University of Leeds
DFT methods for high-throughput investigations of base metal complexes
Abstract
dc:description.abstractThe global energy transition demands the dramatic increase in the extraction and recovery of critical metals, including lithium, nickel, cobalt, and manganese, to meet the demand for battery electric vehicles. One solution to relieve the current supply pressures of critical metals is to recycle spent lithium-ion batteries. This work addresses the challenge of in silico screening for lithium tight-binding ligands for hydrometallurgical recycling processes and performance of DFT methods modelling 3d transition metal complexes. Chapter 2 presents an in-silico screening workflow for identifying lithium binding ligands from the Cambridge Structural Database (CSD). The workflow was split into four stages: (1) Quantitative analysis of ∼12,000 lithium coordination environments; (2) Using results from stage 1 the CSD was scanned for ∼22,000 candidate ligands, which then had a lithium centre added to build candidate complexes; (3) The geometries of the candidate complexes were optimised with GFN2-xTB/UFF, which were then filtered by their geometry; this process was repeated again using HF-GD3BJ/6-31G*-ACP resulting in ∼700 candidates; (4) The binding energies for lithium with the ligand candidates were calculated with M06-2X/def-SVP and were then filtered based on their binding energies; this process was repeated again with M06-2X/def2-TZVP, resulting in a final set of 80 tight-binding lithium ligand candidates. The development of the workflow was supported by two studies, one benchmarking the performance of semi-empirical, HF, and DFT methods predicting lithium complex geometries and the other benchmarking HF and DFT methods predicting lithium-ligand binding energies. Chapter 3 presents work on the development of geometric descriptor used for filtering lithium-ligand coordination vi complexes. Chapter 4 presents the experimental measurements of lithium-ligand binding energies with 7Li NMR and ESI-HRMS, however, this also revealed the practical challenges in measuring lithium-ligand binding energies in aqueous solution. Chapter 5 presents a benchmark study of popular DFT methods predicting the ground state multiplicity and thermal spin crossover temperatures of 3d transition metal complexes. Single-reference DFT methods struggle to model multi-reference systems such as 3d transition metal complexes. Thus, Chapter 5 presents a comprehensive DFT bench marking study assessing the performance of DFT functionals (B3LYP, PBE0, TPSSh, and M06-L) for predicting the ground state multiplicities of a large and diverse set of 127 3d transition metal complexes. Prediction of the correct ground state multiplicity is important for any future in silico screening workflow for identifying nickel, cobalt and manganese selective ligands. DFT was reliably able to reproduce the metal complex geometries and achieve ground state multiplicity prediction accuracies of ∼90%. The predicted adiabatic electronic energy gap between the two lowest Gibbs free energy multiplicities (∆E1−2) proved to be an important descriptor of the reliability of the DFT calculations. DFT prediction of ground state multiplicity was considered unreliable if ∆E1−2 was less than ∼20 kcal·mol−1 and reliable if greater than said ∆E1−2 value. To further probe the reliability of DFT predicting 3d metal complex energetics, 31 Fe(II) complexes with solvent phase thermal spin crossover values were used to benchmark the performance of DFT methods. A range of DFT functionals were tested, including ωB97X, PBE0, BP86, r2SCAN, TPSSh and B3LYP. The best methods were M06-L (R2 = 0.75, RMSE = 91) and r2SCAN (R2 =0.66, RMSE = 52). However, in both benchmarks predicting ground state multiplicity and thermal spin crossover values, DFT is inherently not sufficiently accurate because it is a single-reference method. Hence, the semi-empirical multi-reference method Kestrel, which performed well for predicting thermal spin crossover values (R2 = 0.71, RMSE = 22). This work demonstrates the power of high-throughput computational chemistry in efficient exploration of large chemical spaces but also the current limitations of computational chemistry methods and assays. Typically, greater computational resources (time, RAM, and CPU count) are required for more accurate methods and assays. Consequently, we undertook benchmarking of the accuracy of these methods to establish, for our own purposes, the lowest computational cost that can be adopted before an unacceptable loss of accuracy is incurred.
Degree
thesis:*- Name dc:type.qualificationname
- Ph.D
- Level dc:type.qualificationlevel
- doctoral
- Grantor dc:publisher.institution
- University of Leeds
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Mace, Samuel
- Advisors dc:contributor.advisor
-
- Nguyen, Bao
- Xu, Yingjian
- ORCID person:identifier.orcid
Rights
- Language dc:language.dcterms_iso639_2
- eng