Selectivity is the week's theme as chemists steer reactions away from side products

The epoxidation fork

The clearest statement of the week's theme comes from a preprint on heterogeneous epoxidation catalysis, where the same surface can produce the product you want or burn it, depending on what else is present. Surface metal-oxygen species exhibit up to 100% selectivity toward olefin epoxidation, the preprint reports, but the co-existence of surface metal-oxygen species and O2 leads to combustion via an oxidative carbon-carbon double-bond cleavage pathway. Epoxidation adds an oxygen atom across a double bond to make an epoxide, a building block for other chemicals; combustion breaks the bond entirely and loses the product. The finding is that selectivity is not a property of the catalyst alone but of the conditions around it: deliver the oxidant in a controlled way and the surface makes epoxide, let molecular oxygen join in and the same surface burns the starting material. Controlled oxidant delivery is therefore not a refinement of the process, it is the process. A plant that cannot control what oxygen species are on its catalyst surface cannot control what comes out the other end. m:…4887

A second preprint attacks selectivity from the other side: catching an intermediate before it disappears. Lewis-acid-assisted generation of cationic arylpalladium(II) intermediates enables the capture of fleeting enamines, the preprint reports, and the palladium-catalyzed alpha-arylation of acyclic ketones exhibits broad functional-group tolerance under near-neutral conditions. Enamines are reactive intermediates that form and vanish quickly, which is why they are usually hard to use in synthesis. The method generates the palladium intermediate that can catch them, and does it under near-neutral conditions, which matters because harsh conditions would destroy the functional groups a chemist wants to keep. Broad functional-group tolerance is the practical payoff: a reaction that works only on simple molecules is a curiosity, while one that tolerates the groups found in real drug-like molecules is a tool. The paper's contribution is a way to make a fleeting species stay still long enough to be useful. m:…5873

Light and cobalt set the stereochemistry

Two more preprints in the week use metals to control which mirror image of a product forms, the question of stereochemistry that decides whether a molecule is a drug or its useless twin. A carbazole-decorated biimidazole ligand enables intraligand charge-transfer photoactivity and asymmetric induction in a single nickel catalyst, one preprint reports, and the system enables enantioconvergent light-mediated carbon-carbon cross-coupling of a pyrrolidine-derived trifluoroborate with aryl bromides. Enantioconvergent means that both mirror-image forms of the starting material converge to a single mirror-image form of the product, which is the efficient outcome: nothing is thrown away. Reversible nickel-carbon bond homolysis is the mechanism for stereochemical scrambling in this enantioconvergent process, the authors report, which is the kind of mechanistic detail that lets the next chemist design a better version rather than copying the conditions blindly. m:…4609

A cobalt-catalyzed method reported in a separate preprint makes the same point with a different metal and a different reaction. Cobalt-catalyzed asymmetric enyne cyclo-1,4-dienylation enables the synthesis of enantioenriched five- and six-membered rings with 1,4-diene moieties, and the reaction mechanism proceeds via oxidative cyclometallation, beta-H elimination and reductive elimination. The method remains efficient at the gram scale with excellent enantioselectivity. Gram scale matters because a reaction that works only in milligrams is a demonstration, while one that works in grams is a candidate for real synthesis. The two papers together show the field's current answer to stereochemistry: design one catalyst that does two jobs, using light in one case and a base metal in the other, and let the mechanism do the sorting. m:…4610

Breaking bonds cleanly, and predicting the outcome

The week's fifth paper is about breaking a bond rather than making one. A ruthenium catalyst, RuCw at 5 weight percent, achieves 92.0% conversion of diphenyl ether and 63.6% aromatic yield at 200 °C in catalytic transfer hydrogenolysis, a preprint reports. Transfer hydrogenolysis using 2-propanol yields higher aromatic selectivity than molecular H2 for diphenyl ether hydrogenolysis. The choice of hydrogen source turns out to be a selectivity lever: the alcohol donor gives more of the aromatic products and less of the unwanted ones than hydrogen gas does. Support-dependent rate differences between the two ruthenium catalysts tested are primarily due to the accessible ruthenium population as measured by CO chemisorption, which means the support matters because it changes how much ruthenium is actually available to react, not because of some subtler effect. For a process chemist, that is a useful kind of result: it says where to look when the catalyst underperforms. m:…4608

The last paper in the set does not report a reaction at all; it reports a way to need fewer of them. A preprint on active learning for predicting reaction performance reports that EPIG reduces the number of experiments required for Wittig olefination transfer learning by approximately 50% compared with BALD, and that EPIG achieves the strongest early learning performance on the majority of 118 retrospective active learning tasks across four reaction datasets. The point connects directly to the rest of the week: every selectivity result above was found by running reactions, and a method that halves the experiments needed to learn how a reaction behaves is a method that makes all the others cheaper to find. Taken together, the week's catalysis papers describe a field that is less interested in discovering new reactions than in controlling the ones it has: deliver the oxidant carefully, catch the intermediate, let one catalyst do two jobs, choose the hydrogen donor, and run half as many experiments to learn the same thing. m:…4612