Extraction fidelity — reconstructed vs original

Each stored spectrum is the digitized extract; plotting it reconstructs the trace. Overlaying that reconstruction (blue) on the original NIST points (grey) confirms the unit conversion + resampling are faithful — mean round-trip RMSE 0.36% transmittance across all 1,000 spectra (median 0.30%).

Original NIST spectra vs reconstructed extract for toluene, acetone, ethyl acetate

Reconstruction vs original. Blue (extract) sits on grey (raw NIST) almost perfectly. The only measurable deviations are on small molecules with sharp gas-phase rotational fine structure (water, formaldehyde, methane), which any finite grid undersamples — a sampling limit, not a digitization error.

Provenance & honesty

Every spectrum here is a genuine measurement from NIST SRD 69 (gas phase, MSDC-IR / Sadtler collections), digitized from the original JCAMP-DX records — absorbance converted to transmittance, wavelength axes converted to wavenumber, all resampled to a common grid without smoothing. Being gas-phase, band positions can sit higher than condensed-phase textbook values (e.g. carbonyls). Involatile species (amino acids, nucleotides, polymers) are absent from this gas-phase set and remain reconstructions on their data pages.

Other experimental IR sources

NIST is the main freely-rehostable source, so it is what we store. Each entry above also links out to two larger libraries — SDBS (AIST Japan, ~54,000 compounds) and Wiley SpectraBase — whose licenses permit viewing but not redistribution, so we link rather than copy. Other openly-licensed sets surveyed are small or off-domain: the CC0 Open Spectral Database (beta, limited IR), the CC-BY Open Specy library (~325 FTIR, microplastics/polymer-focused), and the PNNL NWIR quantitative gas-phase set (~500 species, high-resolution, but not freely downloadable). Condensed-phase organic IR at scale simply isn't available under an open, rehostable license — which is exactly why this NIST-based set, fidelity-validated, is worth maintaining.