Probing of Neural Networks as a Bridge from Ab Initio Relevant Characteristics to Differential Scanning Calorimetry Measurements of High‐Energy Compounds. Issue 3 (2nd July 2021)

• Record Type: Journal Article
• Authors: Bondarev, Nikolay V.; Katin, Konstantin P.; Merinov, Valeriy B.; Kochaev, Alexey I.; Kaya, Savas; Maslov, Mikhail M.
• Other Names: Zhou Kun guestEditor.; Liu Bo guestEditor.; Cai Yongqing guestEditor.; Dmitriev Sergey V. guestEditor.; Li Shaofan guestEditor.
• Is Part Of: Physica status solidi. Volume 16:Issue 3(2022)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-07-02
• Subjects: density functional theory -- high-energy-density materials -- machine learning -- neural networks -- quantum chemistry descriptors; Solid state physics -- Periodicals; 530.4105
• DOI: 10.1002/pssr.202100191
• Languages: English
• ISSNs: 1862-6254

Abstract

Abstract : The relationships between the theoretical values calculated using density functional theory and experimental data derived from the differential scanning calorimetry of high‐energy organic compounds are studied. The theoretical values are the number of atoms and bonds of different types and their lengths, minimum eigenfrequencies, atomization energies, ionization potentials, electron affinities, and frontier orbital energies. The experimental data are the amounts of releasing heat (the first peaks higher than 1 kJ g −1 ) and corresponding temperatures. Neural networks and regression, factor, discriminant, and cluster analysis are applied to find the dependencies between theoretical values and experimental data. It is found that the heat amount cannot be predicted in the general cases, whereas the corresponding temperature can be predicted with a neural network with an accuracy of ≈30 °C. Cluster and discriminant analysis provides the way for the classification of high‐energy compounds into three groups. Some of these groups require particular rules for the prediction of experimental data from the theoretical values. Abstract : The results of differential scanning calorimetry and theoretically calculated density functional theory based quantum descriptors for high‐energy compounds are compared. Neural networks are trained to predict experimental results based on calculated values. It is obtained that the thermal stability of high‐energy compounds can be predicted with satisfactory accuracy. A statistics‐based classification of high‐energy compounds is proposed.