Carvalho Felipe Silva, Braga João Pedro
Departamento de Química - ICEx, Universidade Federal de Minas Gerais, 31270-901, Belo Horizonte, MG, Brazil.
J Mol Model. 2022 Mar 23;28(4):99. doi: 10.1007/s00894-022-05055-5.
The Hopfield neural network has been applied successfully to solve ill-posed inverse problems in simple monoatomic liquids structure using scattering experimental data to retrieve the radial distribution function, g(r), and direct correlation function, C(r). In this work, the method was extended to a more complex system: a two-component glassy solid, GeSe. To acquire results with correct peak intensities and behavior for large values of r, it was necessary to carry out the calculations a few times by adjusting the initial conditions to solve a set of coupled equations. However, the new initial conditions are simple and can be defined based on the results obtained at each run. In this sense, the method robustness is also evident while retrieving the radial distribution function for more complex systems from experimental data.
霍普菲尔德神经网络已成功应用于利用散射实验数据解决简单单原子液体结构中的不适定逆问题，以检索径向分布函数g(r)和直接相关函数C(r)。在这项工作中，该方法被扩展到一个更复杂的系统：二元玻璃态固体GeSe。为了获得具有正确峰值强度和大r值行为的结果，有必要通过调整初始条件多次进行计算，以求解一组耦合方程。然而，新的初始条件很简单，可以根据每次运行获得的结果来定义。从这个意义上说，在从实验数据中检索更复杂系统的径向分布函数时，该方法的稳健性也很明显。
