Climate Change - a Chemistry Problem: A Periodic-Table-Wide Density Functional Theory Benchmark and Machine Learning Surrogates for Combustion-Pollutant Adsorbents

The future of climate change relies on us solving a chemistry problem: identifying compounds that can bind, break down, or capture the specific gases driving warming and air-quality collapse, at cost points that permit deployment where the emissions actually occur. We construct an open dataset of adsorption energies computed via Density Functional Theory (DFT) for five combustion pollutants - CO2, NO2, CO, CH4, and SO2 - across 102 chemical elements (510 systems), and train machine learning (ML) surrogate models on it so downstream screens can bypass DFT entirely. The pipeline, implemented in Quantum ESPRESSO, uses a two-stage protocol: a Self-Consistent Field (SCF) single-point sweep across all 510 (element, gas) pairs for triage, followed by full Structural Relaxation for the strongest binders. We train two ML baselines that predict the adsorption energy($E_{ads}$) from features alone: an XGBoost regressor over element descriptors and a lightweight equivariant graph neural network, evaluated with element-stratified splits that test generalization to unseen chemistries. The SCF-triage stage is complete across all 102 elements; relaxation is ongoing. Unlike existing catalysis benchmarks that prioritize sample count on a narrow gas panel, we prioritize periodic-table breadth on a gas panel chosen for climate and health impact - offering, to our knowledge, the first ML-ready adsorption dataset of this kind.