(630e) Experimental and Statistical Approach for the Study of Adsorption Properties At Zero Coverage On the Metal-Organic Framework MIL-47

Adsorption properties of MIL-47 were measured by pulse gas chromatography between 180°C and 250°C for a set of 20 hydrocarbons consisting of aromatic compounds and heterocyclic molecules. The adsorption enthalpy (DH) and Henry constants (K’) were determined at zero coverage; van ‘t Hoff plot correlations (r²) were above 0.99 in all cases. In general, good correlations between adsorption enthalpies and Henry constants were found. Trends of unsaturated bonds’ presence, molecular size and presence of hetero-atoms can be observed in the experimental data (e.g. Henry constants, retention time, entropy contribution...). Strong influence of the adsorbates’ polarizability and electronegative character is observed when hetero-atoms are present. The latter can be related to the interaction with the metal clusters. For a series of benzene alkyl derivates a linear trend is observed. For each additional carbon added, the adsorption enthalpy deceases stepwise: -6.2 kJ/mol for the benzene, toluene, ethyl benzene and propyl benzene series, -6.1 kJ/mol for the benzene, toluene, xylenes and mesitylene series.

Statistical data analysis was applied to model zero coverage adsorption enthalpy and Henry constants. The objective was to establish quantitative structure property relationship (QSPR) models that could qualitatively and quantitatively describe the adsorption properties in the gas phase on this Metal Organic Framework. Principal component analysis is used to divide the experimental data set in a training set (model establishing) and external validation set, using a large descriptor set containing physicochemical and structural parameters of the adsorbates. The descriptor set was reduced by elimination of highly correlated descriptors. Partial least square regression was performed in order to obtain linear models using small descriptor sets containing some structural descriptors and physicochemical properties of the adsorbate molecules. A bootstrap methodology was used to further eliminate non-relevant descriptors on the likeliness of a descriptor coefficient being equal to zero.  The number of bootstraps was set to 500.

The models were able to fit and predict the adsorption properties to a very high degree (r² and q² between 0.99 an 0.95) using only 5 adsorbate descriptors: polarizability, logarithmic value of vapor pressure and the topological descriptors Kappa 1-3. Strong influence of the adsorbates’ polarizability and vapor pressure was observed. An average difference between experimental and calculated values of 2 kJ/mol (expressed by root-mean-square-of-error-prediction) for both training and external validation sets for adsorption enthalpies is obtained.