Now, AI can accurately describe matter at the quantum level! In the latest issue of science, the neural network constructed by deepmind can predict the electron distribution in molecules and calculate the molecular properties. It was only a week since deepmind appeared on the cover of nature and solved two major mathematical problems. This breakthrough has an important impact on the fields of AI, chemistry and materials science. On the one hand, this means that deep learning has great prospects in accurately simulating quantum matter; On the other hand, it has an important impact on the exploration of materials, medicine, catalysts and other substances at the nano scale. Deepmind also said that they will use the achievement of open source to researchers all over the world! No wonder netizens will sigh: DeepMind——YYDS！ Nature says this will be one of the most valuable technologies in the field of Chemistry: Solving the problem of electron interaction with MLP This time, the problem solved by deepmind is related to density functional theory (DFT). DFT is a method to study the electronic structure of multi electron system by calculating the intramolecular electron density. It can describe matter at the quantum level, Through the approximate method, DFT first simplifies the complex electron interaction problem into a non action problem, and then puts all errors in another item to analyze the errors separately. In the past few decades, it has become one of the most commonly used methods to predict the characteristics of various systems in chemistry, biology and materials. However, this method still has some limitations. On the one hand, it has delocalization error. In DFT calculation, the functional will find the electron configuration when the energy is minimized to infer the electron density of the molecule. Thus, the function error will bring electronic error. Most of the existing density functional will mistakenly distribute the electron density on several atoms or molecules, rather than determine it around a single molecule or atom. △ the left figure shows the traditional method, and the right figure shows the method proposed by deepmind Another major error comes from the destruction of spin symmetry. If the chemical bond breaking in the structure is described, the existing functional will give a configuration in which the spin symmetry is destroyed. However, symmetry plays an important role in the study of physical and chemical configurations, so this defect of the current method has caused great errors. It can be seen from the comparison that the PBE method breaks the spin symmetry. Therefore, deepmind proposes a neural network deepmind 2021 (dm21 for short). This framework uses multilayer perceptron (MLP), which can map a set of input vectors to a set of output vectors. After inputting precise chemical data such as spin exponential charge density into a weight shared MLP, it can predict the enhancement value of local charge density and local energy density. After integrating these values, the dispersion correction DFT is added to the function. After training, the first mock exam can be deployed in the self consistent calculation. In the comparison of specific data, the error values of dm21 are lower than those of traditional methods. In other words, dm21 can accurately simulate the transition states of complex systems, such as hydrogen chains, charged DNA base pairs and double radical systems. The experimental results show that the absolute error of dm21 on different benchmarks (gmtkn55 BBB qm9) is less than that of ordinary methods. Therefore, it is not difficult to see that dm21 can construct a more accurate description of electron interaction than DFT method, and deep learning will also have great prospects for accurately simulating matter at the quantum level. AI has shocked the biological and mathematical circles One of the results of this research is James Kirkpatrick, a research scholar at Google deepmind. He said that understanding micro phenomena is of great significance for the study of clean power and micro plastic pollution. This also has a profound impact on researchers' exploration of new materials, drug development and catalysts at the nano level. This is not the first time deepmind has shocked the scientific community with AI. This year, they used alphafold2 to predict 98.5% of human protein, which shocked the biological community. Not long ago, they broke through two major mathematical problems with AI and appeared on the cover of nature, which had a profound impact on knot theory and representation theory. (Xinhua News Agency)