Deep learning method enables to take effective boltsman distribution samples in permanent temperature range

A research team has developed a novel based on deep generative models. Their method enables effective samples of Bolt Zaman distribution in a permanent range of temperatures. These results have been published Physical review posts. The team was led by Professor Paine Ding, Associate Professor of Physics and Chemistry, and Dr. Lee Sho Hoi, Research Assistant Professor of Physics at the University of Hong Kong University of Science and Technology (HKUST).

The distribution of boltsman is one of the most important distributions to the thermal balance systems. It is very important to understand the complex system, such as phase transfer, chemical reactions, and biometroller formation. However, computing the thermodentic quantity for such systems has been a major challenge in this field for a long time.

In order to take samples of traditional numerical methods, including molecular dynamics (MD) and Markov China Monte Carlo (MCMC), when the system’s energy barrier is high, the couple requires widespread fake time to achieve the average, which is due to significant computual costs.

Influenced by the recent progress in deep generative models, Dr. Lee and colleagues suggested a normal framework-the method of distinction (VATD) (VATD)-any tractable density generative model, such as suicide models and normal flow.

The VATD can learn the distribution of bolts in the permanent temperature range, in which the thermoditatic quantity is easily achieved by automatic discrimination with the first and other order derivatives. It effectively closes an analytical partition function.

In most situations, the model guarantees the distribution of theoretically neutral bolts. More importantly, interacting the permanent range of temperatures helps remove the obstacles to the energy, and thus reduces the prejudice in the imitation.

Unlike major generative models in statistical mechanics, VATD only requires potential system energy and does not relieve the MD or Monty Carlo simulations pre -produced datases.

The team endorsed the accuracy and performance of the procedure through numerical experiences on classical statistical physics models, including the icing model and the XY model.

Professor Pan remarked, “This development paved the way for a novel’s phenomenon in complex statistics systems, which include physics, chemistry, material science, and potential applications in life sciences.”