Performance of Metal-Catalyzed Hydrodebromination of Dibromomethane Analyzed by Descriptors Derived from Statistical Learning

Abstract

The catalyzed semihydrogenation of dibromomethane (CH 2 Br 2 ) to methyl bromide (CH 3 Br) is a key step in the bromine-mediated upgradation of methane. This study presents a cutting-edge strategy combining density functional theory (DFT), catalytic tests complemented with the extensive characterization of a wide range of metal catalysts (Fe, Co, Ni, Cu, Ru, Rh, Ag, Ir, and Pt), and statistical tools for a computer-assisted investigation of this reaction. The steady-state catalytic tests identified four classes of materials comprising (i) poorly active (<8%) Fe/SiO 2 , Co/SiO 2 , Cu/SiO 2 , and Ag/SiO 2 ; (ii) Rh/SiO 2 and Ni/SiO 2 , which exhibit intermediate CH 3 Br selectivity (<60%); (iii) Ir/SiO 2 and Pt/SiO 2 , which display great propensity to CH 4 (>50%); and (iv) Ru/SiO 2 , which exhibits the highest selectivity to CH 3 Br (up to 96%). In-depth characterization of representative catalysts in fresh and used forms was done by X-ray diffraction, inductively coupled plasma optical emission spectroscopy, N 2 sorption, temperature-programmed reduction, Raman spectroscopy, electron microscopy, and X-ray photoelectron spectroscopy. The dimensionality reduction performed on the 272 DFT intermediate adsorption energies using principal component analysis identified two descriptors that, when employed together with the experimental data in a random forest regressor, enabled the understanding of activity and selectivity trends by connecting them to the energy intervals of the descriptors. In addition, a representative analytic model was found using the Bayesian inference. These findings illustrate the exciting opportunities presented by integrated experimental/computational screening and set the fundamental basis for the accelerated discovery of superior hydrodebromination catalysts and beyond.

Nuclearity and Host Effects of Carbon‐Supported Platinum Catalysts for Dibromomethane Hydrodebromination

“…The values for the NP-based systems, ≈24 and 27 kJ mol −1 , are lower than those attained over platinum nanoparticles supported on SiO 2 (34 kJ mol −1 ). [31] Rather, these apparent activation energies are in the same range as the values obtained over SA-based catalysts (37 and 39 kJ mol −1 ), in line with the comparable TOF over these materials. From the reaction profiles, the energies required for the most energetic steps are taken as representative for the apparent activation energies.…”

“…[1] This goal lower due to coking and over hydrogenation to CH 4 , respectively. [31] Despite the promising initial selectivity to CH 3 Br, coking and sintering cause the rapid deactivation of Ru-based catalysts. On the contrary, platinum nanoparticles of ≈2 nm show the highest stability all systems investigated.…”

“…[ 30 ] Among the platinum group metals, selective CH 2 Br 2 hydrodebromination to CH 3 Br was reported over Ru/SiO 2 (up to 96%, Table 1 ), whereas the CH 3 Br selectivity over Rh/SiO 2 (<48%), Ir/SiO 2 (<40%), and Pt/SiO 2 (23%) was lower due to coking and over hydrogenation to CH 4 , respectively. [ 31 ] Despite the promising initial selectivity to CH 3 Br, coking and sintering cause the rapid deactivation of Ru‐based catalysts. On the contrary, platinum nanoparticles of ≈2 nm show the highest stability all systems investigated.…”

“…[28][29][30] Therein, limited progress has been made toward selective hydrodebromination, where carbon losses in the form of CH 4 and coke represent a major challenge. [30,31] A recent study evaluated the performance of SiO 2 -supported nanoparticle-based (NP-based) metal catalysts (1 wt% of Fe, Co, Ni, Cu, Ru, Rh, Ag, Ir, Pt). Therein, at comparable reaction conditions, iron-, cobalt-, copper-, and silver-based catalysts displayed CH 2 Br 2 conversion levels of 4-7%, showing consistency with the reported poor hydrodebromination ability of these elements.…”

Turning chemistry into information for heterogeneous catalysis

“…This automatization procedure has been used to generate the databases for further machine learning studies. [55] T A B L E 1 Minimum metadata stored in ioChem-BD for each calculation…”