AI tackles one of quantum chemistry’s toughest challenges

New research using neural networks, a form of AI inspired by the brain, offers a solution to the difficult challenge of modeling the states of molecules.

Research shows that this technique can help solve fundamental equations in complex molecular systems. It could lead to practical applications in the future, helping researchers prototype new materials and chemical syntheses using computer simulation before trying to make them in the lab.

Led by scientists from Imperial College London and Google DeepMind, the study is published in Science.

Excited molecules

The team studied the problem of understanding how molecules transition into and out of excited states. When molecules and materials are stimulated with a large amount of energy, for example by being exposed to light or high temperatures, their electrons can be propelled into a new temporary configuration, called an excited state.

The exact amount of energy absorbed and released as molecules transition between states creates a unique fingerprint for different molecules and materials. This affects the performance of technologies ranging from solar panels and LEDs to semiconductors and photocatalysts. It also plays a critical role in biological processes involving light, including photosynthesis and vision.

However, this fingerprint is extremely difficult to model because the excited electrons are quantum in nature, meaning that their positions in molecules are never certain and can only be expressed as probabilities.

Dr David Pfau, Principal Investigator at Google DeepMind and the Department of Physics at Imperial College, said: “Representing the state of a quantum system is a huge challenge. Each possible configuration of electron positions needs to be assigned a probability.

“The space of all possible configurations is huge. If you tried to represent it as a grid with 100 points along each dimension, then the number of possible electronic configurations for the silicon atom would be larger than the number of atoms in the universe. This is exactly where we thought deep neural networks could help.”

Neural networks

The researchers developed a new mathematical approach and used it with a neural network called FermiNet (Fermionic Neural Network), which was the first example of deep learning being used to calculate the energy of atoms and molecules from first principles that was accurate enough to be useful.

The team tested their method with a series of examples and obtained promising results. On a small, complex molecule called a carbon dimer, they obtained a mean absolute error (MAE) of 4 meV (millielectronvolt, a tiny measure of energy), which is five times closer to experimental results than previous benchmark methods reaching 20 meV.

Dr Pfau said: “We tested our method on some of the most challenging systems in computational chemistry, where two electrons are excited simultaneously, and found that we were within about 0.1 eV of the most demanding and complex calculations performed to date.

“Today, we are making our latest work open source and hope that the research community will build on our methods to explore the unexpected ways in which matter interacts with light.”