Full information of system properties inferred from individual particle dynamics

C Liang and D Huang and SY Lu and Y Feng, PHYSICS OF PLASMAS, 31, 113702 (2024).

DOI: 10.1063/5.0239733

Using the machine learning method, the screening parameter kappa and the coupling parameter Gamma of two-dimensional (2D) dusty plasma are determined simultaneously purely from position fluctuations of individual particles using both simulation and experiment data. To train, validate, and test convolutional neural networks (CNNs), Langevin dynamical simulations are performed with different kappa and Gamma values to obtain position fluctuation data of individual particles. From the test with the simulation data, the trained CNNs are able to accurately determine the values of kappa and Gamma simultaneously, with the typically averaged mean relative error varying between 10 % and 17 %. While using the trained CNN with the 2D dusty plasma experiment data, the distribution of the determined kappa(NN )or Gamma(NN) values always exhibits one prominent peak, and the peak locations well agree with the kappa and Gamma values determined from the widely accepted phonon spectra fitting method. The obtained results clearly demonstrate that, using machine learning methods, the two global characterization parameters of kappa and Gamma in 2D dusty plasmas are able to be accurately determined simultaneously purely from the position fluctuations of local individual particles.

Return to Publications page