- DeepMind, a scientific paper, demonstrates how neural networks can improve approximation to the Density Functional (a method that describes electron interactions in chemical systems).
- This is a great example of deep learning’s potential in simulating matter at the level of quantum mechanics.
- DeepMind will also release the code as a paper to make it available to others.
DeepMind published a paper in the scientific journal Science demonstrating how neural networks can help describe electron interactions in chemical systems better than other methods.
Density Functional Theory was developed in the 1960s and describes the mapping of electron density to interaction energy. The exact nature of the mapping between electron densities and interaction energy, also known as the density functional, has been unknown for more than 50 years. DeepMind’s breakthrough in the field has demonstrated that neural network can be used for a more precise map of electron density and interaction than previously possible.
DeepMind trained the model to learn functionals by expressing it as a neural network and including exact properties in the training data. This allowed DeepMind to avoid two critical systematic errors — spin symmetry breaking and delocalization error — which resulted in a better description of a wide range of chemical reactions.
This will allow researchers to quickly access our code and better approximate the exact Density Function. It is another step in deep learning’s promise of accurately simulating matter on the quantum mechanical level. This may allow material design in a computer, allowing researchers to explore questions regarding materials, medicines and catalysts at the nanoscale.
DeepMind Research Scientist James Kirkpatrick says, “Understanding technology on the nanoscale is becoming more crucial in helping us to tackle some of our major challenges of 21st Century, from clean electricity and plastic pollution.” “This research is a first step towards understanding the interactions between electrons and the glue that holds molecules together.
DeepMind made the paper and the open-source code available to accelerate progress in the field.
DeepMind
DeepMind is a scientific research company committed to solving intelligence to advance Science, humanity and the planet. It works closely with scientists, engineers, ethicists and scientists to develop advanced artificial intelligence.