Data reduction of matrices from the high performance liquid chromatography/swept-potential electrochemical detector system and their examination by the factor analysis approach

In the analysis of multicomponent samples, analytical chemists are faced with the problem of incomplete separation of the response signals of the multiple analytes in the samples. Recent progress has been made in the design and development of instruments with multidimensional output in order to achieve resolution of the response signals by profile differences. If resolution by profiles is not possible, another approach is to use suitable chemometrics techniques to complement the instrument's limitations. In the high-performance liquid chromatography coupled with fast sweep electrochemical detectors, such mathematical approaches to process multidimensional data are lacking.; The purpose of this study was to develop a curve-resolution computer program for analyzing data from high-performance liquid chromatography using a swept-potential electrochemical detection system (HPLC/SPED) which is currently used in this laboratory. There was also a need to develop a digital filtering program in order to improve the signal-to-noise ratio where necessary. Yet another need was to be able to display the three-dimensional data (current, potential, and retention time) on a two-dimensional surface.; For curve resolution, a computer program based on factor analysis was developed. The autocorrelation coefficients (AUTO), the coefficients of multiple determination (RATIO) and single-vector uniqueness testing were investigated as predictors of the number of components in a matrix. For digital filtering, three filter transfer functions, namely, the ideal low pass filter (ILPF), the exponential low pass filter (ELPF), and the butterworth low pass filter (BLPF) were investigated to determine which was the most suitable for the HPLC/SPED data.; In factor analysis, all the predictors gave very satisfactory results. For systems where the least concentrated component was about 1/50 of the others, the single-vector uniqueness testing performed better than either AUTO or RATIO. In the digital filtering program, both ELPF and BLPF performed better than ILPF on the basis of the root mean square error.