Metal-organic frameworks as kinetic modulators for branched selectivity in hydroformylation

Finding heterogeneous catalysts that are superior to homogeneous ones for selective catalytic transformations is a major challenge in catalysis. Here, we show how micropores in metal-organic frameworks (MOFs) push homogeneous catalytic reactions into kinetic regimes inaccessible under standard conditions. Such property allows branched selectivity up to 90% in the Co-catalysed hydroformylation of olefins without directing groups, not achievable with existing catalysts. This finding has a big potential in the production of aldehydes for the fine chemical industry. Monte Carlo and density functional theory simulations combined with kinetic models show that the micropores of MOFs with UMCM-1 and MOF-74 topologies increase the olefins density beyond neat conditions while partially preventing the adsorption of syngas leading to high branched selectivity. The easy experimental protocol and the chemical and structural flexibility of MOFs will attract the interest of the fine chemical industries towards the design of heterogeneous processes with exceptional selectivity.

The images were processed with Leica Application Suite 4.12.0. FT-IR spectra were measured on a ThermoFischer Nicolet iS50 FT-IR using its ATR cell.

SM 6
Based on literature procedure: [4] Zn(NO3)2 After cooling down the oven to room temperature, the vials were removed from the oven and the mother liquor was decanted. The crystals were collected and washed with fresh DMF (3 × 15 mL) and soaked in CHCl3 (15 mL) for 3 days with the replacement of the fresh CHCl3 each 24 h. The obtained crystals were stored in toluene until use.

PPh2-BDC loading
The MOF sample (3 mg dry mass) was placed in a NMR tube and suspended in DCl solution (0.1 mL; 20% in D2O) using an ultrasonification bath. The solids were then dissolved by adding DMSO-d6 (0.5 mL) and analyzed by NMR.
The solid of the reaction mixture was filtered by membrane filter, washed with DMF, H2O and EtOH and dried in the vacuum oven. Yield: 388 mg (81% calculated on dry MOF).

SM 11
Nitrogen Physisorption before and after catalysis (Table 1 Entry

Pore Size Distribution
Below are the Horvath-Kawazoe differential pore volume plots of the samples before and after catalysis under conditions of Table 1

Hydroformylation
General procedure for hydroformylation of 1-hexene ( Table 1 in the main text and  Supplementary Tables 5-9) A stock solution of Co2(CO)8 in 1-hexene was prepared inside the glove box. In a 1.5 mL GC crimp vial the MOF was weight in and activated at 150 °C overnight. The MOF was then suspended in 250 µL Co/hexene stock solution and the vial was closed. The vial put in a 50 mL Premex® autoclave and purged with Ar. The autoclave was then briefly opened (under a flow of Ar) and the septum was pierced with a needle. The autoclave was closed, then syngas (CO:H2 = 1) was introduced and the reaction mixture was heated to 100°C for 18 hours.
The reaction mixture was cooled to room temperature and the pressure was slowly released.
The samples were topped up with 1 mL acetonitrile and 50 µL were transferred in a GC vial filled with acetonitrile and p-cymene. The conversion and branched:linear ratios were obtined by GC-FID with p-cymene as external standard using the same response factor (Rf) for all aldehyde products. In addition to the hydroformylation products, the chromatograms showed traces of unknown compounds, which never exceeded 5% of the total area of 1-hexene and hydroformylation products.
General procedure for the substrate scope ( Table 3 in the main text) The MOF (10 mol% to the olefin for MOF-74(Zn), 1 mol% to the olefin for UMCM-1 derivatives) was placed in a crimp vial, which was closed with a crimp cap and pierced with a needle. In the case of MOF-74(Zn), the vial was placed into a round-bottom flask and activated at 150 °C in vacuum for 24 h. The round-bottom flask was allowed to cool to room temperature but remained under vacuum until it was introduced into a nitrogen-filled glovebox. UMCM-1 derivatives were stored in the glovebox and weighed in without pervious activation. Co2(CO)8 (1.5 mol%) was dissolved in the olefin (500 µL), and the whole solution was added to the MOF.

SM 16
The vials were placed into a 50 mL Premex® autoclave and purged with Ar several times. The valves to and from the autoclave were closed and the syngas line flushed once. Syngas pressure (CO:H2 1:1, 30 bar) was applied and the autoclaves heated at 100 °C for 17 h. The autoclave was allowed to cool down to room temperature before the pressure was released slowly over 15 min. The autoclave was flushed with nitrogen before it was opened to remove additional syngas.

Analysis of the conversion and branched to linear ratio
The content of the reaction vials was transferred into a 5 ml volumetric flask and filled with THF. The MOF was extracted for 30 min before 200 µl of the suspension was added to 800 µl of a solution of the internal standard (p-cymene 0.048 M in THF). This suspension was filtered and analyzed by GC-FID. The branched to linear ratio was calculated from the ratio between the integrals of the isomers assuming the same Rf. Conversion and aldehyde yield in Table 3 and Supplementary Table 10 were calculated with GC-FID upon calibration of the olefins and the linear aldehydes with p-cymene. The Rf of all olefin isomers were assumed the same. The Rf of all aldehydes were assumed the same. The total oxo products yield was determined as described below. Table 3

of the main text
The total yield of oxo products (aldehydes + aldol products), which is an indication of the performance of hydroformylation, was calculated since we always observed aldol products at the high conversion in Table 3 (See GC-MS data above). Desorbing the aldehydes from the pores of the MOFs without decomposing them after catalysis was a challenge.
The method to yield determination of the oxo products was: 1) measure the mass of the raw product after catalysis;

SM 17
2) calculate the mass of the pure oxo products by removing the mass of non converted olefin and of impurities detected by GC; 3) calculate mol of oxo products with the corrected mass and the molar mass of the corresponding aldehydes. This is a safe assumption since aldol products have molar mass multiple to that of the aldehyde.
The vial was placed in a 50 mL Premex® autoclave and purged with Ar. The autoclave was then briefly opened (under a flow of Ar) and the septum was pierced with a needle. The autoclave was closed, then syngas (10 bar, CO:H2 = 1) was introduced and the reaction mixture was kept at room temperature for 72 hours. After pressure release the sample was immediately placed inside the glove box and the supernatant reaction solution was decanted off. The residual MOF was quickly washed with 2 x 0.5 mL toluene and left open to the atmosphere to evaporate the excess solvent and submerged in toluene (5 mL) for 24 hours in order to leach the excess Co-species. This procedure was repeated twice.

Hydroformylation with HCo(CO)3(MixUMCM-1-PPh2)
Inside the glove box, a 1.5 mL GC crimp vial was charged with pre-treated MOF (4.5 mg dry mass). The MOF was then suspended in 250 µL neat 1-hexene. The vial was then closed and put in a 50 mL Premex® autoclave and purged with Ar. The autoclave was then briefly opened (under a flow of Ar) and the septum was pierced with a needle. The autoclave was closed again, the syngas (CO:H2 = 1/1) was introduced and the reaction mixture was heated to 100°C for 17 hours.

Hydroformylation with Co@MOF
Co@MOF (amount in Supplementary Table 11) was added to a 2 ml crimp vial in a nitrogenfilled glove box. 1-Hexene (500 µl, 4.0 mmol, 1.0 eq.) was added to all vials before they were closed with a crimp cap and taken out of the glove box. The vials were placed in the autoclave which was flushed several times before syngas pressure was set to 30 bar at room temperature.
The reactor was heated to 100°C leading to a pressure of 35 bar and the substrates were allowed to react for 16 h. The autoclave was cooled down to room temperature and the remaining syngas pressure was slowly released to avoid spilling.

Recycling of Co@MOF
After the first catalytic run, the reaction mixture was removed with a syringe. The MOF was washed once by 1-hexene (1 mL) and then the solvent extracted with a syringe. Fresh 1-hexene (500 µl, 4.0 mmol) was added again, the vials were closed and the reaction was carried out as stated above.

Recycling MOF
After the first catalytic run, the MOF was filtered off and washed with CHCl3 (3 x 10 ml) in the case of MixUMCM-1-NH2 (28%). MOF-74(Zn) samples were washed with THF (3 x 10 ml) and EtOH (3 x 10 ml) before they were purified by Soxhlet extraction (5 d, THF). The purified MOFs were used for hydroformylation following the standard procedure of Table 1 in the main text.

GC-MS Chromatogram (Table 3 Entry 1 without MOF in Main Text)
Supplementary Figure 10 Extracted ion chromatogram for the mass of the aldol product (210 u) and mass spectra of the two found compounds compared to the best fitting substance in the database. The two products are likely isomers of the depicted aldol product.

Interaction Energy Calculations:
All calculations were performed using the code CP2K [8] at density functional level of theory.
The semi-local PBEsol functional was adopted [9] using the DZVP-MOLOPT-SR-GTH gaussian basis set for all the atom types, [10] and a cutoff of 500 Ry for the plane wave auxiliary In CP2K the interaction energy can be calculated defining 2 fragments A and B. Two fragments corresponding to the MOF ( " ) and the catalyst ( # ) were defined in each case.
The adsorption/binding sites tested are shown Supplementary Figures 11 to 17 .

Monte Carlo Simulations
Raspa 2.0 package was used for Monte Carlo simulations. [11] The interactions are computed using Lennard-Jones potential with a cutoff of 14.0 Angstroms for dispersions and a Coulombic potential for charges. As for framework's atom types, the parameters for dispersions are taken from DREIDING force field [12] integrated with UFF [13] for the missing atom types (Mg, Co, Ni and Zn). The choice of the force field is motivated by the good match with adsorption experiments of alkanes in MOF-type frameworks. [14] The point charges are computed using the REPEAT scheme [15] to fit the PBEsol electrostatic potential from geometry optimized structures. These periodic DFT calculations were run using CP2K software package, utilizing the DZVP-MOLOPT-SR-GTH gaussian basis set for all the atom types and a cutoff of 500 Ry for the plane wave auxiliary basis set. The position of the frameworks' atoms is kept fixed in

SM 37
all the calculation. 1-hexene and aldehydes are modelled using TraPPE force field, [16] [17] where the only missing parameters are for the O-CH-CH-CHx torsion potential of the carbonyl group of branched aldehydes. These parameters were computed from quantum mechanics using the MP2/6-31G* method in Gaussian09 (Supplementary Figure 18) coherently with the TraPPE parametrization. Due to the similarity of the two species, only one of the two branched aldehydes is considered: 1 (2-methylhexanal) which is experimentally obtained with the highest ratio.
CO and H2 molecules are treated as rigid particles. The CO interaction parameters adopted in this work were specifically designed for adsorption in MOF-type materials. [18] The parameters for H2 are taken from the work of Marx et al. [19] and already validated for adsorption in MOFs. [20][21] Supplementary Figure 18 Torsional scan of the carbonyl group in a representative model for branched aldehydes. The torsional potential was computed using the MP2/6-31G* method and used to fit the parameters for the TraPPE force field. The plot shows the agreement of the fitting and the conformation of the molecule in the maximum and minimum points In the following paragraph we describe the protocol used to compute the affinity of the reactive species with the frameworks. First, the amount of 1-hexene inside the bulk frameworks at 30 bar and 100°C was obtained for each considered MOF by performing a Grand Canonical Monte Carlo (GCMC) simulation, [22] where the fugacity of the solvent is derived using the Peng Robinson equation of state. [23] The results of GCMC simulation were averaged for 5,000 cycles, after other 5,000 cycles of initialization. Depending on the concentration of 1-hexene in the framework we computed the volume of the cubic box that simulates the homogenous phase, imposing the same number of molecules and the homogenous density of 1-hexene as computed from GCMC in the empty box (0.00395 molecules per cubic Angstrom). Note that this corresponds to a macroscopic density of 6.56 mol/L, which is slightly smaller than the reported experimental value of 7.11 mol/L. However, to be internally consistent with the simulations we keep the value of 6.56 mol/L as the reference density for the homogeneous phase.
To compute the affinity with the frameworks, one molecule for each reacting component (i.e., H2, CO, linear heptanal and branched aldehyde 1) was added to the mixture with saturated 1hexane. The two simulation boxes, for the homogeneous and the crystal bulk phases, can exchange molecules according to the following rules: 1-hexene and aldehydes were allowed to swap identity between the two boxes and the gas molecules were allowed to be removed and reinserted in the other box. These two moves were possibly selected within the Monte Carlo cycle and attempted, similarly to the other standard molecular moves, i.e., translation, rotation, intramolecular displacement and reinsertion. The choice for the swap rules follows from the consideration that 1-hexene and aldehydes have a similar size and therefore there is a higher probability that a change of identity between them will be accepted, while for the smaller H2 and CO molecules, it is easier to find interstices between the solvent molecules to be inserted.

SM 39
For each pair of systems (i.e. MOF + homogenous), we ran ten independent simulations, executing 5,000 cycles of equilibration and 5,000 cycles of production. The final average and standard deviation for the reactant/product occupation is obtained by considering the output these ten runs for the block averaging, i.e., as the result of 50.000 production cycles. Since there are two molecules for each reactive species for each pair of systems (i.e., the framework and the homogenous box) we define as occupancy percentage (%occup.) the averaged probability to find the specie in the MOF's pore volume, divided by two. Therefore, one can observe %occup.=50% if the affinity is the same with

SM 40
The MOF-aldehyde affinity is therefore enthalpic and apparently not dependent on the type of isomer (linear or branched), i.e., it is not due to a steric confinement as in the case of "shape selectivity" seen for hydrocarbons in zeolites. In the non-charged system, the affinity of the gas molecule that are only weakly polar, remains almost unvaried and suggests that there is a minor effect due to non-covalent interactions. We conclude that the higher affinity of the gas molecules with the liquid 1-hexene is due to an entropic motivation. The 1-hexene saturated in the pore volume has a higher density and the confinement effect is higher inside the MOF: this results in a lower probability of forming interstices where the small gas molecules can fit, i.e., the cavitation contribute to the solvation energy of H2 and CO.

Pore Volume Calculations
The pore volume in the bulk frameworks, that was used to calculate the density of the saturated 1-hexene is computed with the "probe occupiable pore volume" (VOLPO) routine [24] as implemented in the Zeo++ v0.3 software package. [25][26] For the probing of the volume, 500.000 samples were used, together with the high accuracy (-ha) option in Zeo++. The radii of the atoms in the framework were taken as half of the Lennard-Jones sigma parameter of the force field (DREIDING, integrated with UFF for the missing atom types). This choice is coherent with the potential used for the Monte Carlo simulations.
As for the probe size, a diameter of 3.703 Å was utilized. This value is the average of the sigma parameters for the CH3 (3.75 Å) and the CH2 (3.675 Å) beads in TraPPE force field, that are, respectively, the head and the tail of the 1-hexene molecule.

DFT Calculations
Supplementary

Kinetic Analysis
The kinetic analysis was based on the empirical rate of formations of the branched and the linear aldehydes reported in the paper. [27] The two equations are shown below and are also reported in the full text.
The concentration of H2 and CO at different pressures in 1-hexene were calculated using the Soave modifications of the Redlich-Kwong equation (SRK) [28] and are reported in Supplementary Table 19.

SM 47
The concentrations of 1-hexene, CO and H2 within the pores of the MOFs were calculated by multiplying the concentration in the homogeneous phase by a factor Z derived from the Monte Carlo simulations (Supplementary Table 20). The Z factor for H2 and CO were calculated by using equation (4). This is consistent with the fact that an %occup. of 50 would give a Z factor of 1 and therefore no preference of a molecule to be either in the homogeneous or the MOF phase, while with %occup. of 0 one would find null concentration inside the MOF as both CO and H2 are found with %occup. < 50%.
The concentration of Co2(CO)8 was calculated from the catalyst loading relative to the 1hexene concentration. (H2:CO = 1) Their effect on the rate of formations of the branched aldehyde RB and of the linear one RL for catalysis within the pores of UMCM-1-NH2 is also shown.