Accuracy Ratings¶
Warning
Accuracy ratings are a work in progress. These values are heuristic estimates based on the level of theory and convergence settings. They should be used as a guide, not as an absolute measure of error. If you have suggestions on how to improve our rating system, please let us know!
Simmate assigns an Accuracy Rating to each workflow as these are helpful to beginners and also provide a quick way to sort workflows for those more experienced. This rating is a value from 0 to 5 that estimates how well the workflow's results (such as energy or geometry) are expected to predict experimental data.
Rating Scale (0-5)¶
The current rating system is based on the following classification:
| Rating | Category | Description | Examples |
|---|---|---|---|
| 5 | Hybrid DFT | Uses hybrid functionals (e.g., HSE06). Currently the most accurate practical predictors for experimental results. | matproj-hse, matproj-hsesol |
| 4 | Meta-GGA | Uses Meta-GGA functionals (e.g., r2SCAN). More accurate than standard GGA for many systems. | matproj-scan |
| 3 | Standard GGA+U | Robust, modern production settings using GGA or GGA+U (e.g., PBE v2). The industry standard for high-throughput. | matproj, mvl-slab, matproj-pbesol |
| 2 | Legacy / Older | Uses older standard settings or less rigorous convergence criteria. | mit (Legacy Materials Project) |
| 1 | Screening / LDA | Better than tutorial settings but uses faster approximations like LDA or loose convergence. | quality04, lda presets |
| 0 | Tutorial / Fast | Minimal settings intended for learning, testing, or "rough" initial relaxations. Not for publication. | quality00, evo-tutorial |
How Ratings are Determined¶
The accuracy_rating is determined by analyzing several factors:
- Level of Theory: Functional type (LDA < GGA < Meta-GGA < Hybrid).
- Convergence Criteria: Energy (
EDIFF) and force (EDIFFG) tolerances. - Basis Set Quality: Plane-wave cutoff (
ENCUT) and k-point density (KSPACING). - Pseudopotentials: Whether high-accuracy or low-accuracy (reduced electron count) potentials are used.
- Benchmark History: Known performance of the specific preset (e.g. Materials Project vs MIT Project) in the scientific literature.
For more details on the specific settings of a workflow, you can inspect the _incar or _pwscf_options attributes of the workflow class in the source code or web ui.