by Dr., Resonance Science Foundation Research Scientist
As many theoretical and computational chemists and physicists know, quantum chemical calculations involving more than an electron and nuclei are very difficult to solve. They belong to a field called many body problems and require an extensive amount of computational infrastructure and hours of calculations depending on the size (the number of particles) of the system.
Here is where artificial intelligence – a combination of artificial neural networks and machine learning – comes into play. Neural networks have been around for more than 50 years, and they are more actualized than ever before. This is because they can learn through something called backward propagation, reaching a high level of predictability and increasing accuracy by training the network.
Quantum theoretical models, together with their computational packages, have been outstandingly successful in describing the quantum regime. While these models and packages supply fast and accurate predictions of atomic chemical properties, they do not capture all the electronic degrees of freedom of a molecule, limiting their applicability in chemical reactions and chemical analysis. Molecules and nanoparticles also require much more time to reach convergence, as compared to atoms. Calculations may even take weeks or months!
Initially used to predict pattern recognition such as market behavior and facial recognition, AI is now used to predict physical-chemical molecular properties in order to design drugs or new materials, among others. In order to perform accurately, AI must incorporate the fundamental laws of quantum physics. Deep machine learning has met this challenge with the proper algorithm capable of predicting the quantum states of molecules – also known as wave functions – where all properties emerge. Such an algorithm allowing AI to solve the fundamental equations of quantum mechanics has been published in Nature Communications under the title, “Unifying machine learning and quantum chemistry with a deep neural network for molecular wavefunctions.” The work was done by researchers at the University of Warwick, the Technological University of Berlin and the University of Luxembourg. Their newly developed AI algorithm can supply accurate predictions within seconds on a laptop or mobile phone.
Their code, named SchNOrb, is available upon request.
RSF in perspectiveArtificial intelligence raises deep concerns regarding data processing, information and the nature of reality. In a coming article, a discussion about the impact, meaning and consequences of AI and data science will be placed in the context of the Generalized Holographic Theory … stay tuned!