openmaterials.ai / codes / Materials Project API

mp-api: query the Materials Project database of computed properties

The client for the Materials API, the REST interface through which the Materials Project serves its DFT-computed structures, energies and electronic properties as JSON.

What it is and how it works

A database of computed properties

Every material is a resource with its own URL, and an HTTP verb returns JSON. The database then held data for over 33 000 compounds, computed with VASP in over 15 million CPU-hours at NERSC (Jain 2013).

REST semantics

Each object, such as a material, maps to a URL of the form materialsproject.org/rest/v1/[unique-id]. A GET returns JSON. Transfer was one way, to the user (Jain 2013).

How the data are made

FireWorks manages workflows at supercomputing centers and custodian repairs failed VASP runs with rule-based fixes. Structures come mainly from the ICSD (Jain 2013).

Corrections and validation

Reaction energies mix methods for solid-gas, d-block oxide and aqueous reactions via reference reactions. Oxidation states and bond lengths are checked (Jain 2013).

Derived quantities

Energy above hull from the convex hull of formation energies measures stability. Band gaps, transition type and effective masses come from band structures (Jain 2013).

pymatgen and the web apps

pymatgen holds a high-level interface to the Materials API. The Materials Explorer, Phase Diagram App and Reaction Calculator read the same data (Jain 2013).

What people do with it

  • 33 000compounds in the database at publicationJain 2013
  • 15 millionCPU-hours at NERSC behind the databaseJain 2013
  • 85,014formulas queried through the MP APIBartel 2020

Metastability of known inorganic crystals

Of 29,902 ICSD phases retrieved through the Materials API, 50.5% are metastable, with a median of 15 meV/atom and a 90th percentile of 67 meV/atom. Nitrides sit at a median of 63 meV/atom. Sun 2016

Amorphous limit for synthesizable polymorphs

All polymorph entries for 41 systems came from the Materials Project. Cutoffs of 0.025, 0.05 and 0.1 eV/atom would exclude 63, 39 and 26% of known polymorphs. Amorphous limits exclude none. Aykol 2018

Stability from learned formation energies

85,014 formulas from the MP API trained seven models. Composition models cut formation energy MAE 27 to 74%, yet four of five stay at 0.10 to 0.14 eV/atom on decomposition energy. Bartel 2020

How it is used

  1. Choose the query

    A material id, a formula or a property range picks the resources to fetch.

  2. Request

    An HTTP GET on materialsproject.org/rest/v1/[unique-id] returns JSON.

  3. Load into pymatgen

    The pymatgen interface turns the JSON into Structure and ComputedEntry objects.

  4. Analyse locally

    Phase diagrams, metastability statistics or training sets for models.