I was playing with the upstream nomad GUI search, to see if there is enough data for some thesis machine learning ideas and I have some usability suggestions. This is not really a bug so just posting random thoughts about the different search categories which I explored.
- Elements/formula search: I really like this, this is a huge improvement from pre v1 I can’t think of anything more needed.
- Structure: also very good, seems a bit limited by the normalizer system site limit at least on my datasets (and I wonder if matid could be able to detect for example structures with vacancies, etc… but that is not NOMAD related)
- DFT: most of the search criteria are very useful but some seems fringe at best (exact exchange fraction)
- I really miss here spinpolarization checkbox (select only spinpolarized, select only non-spinpolarized calculations)
- For example the U value search looks OK, but the criterion of exact U value might be too specific, example if I would want to search for everything that does not use Hubbard correction, selecting min 0 and max 0 doesn’t work. Also how is it supposed to work if different atoms have different +U values?
- I could use a dispersion correction treatment (van_der_waals_method) search
In general it seems problematic to search for a negative criteria. Like for example find me anything that has no +U as mentioned above (it looks like that is is often not possible to distinguish between there is no +U and there is no parser support to recognize if there is +U, few other search options have similar problem, e.g., Relativity Method: not able to distinguish none vs unknown…)
- Molecular dynamics: looks OK, can’t think of anything needed there.
- Geometry optimizations:
- the histograms doesn’t seem useful at all (all it takes is one calculation with too big value to have everything in the first column), maybe x log axis could help?
- search for “Final Force Maximum” in newtons is just plain annoying, there should be a unit selection, so one can search in some common units (eV/angstrom, Ry/Bohr, Ha/Bohr), etc.
- I could use distinguishing between just atomic optimization and full cell optimization (including lattice parameters), for the latter case also a “Final stress maximum” criterion would be cool.
Another useful criterion (for example when searching for stuff that could be used to train a force field) would be to select calculations where stress or forces are calculated. Right now one can select geometry optimization or molecular dynamics where forces are guaranteed but I did not figure out how to select for example only single_point calculations that have forces (like the ones used for phonons), or stress. This could go somewhere in properties, but dunno where exactly. I believe this was supported pre-v1.
Yeah, but I don’t want to criticize too much, I’m really happy with the current NOMAD Archive (and Oasis) state and its very useful to me in general.