GitLab

The current structure normalizer, which classifies the structure e.g. in surface, volume, 1D, 2D, should be extended by the MatID structurecluster class. To avoid false positives and false negatives, a decision tree is integrated.

• Add a git submodule for the MatID development version that has the new StructureClusterer class.
• Add the new Topology metainfo from the MD branch. This is liable to change in the future, but it's easy to adapt afterward (atomutils.py, normalizing/material.py, datamodel/results.py).
• Add the GUI updates for visualizing topologies from the MD branch (archive.js, Quantity.js, Structure.js, BrillouinZone.js, Overview.js, MaterialCardTopology.js)
• Modify nomad/normalizing/material.py to include a new function that adds the new topological (=hierarchy of identified subsystems) data into results.material.topology. For an easy start, we can simply:
  • Run this clusterer only after we see that the results.material.structure_type == unavailable.
  • Concentrate on a very simple case first, e.g. getting reasonable results for systems that comprise of several stacked surfaces (=heterostructures) or adsorbtion systems.
• Add unit test for the surface examples (6?):
  • Test that correct elements and number of elements are present
  • Test that topology contains only surfaces (molecules/outliers should not be stored in the topology for now.
  • Test that the cluster indices match up to some percentage (90?)
  • Test that structural_type = surface