The Materials Project is pioneering the fourth paradigm of data-intensive science in the materials science community. This data-centered framework will allow scientists to collaborate and derive scientific insight in materials science in a way that is not possible with current scientific tools.
Computation is currently accepted as the third paradigm of science, alongside experiment and theory.
In sciences from biology to particle physics, hundredfold to thousandfold increases in the data from simulations and post-analysis of experiments has caused a fourth paradigm to emerge, as proposed by Jim Gray.
The fourth paradigm lays the foundation for exploratory data-driven science to complement traditional hypothesis-driven research through emphasis on a Collaborative, Networked, and Data-driven framework.
We computed core properties of over 80,000 materials, screened 25,000 of these for Li-ion batteries, and have used this data to predict several new battery materials. But this is just the tip of the iceberg for the potential of this approach. The true potential lies in opening the data and its derivations to unfettered exploration by the larger scientific commmunity. As a first step in enabling this, we have built the Materials Explorer, an intuitive and detailed user interface for specifying materials and properties. We also provide access to advanced applications to compute common electrochemical and other properties.
We leveraged the fundamental scalability of the Web architecture, generally known as Representational State Transfer (REST) (Fielding 2000). Right now, registered users can store searches as URLs, called permalinks, that make sharing of results natural and easy. We want to open the floodgates to third-party exploratory application mashups, using the Materials Project REST API and JSON as the common data interchange format.
We designed the Materials Project from the ground up to treat all inputs, outputs, processes, and derivations of the data in a uniform manner. This allows the system to be fundamentally searchable using a rich, unified syntax. By saving derived values, re-computation is avoided. We aim to record all the relevant details of the scientific process to enable other users to re-try or modify calculations. We are currently exploring the statistical techniques for structure prediction that can reduce the search space by orders of magnitude.