(Received: May 12, 2004; Accepted for publication: May 31, 2004; Published on Web: December 28, 2004)
Various computer-aided methods are used to estimate chemical structures from infrared spectra: an inductive method using a spectral database, an empirical method using a knowledge base, and a theoretical method using computational chemistry. With an aim to increase an ability for estimating structures from infrared spectra, various computer methods were intensively investigated. At first an inductive method using a spectral database was studied, and several problems were pointed out. Next a linear learning machine and a neural network were applied to estimating functional groups from infrared spectra as an empirical method, and several problems were also mentioned. Third a non-empirical molecular orbital calculation as a theoretical method was applied to examine the agreement between calculated and experimental vibrational frequencies and infrared intensities, and several problems were also underlined. It was concluded that there are many difficulties in estimating chemical structures from infrared spectra with the power of existing computers.
Keywords: Infrared spectra, Structure estimation, Spectral database, Neural network, Non-empirical molecular orbital calculation
Text in Japanese