Structure Sensitive Reaction Kinetics of Chiral Molecules on Intrinsically Chiral Surfaces
Kareem Abdelmaqsoud, Michael Radetic, Carlos Fernández-Cabán, Michael Widom, John R. Kitchin, and Andrew J. Gellman, “Structure Sensitive Reaction Kinetics of Chiral Molecules on Intrinsically Chiral Surfaces”, The Journal of Physical Chemistry C 128(33), 13879-13887 (2024). https://doi.org/10.1021/acs.jpcc.4c04224
This summary was written by a language model from the paper and has not been checked line by line. Rely on the published paper for details.
Question
Metal surfaces made of low Miller index terraces separated by kinked steps lack mirror symmetry, so they are intrinsically chiral and come in R and S forms. Such surfaces can make the two enantiomers of an adsorbed chiral molecule react at different rates, which is the basis of enantiospecific catalysis and separation. Because there are infinitely many chiral orientations of an fcc crystal (48 stereographic triangles, 24 of each handedness, all equivalent by symmetry), the design question is which surface structure maximizes enantiospecificity. The paper asks whether data-driven models trained on kinetic measurements across hundreds of orientations can describe this structure sensitivity and predict the most enantiospecific Cu surface (Abdelmaqsoud et al. 2024).
Approach
The probe reaction is the decomposition of D- and L-tartaric acid (TA) on Cu, which proceeds by a vacancy-mediated surface explosion: a slow initiation step followed by an autocatalytic explosion step, each with its own rate constant. Previously reported experiments used three curved Cu surface structure spread single crystals (S4Cs) centered on (100), (110) and (111), each spanning orientations up to 14 degrees apart. Starting from a saturated TA monolayer, the crystals were heated to an isothermal 433 K in short cycles and the coverage was mapped by spatially resolved XPS (O 1s) at 169 points per crystal. Electron backscatter diffraction pole figures were used to assign Miller indices to each point. The enantiospecificity of each orientation is the difference between the D-TA and L-TA decomposition half-times, built using the diastereomeric relationship (L-TA on an S surface behaves like D-TA on an R surface), giving 507 data points (Abdelmaqsoud et al. 2024).
Two linear models were built. The first represents each surface by generalized coordination numbers (GCN). Using Pymatgen slabs for 270 orientations spanning the stereographic triangle, the authors found 58 distinct GCN values among surface atoms (atoms with coordination below 12), and each surface became a 58-element vector of the fraction of surface atoms with each value, interpolated linearly to the experimental orientations. Because GCN features cannot tell R from S, the data were split into two diastereomeric sets and separate models predicted the half-times of each enantiomer, with their difference giving the enantiospecificity. The second model expands the half-time difference over the sphere of orientations in five chiral cubic harmonics, combinations of spherical harmonics that have the cubic symmetry of fcc Cu and are antisymmetric under inversion and reflection; the simplest of these is ninth degree. Both models used ridge regression with an 80/20 train/test split and the regularization strength chosen by grid-search cross-validation (0.01 for GCN, 0.12 for harmonics) (Abdelmaqsoud et al. 2024).
Key findings
The measured half-time differences change sign between regions of each crystal following its 4-fold, 2-fold or 3-fold symmetry and vanish along the achiral high-symmetry directions. Handedness alone does not fix the sign: R surfaces do not always favor the same enantiomer, so both structure and orientation matter. The GCN model gave mean absolute errors of 141 s (train) and 138 s (test); the harmonics model, with only 5 features against 58, did slightly better at 129 s and 133 s. Both reproduced the symmetry and orientation of the data on all three crystals. Extrapolating over orientations with Miller indices up to 30, the GCN model predicted Cu(11,3,1) and the harmonics model Cu(17,5,2) as the most enantiospecific. These are only 1.8 degrees apart, so one S4C could test both, and the predicted locations stayed the same when the models were retrained on different random splits, although the predicted magnitudes varied. The harmonics model gives smoother predictions because its basis functions vary smoothly with orientation, while GCN features do not. The GCN model is more interpretable: a kink motif made of atoms with GCN values of 4.8, 5.6, 6.4 and 9.3 correlates with larger predicted half-time differences, and (11,3,1) has the highest fraction of atoms in that motif (Abdelmaqsoud et al. 2024).
Significance
The work gives two complementary ways to map enantiospecific kinetics across the full space of chiral orientations of a cubic metal, one physically interpretable and one built from symmetry alone, and uses them to choose the next experiment. The authors stress that extrapolation is uncertain and must be checked experimentally. They also note that TA decomposition may reconstruct the surface, as seen for aspartic acid on Cu, which could disrupt the identified kink structure; the GCN model assumes the reconstructed surface still reflects the starting structure, while the harmonics model makes no structural assumption, and the agreement of the two is taken as support. The framework is presented as applicable to any enantiospecific reaction on any chiral material with a cubic lattice (Abdelmaqsoud et al. 2024).