No abstract

Summary Transmission and diffuse reflectance FTIR spectra of samples of Pinus radiata wood have been used as the basis of PLS-1 (partial least squares) and PLS-2 (projection to latent structures) models for the prediction of extractives, lignin, total carbohydrate and basic density. Very little difference was observed between models based on transmission spectra and those based on diffuse reflectance spectra. Because of the ease of sample preparation and presentation the diffuse reflectance technique was chosen as the method of preference. Similarly very little difference was observed between models prepared using PLS-1 and models prepared using PLS-2 correlations of diffuse reflectance spectra. Multiple correlation coefficients between the four measured properties and the diffuse reflectance spectra using PLS-2 modelling with four principal components are respectively: extractives, 0.87; Klason lignin, 0.84; total carbohydrate, 0.58; and density, 0.87.

Principal component analysis is used to examine large multivariate databases. The graphical approach to exploratory data analysis is described and illustrated with a single example of chemical composition data obtained on environmental dust particles. While the graphical approach to exploratory data analysis has certain advantages over the numerical procedures, the empirical approach described here should be viewed as complementary to the more robust treatments that statistical methodologies afford.

Adaptive femtosecond control experiments are expanding the possibilities for using laser pulses as photophysical and photochemical reagents. However, because of the large number of variables necessary to perform these experiments (usually 100-200), it has proven difficult to elucidate the underlying control mechanisms from the optimized pulse shapes. If adaptive control is to become a widespread tool for examining chemical dynamics, methods must be developed that reveal latent control mechanisms. This manuscript presents a generally applicable method for dimension reduction of adaptive control experiments based on partial least squares regression analysis (PLS) of the normalized covariance matrix of the total data set. When applied to experimental results obtained in our laboratory, it shows that only seven fundamental dimensions from an original 208-dimension search space are needed to account for approximately 90% of the variance in the observed fitness of 11,700 laser-pulse shapes explored during the optimization experiment. Furthermore, the seven dimensions have a remarkable regularity in their functional form. It is anticipated that this work will facilitate theoretical treatments directly linking the optimal fields to control mechanisms, allow quantitative comparisons of independent control results, and suggest new experimental methods for rapid adaptive searches.

No abstract