Raman Spectra of Hydrated Calcium Sulfate (Gypsum) from Early Measurements to the Use of Artificial Intelligence

This study explores the significant role of Artificial Intelligence in the evolution of Raman spectroscopy for mineralogical analysis. We focus on gypsum (CaSO4⋅2H2O) as a case study, with a particular emphasis on its water of crystallization. Gypsum was among the first minerals studied with Raman spectroscopy (in 1945), utilizing a pioneering yet sophisticated technique developed by Franco Rasetti. Today, we have advanced laser sources, diffraction gratings, CCD and CMOS devices for spectrum recording, and various software tools for analysis. Furthermore, databases like RRUFF, for instance, provide reference spectra for material analysis and comparison. Beyond the evident technological advancements, we now wish to highlight the transformative role of Artificial Intelligence (AI) as the latest pillar of a bridge connecting early measurements to recent data freely available in RRUFF. This pillar enables advanced data analysis and comparison. AI's capability to discriminate subtle differences is illustrated through a detailed comparison between the Raman spectra of gypsum and bassanite (CaSO4⋅0.5H2O), a closely related hemihydrate, focusing on their distinctive vibrational signatures. By analyzing the portion of the spectrum related to the water of crystallization, it has been highlighted that deconvolution is necessary for peak evaluation. The approach we demonstrate, based for simplicity on Gaussian and q-Gaussian components, shows Google Language Model's absolute capability to perform deconvolution and provide the relevant Python program. This demonstrates that an AI trained on Raman spectra can perform their deconvolution, thus becoming autonomous in the analysis and comparison of spectra from massive databases.