(349e) Predicting Relative Stabilities of Divalent Cations in Metal-Exchanged Zeolites

Metal-exchanged zeolites possess well-established catalysis applications, yet identifying their Al distribution remains a key challenge. Density Functional Theory (DFT) computed cation-exchange energies of metal exchanged zeolites provide insights into the Al distribution. However, the number of cation-exchange locations dictated by unique Al configurations grow exponentially as the number of atoms in the zeolite unit cell grows. Further, different cations preferentially exchange at specific Al arrangements. For example, Cu²⁺ ions prefer Al pairs in six-membered rings of CHA zeolite, whereas larger ions may prefer Al pairs in larger rings. However, when generalizing to other zeolites, enumerating all Al pairs, cation locations and cation identities using DFT becomes computationally intractable. Here we develop a method to predict the relative energies (𝚫E) of divalent metal ions (M²⁺) exchanged at Al pairs in zeolites (Z₂M) using structural descriptors and Machine Learning (ML) models. First, we performed DFT geometry optimizations on selected structures of four zeolite topologies (CHA, TON, BEA, and MFI) and two ions (Cu²⁺ and Pd²⁺). Next, we trained a ML model for predicting 𝚫E using the features extracted from DFT-optimized structures. The model shows the local environment of M²⁺ is highly correlated (R²_test=0.98, MAE=11 kJ/mol) with the 𝚫E of the corresponding Z₂M site. This correlation suggests that 𝚫E is accurately predicted through distance-based descriptors. To avoid reliance on DFT-computed geometries for predicting 𝚫E, we developed a physics-based force field (FF), optimized the FF parameters using our DFT data, and used the FF to determine the Z₂M geometries. We next used the FF-optimized geometries to generate descriptors for our ML model and predicted the 𝚫E values for each species. This model, which combines a forcefield for geometry optimizations with a ML model for energy evaluations, captures the 𝚫E trend (R²_test=0.92, MAE=21 kJ/mol) for relative stabilities of divalent ions, without the need for DFT-calculations.