|
|
Loading ...
|
|
|
|
|
pages views since
05/19/2016 : 150001
· Members : 7
· News : 841
· Downloads : 0
· Links : 0
|
|
|
|
|
AI Unlocks Crystal Patterns to Drive Innovations of the Future
|
|
|
|
Posted by Okachinepa on 12/09/2024 @
Courtesy of SynEvol:
Credit:University of Reading
To forecast how atoms would arrange themselves in crystal structures, a novel artificial intelligence model called CrystaLLM has been created. This innovation may hasten the development of novel materials for solar cells, computer chips, and batteries, among other technologies.
CrystaLLM, developed by researchers at the University of Reading and University College London, functions similarly to artificial intelligence chatbots in that it analyzes millions of crystal structures to learn the "language" of crystals.
The technology will be made available to the scientific community to help advances in material discovery after being published today (December 6) in Nature Communications.
Predicting crystal formations is similar to solving a challenging, multifaceted jigsaw with hidden components, according to Dr. Luis Antunes, who oversaw the study while doing his PhD at the University of Reading. Predicting crystal structures involves testing innumerable atom configurations, which demands enormous processing power.
Similar to a skilled puzzle solver who identifies winning patterns rather than attempting every potential move, "CrystaLLM offers a breakthrough by studying millions of known crystal structures to understand patterns and predict new ones."
Currently, the method for determining how atoms will form crystals is based on laborious computer simulations of the atoms' physical interactions. CrystaLLM functions more simply. It learns by scanning millions of crystal structure descriptions included in Crystallographic Information Files, the standard format for representing crystal structures, rather than by doing intricate physics computations.
These crystal descriptions are handled by CrystaLLM in the same way as text. It gradually picks up patterns about the structure of crystals as it reads each description and makes predictions about what will happen next. The system learned the physics and chemistry rules on its own without ever being taught them. Just reading these descriptions taught it things like how atoms arrange themselves and how their size influences the structure of the crystal.
Even with materials that CrystaLLM has never encountered before, it was able to produce realistic crystal formations during testing.
The research team has developed a free website that allows researchers to generate crystal structures using CrystaLLM. Better batteries, more effective solar cells, and speedier computer processors might all be developed more quickly if this model is incorporated into crystal structure prediction workflows.
|
|
|
