AIB-21: Adsorption Isotherm Benchmark

AIB-21 is a curated adsorption isotherm dataset comprising 8,486 data points across 174 isotherms spanning 21 adsorbent–adsorbate systems, drawn from the NIST/ARPA-E Database of Novel and Emerging Adsorbent Materials (NIST ISODB). The dataset covers a temperature range of 23–800 K and a pressure range of 3.0×10⁻⁶–299.4 bar, spanning conditions relevant to cryogenic gas storage, post-combustion CO₂ capture, and high-temperature industrial separations. Systems were selected from the raw NIST ISODB corpus (613,164 points, 11,157 systems) through an eight-stage quality-control pipeline enforcing verified point-level monotonicity, sufficient thermal coverage (≥4 temperatures spanning ≥20 K), sufficient pressure coverage (≥1 log decade, ≥8 effective unique pressures), and unit-aware grouping to prevent merging of incompatible measurement conditions. The 21 systems span five adsorbate classes, light nonpolar/quadrupolar gases, polar protic species, aliphatic hydrocarbons, and aromatics, across thirteen materials including microporous zeolite, rigid and flexible metal–organic frameworks, and non-framework porous solids. This dataset was assembled to support benchmarking of adsorption isotherm models, classical and machine-learned, under interpolation and extrapolation conditions, and is released here as a standalone, citable resource. Source data: Siderius, D. (2020). NIST/ARPA-E Database of Novel and Emerging Adsorbent Materials (Version 1.0.4) [Dataset]. National Institute of Standards and Technology. https://doi.org/10.18434/T43882 Associated code and analysis pipeline: https://github.com/SatyaK-0/AIB21-AdsorptionBenchmark Associated manuscript: Kokonda, S., Mukherjee, K., & Colón, Y. J. Interpolation Is Not Understanding: A Benchmark Revealing the Thermodynamic Limits of Symbolic Regression for Adsorption Extrapolation Compared to Classical Models. Manuscript in preparation.