This project aims to develop an automated process to simplify separation and assignment of individual spectra from complex mixtures. Three methods were developed to extract related groups of spectroscopic transitions for easier assignment of chemical species. All have the same core idea of correlating intensities between two spectra recorded with one varying experimental parameter such as concentration. Each spectroscopic transition forms a point on a plot of Intensity in Spectrum 1 (I1) vs Intensity in Spectrum 2 (I2), and on this I1 vs I2 plot, related transitions cluster to form rays. The approaches studied aim to extract data groupings from the original complex spectrum to give separated spectra, hypothesizing that each belongs to one chemical species. Techniques tested were (i) a version of extended cross-correlation, based on Robert Field’s work at MIT that uses exponential functions to determine which ray each data point lies on, (ii) grouping points based on intensity ratios between two spectra, and (iii) analysis of point density within rays on the I1 vs I2 plot, which allows for identifying related transitions based on the number of data points present. Comparison of results of all three methods with previously assigned data provides initial confirmation of the potential of each method to be automated and applied to a range of chemical systems.
These techniques were tested on spectra of clusters containing combinations of 1,1-difluoroethylene (DFE) and CO₂. Assigned clusters include (DFE)₁(CO₂)₂ and (DFE)₁(CO₂)₃. Clusters that are hypothesized are (DFE)₂(CO₂)₁, (DFE)₃(CO₂)₁, and (DFE)₁(CO₂)₄. Formulas of these clusters are inferred from the I1 vs I2 plot, where the angle of each ray is related to cluster composition. Results from these methods are also being correlated with graphical principal component analysis-based methods, and the approaches appear complementary.

I1 vs I2 plot of assigned and likely clusters. Inset: Number of rotational transitions vs ray angle.

Brooks Pate

Appearances