Final Defense: Kaustubh P. Bawankule
Tuesday, December 1, 2026 at 9:00 AM to 11:00 AM
Event Details
Location
Dudley 1119
Add To Calendar
Event Description
"MULTISCALE APPROACH TO RARE EARTH ELEMENT SEPARATION: COMPLEMENTARY INSIGHTS FROM EXPERIMENTS, MOLECULAR SIMULATIONS, AND MACHINE LEARNING"
Kaustubh P. Bawankule, MSE PhD Candidate
Advisor(s): Dr. John A. Howarter
ABSTRACT
Rare earth elements (REEs) are critical minerals whose importance far exceeds the security of their supply: they are essential to permanent magnets, catalysts, and defense systems, yet their refining remains highly concentrated. The constraint is chemical rather than geological: the lanthanide contraction leaves adjacent elements nearly indistinguishable in charge, size, and coordination behavior. Securing supply therefore requires both technospheric mining of secondary feedstocks and more efficient separation of the mixed streams they yield. In solvent extraction, the dominant separation route, performance is governed by the supramolecular organization of the organic phase, which is difficult to resolve experimentally. This work examines what governs REE recovery and separation across length scales, and whether these insights can be generalized into transferable design rules, by integrating experiments, molecular simulations, and machine learning.
Phosphogypsum, generated in hundreds of millions of tonnes annually, is a notable secondary feedstock, yet reported leaching recoveries vary widely. A response surface study of HCl leaching showed that REE release is limited by the capacity of the liquor to dissolve the calcium sulfate host, which makes the liquid-to-solid ratio the dominant variable. Leaching, however, recovers the REEs only as a mixed liquor that must still be separated.
Molecular dynamics simulations showed that the organic phase used for this separation is far from an ideal solution: diglycolamide extractants self-assemble with co-extracted water and acid into metastable, polydisperse aggregates modulated by acidity, diluent polarity, and counterion identity. Accelerated weight histogram simulations at the liquid–liquid interface further showed that Nd3+–diglycolamide coordination becomes progressively more stable with depth into the organic phase.
These molecular insights, however, remain system-specific, whereas REE extraction is strongly multifactorial. A machine learning framework trained on approximately 3,000 experiments from about 75 publications predicted extraction for entirely unseen ligands with an R2 of 90.5%, and identified diluent properties, ionic radius, and the spatial localization of key ligand properties as its principal drivers. Symbolic regression was then used to derive what is, to our knowledge, the first interpretable analytical model of REE liquid–liquid extraction, allowing high-throughput in silico screening of both ligands and operating conditions.
Intralanthanide selectivity thus arises from a high-dimensional extraction landscape and can be tuned only through concerted optimization of its dominant drivers. Collectively, this work traces REE separation from the macroscopic recovery of a secondary feedstock, through the molecular organization of the organic phase and its interface, to transferable design rules, and establishes a multiscale basis for the rational design of more selective separation systems.