The road to standardized results data [WIP]

[rartino]

Note: I'm drafting this issue (but it is convenient to work on it at GitHub). I will remove the [WIP] suffix when it is ready.

It is of interest to many database providers to make values of physical, and other, properties resulting from computation and/or experiments available via OPTIMADE. We have had many discussions over the years on how this should be represented in a way that allows for standardized federated access and query. Right now, it is fairly common to make such data, when tied to specific structures, available by provider-specific properties in the /structures endpoint.

However, as an official mechanism this has limits. It is quite unclear how to handle, e.g., multiple parallel computations (or experiments) of the same property for the same structure. One can "duplicate" the structures entry so there are two identical structures, bulk_modulus calculated with LDA and one by PBE. This is not ideal from the view of data normalization, nor will the interpretation of such data be particularly clear for a client (e.g., when requesting results as bulk_modulus > 36). The other alternative is to define a large array of property names, e.g. bulk_modulus_computed_dft_lda, bulk_modulus_computed_dft_pbe, but there is no way to properly systematically represent all possible set of parameters this way.

...

We have outlined the following steps forward:

• one
• two
• three