Coupling Chemistry and Machine Learning Across the Lithium Supply Chain: Resource Geochemistry, Extraction, Separation, and Recycling

Mustafa Rezaei, Gabriela Sanchez Lecuona, Omid Abdolazimi

Lithium supply must expand sharply over the coming decades to support the electrification of transport and the growth of stationary energy storage. At almost every stage the governing chemical challenge is the same: concentrating and separating Li⁺ from chemically similar competing ions at acceptable economic and environmental cost. This review examines the supply chain through two interconnected lenses the chemical mechanisms that govern lithium recovery, and the data-driven methods used to characterize resources, model processes, and optimize recovery. We first summarize the geochemistry that controls where lithium concentrates, from lithium-cesium-tantalum pegmatites and closed-basin brines to volcano-sedimentary claystones and oilfield waters, and show how machine-learning resource assessment has revised occurrence estimates, notably for the Smackover Formation of Arkansas. We then review the chemistry of the three primary recovery routes spodumene conversion and acid roasting, evaporative brine processing, and direct lithium extraction by aluminate sorbents, ion sieves, membranes, and electrochemical systems together with the purification steps that turn concentrates into battery-grade chemicals, where magnesium and calcium removal dominates refining cost. Recycling is treated as an emerging fourth resource: pyrometallurgical, hydrometallurgical, early-stage selective, and direct-regeneration routes are compared against the recovery targets of the European Union battery regulation. Throughout, we assess where machine learning, surrogate modeling, and process optimization have delivered validated gains from sorbent design to battery sorting and where claims outrun evidence. We close with an integrated framework linking chemical mechanism to data-driven development, and identify data scarcity, absent benchmarks, and limited transferability as the binding constraints of the next decade.