> For the complete documentation index, see [llms.txt](https://docs.materialsproject.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.materialsproject.org/services/ml-and-ai-applications.md).

# ML & AI applications

*These pages are under active development as we improve the quality of our documentation on ML applications and ML-dataset accessibility.*

### Motivation

Predicting the properties of materials, such [as those computed by the Materials Project](https://docs.materialsproject.org/methodology/total-energies/) and [its collaborators](https://next-gen.materialsproject.org/contribs), can be an expensive and time-consuming undertaking. Synthesizing, characterizing, and experimentally validating novel materials in the physical world is vastly more expensive.

New techniques based on ML are a promising avenue for reducing the time and cost of materials development by multiple orders of magnitude. These include direct prediction of physical properties from structural or chemical composition information, and direct prediction of the materials potential energy surface by "foundation potentials" (FPs), AKA universal machine learning interatomic potentials (MLIPs).

Using information such as crystal structure, chemical composition, and electronic bandstructures as inputs, ML models can be trained to predict otherwise expensive or computationally prohibitive properties. More, ML can help scientists generate novel ideas for highly complex tasks like solid-state synthesis and crystal structure determination with heretofore unobtainable physical insights. In practice, this means scientists can screen enormous chemical and configurational search spaces in the pursuit of stable, cheap, and highly performant materials.