Navigating high dimension spaces with artificial intelligence

We humans, have been engineering our environment for thousands of years by shaping natural materials and inventing new materials.  However, the number of materials we have developed and exploited is relatively few.  Nature has provided us with vastly rich chemistry, with over 90 different elements.   Combining these elements in different proportions creates even a larger number of new materials.  Not just chemistry, but the arrangement of elements in a material matters.  If we select 30 common, non-toxic elements there are billons of possible yet to be explored chemistries and realizing these chemistries in different structural arrangements, by tweaking synthesis condition, we have library of materials with vast range of functionalities. 

Many of the challenges we face today could be addressed with a new material with novel functionality. 

We are developing an approach that combines an accelerated brute searches, with insights from physiochemical theories, and predictions from previous experiment via machine-learning to search vast unexplored multi-dimensional composition- synthesis parameter spaces with 100-1000 times faster.  I will illustrate this approach with a recent example of a search of materials with wear-resistance comparable to diamond-like carbon.

Apurva Mehta is a Lead Scientist at SLAC National Accelerator with 30 years of experience in advance X-ray-based metrology.  Over the last decade, she has been combining high throughput computational and experimental methods in an artificial intelligence framework to accelerate discoveries and insights.