor (Meisner effect) Micromagnetic (large length scales e.g. vortices, domains) Piezomagnetism

DFT, was spin orbit coupling turned on?

Recommendation: Make a group of interested participants, and github repository

[gjc29]

DISCUSSION OF NON-EQUILIBRIUM AND MAGNETIC PROPERTIES ON WEDNESDAY

Create /properties endpoint that contains

cheminfo stability electronic magnetic thermo phonon transport optical defect mechanical superconductivity radiological chemical acoustic

may need aliases from e.g. optical band gap to electronic band gap

then discuss electronic that could contain _bandgap [scalar: units eV] dos [list of 2D points: units 1/eV] _dosfermi [list of 2D points: units 1/eV]

[gjc29]

Following further discussion it was agreed that snapshots used for training ML potentials should be stored within a trajectory if they have the same composition, and collection for changing composition. The trajectories working group should include a statement about how trajectories do not have to be indexed by time-step, but could be any arbitrary index (e.g. random number seed or nested sampling progression), and properties working group should formulate a paragraph with best use cases.

[merkys]

@gjc29 and I have shortlisted the following ("least controversial") properties from MPDS property list for inclusion into OPTIMADE:

No category:

LPFcode_1	LPFcode_2	LPF_sortkey	Symbol	SI unit	Subtitle	Datasheet	Property
P0D	N0	1	D	Mg m‰…	density	density	density

Electronic and electrical properties:

LPFcode_1	LPFcode_2	LPF_sortkey	Symbol	SI unit	Subtitle	Datasheet	Property
P5B	N1B	1	E¿	eV	energy band structure	energy at Fermi level	energy at Fermi level
P5C	N1C	1	N(E¿)	eV‰ƒ	electron density of states	electron density of states at Fermi level	electron density of states at Fermi level
P5D	N1D	1	EÑ	eV	energy gap	energy gap	energy gap
P5T	N1M	1	mÏ*	m¡	effective mass	effective mass of electrons	effective mass of electrons in conduction band

Optical properties:

LPFcode_1	LPFcode_2	LPF_sortkey	Symbol	SI unit	Subtitle	Datasheet	Property
P6O	N4J	81	WF	eV	work function	work function	work function

[gjc29]

discussion of properties led to that we should only include density (3D), electron density of states at Fermi level, and work function, and that these should be in name spaces(s) and not the core. If they cannot be defined/calculated they should be left null