Nonlinear multivariate curve resolution alternating least squares (NL-MCR-ALS)

Bilinearity is the basic principle of multivariate curve resolution. In this paper, we consider a case when this premise is violated. We demonstrate that the alternating least squares approach can still be used to solve the problem. The developed theory is applied to calibration of spectral data that includes the so‐called saturated peaks, which are flattened because of samples with ultrahigh absorbance. We demonstrate that in spite of serious violations of the Lambert–Beer law, the results of prediction are quite satisfactory, and the accuracy is better than in other competing methods. Copyright © 2014 John Wiley & Sons, Ltd. Bilinearity is the basic principle of multivariate curve resolution. In this paper, we consider a case when this premise is violated. We demonstrate that the alternating least squares approach can still be used to solve the problem. The developed theory is applied to calibration of spectral data that includes the so‐called saturated peaks, which are flattened because of samples with ultrahigh absorbance. We demonstrate that in spite of serious violations of the Lambert–Beer law, the results of prediction are quite satisfactory, and the accuracy is better than in other competing methods.