New Dielectric Materials

We just published data on two materials, Zr₂Bi₂O₇ and CsTaTeO₆, which were the result of a workflow to discover new dielectric materials. Dielectrics are in used CPUs and SSDs among many other electronic devices. We managed to experimentally synthesize both these materials and measure their dielectric properties (thanks Wes!), so I wanted to take the opportunity of having two crystal structures that are special to me to write a short tutorial on how to render 3D crystal structures in a browser using one of my side-projects called elementari.

elementari currently only understands pymatgen’s JSON-like structure representation. Converting the experimental CIF files to this format is easy:

from glob import glob

import pymatgen.transformations.advanced_transformations as pat
from pymatgen.core import Structure

for cif_file in glob("*.cif"):
    struct = Structure.from_file(cif_file).add_oxidation_state_by_guess()
    # remove partial occupancies, elementari does not support them (yet)
    ordered = pat.OrderDisorderedStructureTransformation().apply_transformation(struct)
    ordered.to(cif.replace(".cif", ".json"))

The resulting JSON files can then be visualized with elementari using the Structure component which takes only a few lines of code:

<script>
  import { Structure, StructureCard } from 'elementari'

  // parse all JSON files in this directory
  const structs = import.meta.glob(`./*.json`, { eager: true, import: 'default' })
</script>

<!-- iterate over JSON objects and pass them to the <Structure /> component for rendering -->
{#each Object.entries(structs) as [name, structure]}
  <StructureCard {structure} />
  <Structure {structure} />
{/each}

Here’s what this code renders:

Note the structures are rendered without bonds. That’s because the converted CIFs give us pymatgen Structures, not StructureGraphs which contain adjacency matrices to convey bond information. elementari can render bonds but StructureGraph support (i.e. converting the adjacency matrix into a list of bonds) is still WIP. While elementari contains JS code for on-the-fly bond detection (based on max distance or nearest neighbor heuristics), it doesn’t work well in many cases (don’t use it). Hopefully something I’ll fix soon.