Cutting through the Noise: Extracting Dynamics from Ultrafast Spectra Using Dynamic Mode Decomposition.

Area of Science:

• Physical Chemistry
• Spectroscopy
• Chemical Dynamics

Background:

• Coherent multidimensional spectroscopy, particularly ultrafast two-dimensional infrared (2D IR) spectroscopy, probes molecular structure and subpicosecond dynamics in solution.
• Traditional analysis methods, like center/nodal line slope, infer dynamics from spectral line shape evolution but are sensitive to subjective choices and model assumptions.
• Extracting dynamics from complex spectra with overlapped peaks and cross-peaks presents significant challenges for conventional approaches.

Purpose of the Study:

• To introduce and evaluate Dynamic Mode Decomposition (DMD) as a novel, data-driven method for extracting molecular dynamics from ultrafast 2D IR spectra.
• To demonstrate DMD's capability in analyzing complex spectral features, including overlapped transitions and cross-peaks, which are difficult for traditional methods.
• To assess the performance of DMD combined with conditional generative adversarial neural networks for dynamics recovery under low signal-to-noise conditions.

Main Methods:

• Application of Dynamic Mode Decomposition (DMD), a data-driven technique, to analyze ultrafast 2D IR spectral data.
• Evaluation of DMD performance using both simulated and experimental spectra, specifically those containing overlapped peaks.
• Integration of conditional generative adversarial neural networks (cGANs) with DMD to enhance dynamics retrieval in low signal-to-noise scenarios.

Main Results:

• DMD successfully extracts spatiotemporal structures directly from complex 2D IR spectra, providing a more objective analysis.
• The method accurately retrieves dynamics from overlapped transitions and cross-peaks, overcoming limitations of traditional spectral analysis techniques.
• The combination of DMD and cGANs demonstrates robust recovery of molecular dynamics even at significantly reduced signal-to-noise ratios.

Conclusions:

• Dynamic Mode Decomposition offers a powerful, assumption-free approach for analyzing molecular dynamics from multidimensional spectroscopic data.
• DMD effectively handles spectral complexity, including overlapped features, providing more reliable dynamic information than conventional methods.
• The DMD framework is versatile and can be extended to other forms of multidimensional spectroscopy, enhancing its broad applicability.