FlashAIR-QM9d: QM9-Derived DFT IR Spectra and Molecular Database for FlashAIR Framework

This dataset contains the QM9-derived molecular structures and DFT-calculated infrared (IR) spectra used for training and benchmarking the FlashAIR framework presented in the accompanying paper. The database includes 133,885 unique organic molecules (H, C, N, O, F) with up to nine heavy atoms, along with their harmonic vibrational frequencies and IR intensities computed at the B3LYP/6-311+G(d,p) level using Gaussian 16. All spectra are preprocessed onto a standardized 825-point wavenumber grid (550–3846 cm⁻¹, 4 cm⁻¹ spacing) with Lorentzian broadening (FWHM = 24 cm⁻¹). The dataset serves as the primary reference for the fragment-based spectral library construction in FlashAIRa and the Gaussian Process Regression training in FlashAIRx. This database is an extended and preprocessed version derived from the QM9 dataset (Ramakrishnan et al., Sci. Data 2014).