For the MIL-88B synthesis FeCl₃·6H₂O (180 mg, 1.1 mmol), 1,4-H₂BDC (120 mg, .72 mmol) with benzoic acid as a competing ligand (Yang et al., 2019) (22 mg, .18 mmol) in 6 mL DMF which was heated at 145°C for 1.5 h in solvothermal reaction. The collected product was washed with DMF, water and IPA, centrifuging at 6,000 rpm for 15 min, before the final drying and activation.

A palladium (Pd) suspension of 280 mg palladium acetate particles and 580 mg of polyvinylpyrrolidone (PVP) acting as a stabilising agent, was mixed in 50 mL of methanol. .29 mL of 100% acetic acid was added to the solution containing palladium acetate prior to mixing with the zinc source, .745 g (∼2.5 mmol) of zinc nitrate hexahydrate. A second solution of 1.642 g (∼20 mmol) of 2-methylimidazole and 1.98 mL (∼20 mmol) of n-butylamine was created in 50 mL of methanol. When dissolved, the two solutions were mixed, inducing sol-gel formation (Hunter-Sellars et al., 2021). After 24–72 h, this mixture was then centrifuged for 10 min at 3500 rpm (to avoid sintering of particles), and then cleaned using methanol before the final drying and activation.

Low porosity MIL-100 was obtained following the procedures outlined in Kyong-Seo et al. (Seo et al., 2012), in which n-butylamine, 1.29 mL (∼13 mmol) was used as a capping reagent in the iron nitrate, 8.016 g (∼20 mmol), trimesic acid, 2.752 g (∼13 mmol), hydrothermal reaction (20 mL water, 160°C). After 12 h at 160°C, the mixture was cooled, and the product was obtained by centrifugation at 10 min at 3,500 rpm. The product was then cleaned using ethanol and D.I. water and allowed to dry before a final activation.

To assess the activity of the MOF and ZIF catalysts, reactions were carried out in batch and continuous flow modes. The post-reaction mixture was then passed through a .2 μm PTFE filter before being analysed using a GC-FID/MS (Shimadzu GC-2030/GCMS-QP2020 NX) equipped with a Shimadzu SH-Rxi-5ms column (30 m, .25 mm ID, .25 μm) Crossbond^® (5% diphenyl/95% dimethyl polysiloxane).

In a 10 mL round bottom flask, levulinic acid, 1 mmol, was added to 3 mL of isopropanol, along with 1 mmol NaOH and 3 mg of catalyst. TFH reactions were carried out at 100°C, continuously stirred (250 rpm), and refluxed for 15 h. Where pH measurements were taken, a Mettler Toledo (FiveEasy) pH probe was inserted into the reaction mixture at intervals over the course of the reaction. The pH was left to stabilise before taking any readings.

All three catalysts were reused for five cycles. Catalytic recycling was performed using the following process. Post-reaction, the catalyst was separated from the product mixture by centrifugation at 7,000 rpm for 20 min, with further centrifugation occurring after each washing step. The catalyst was then washed with IPA, twice with a 50/50 IPA/water mixture, and finally with IPA again before drying and reuse. Catalysts were activated initially under vacuum at 180°C, but not between cycles.

ZLC measurements were made using a Shimadzu Prominence HPLC, with the sample held within a re-fitted Phenomenex guard cartridge. Experiments were made using .1 mM of adsorbate in isopropanol (IPA). Following equilibrium and saturation of the adsorbent pores, the mobile phase was switched to pure IPA and the desorption curve was used to determine the counter-diffusion constants. Data was collected via an in-line UV/Vis detector and temperature control was carried out using the column oven, with measurements made at 20, 40, 60, and 80°C, and .5, and 1.0 mL/min flow rates to ensure there was consistency in diffusion time constants with varying flow rate (Brandani and Mangano, 2021). Blank measurements were taken using a column of non-porous glass beads, to correct for system effects. A revised ZLC model, Eq. 1, used in this work, has been made to account for fluid hold-up effects, either by dead volume or by adsorption interactions (Verbraeken et al., 2021): ln C C 0 = ln 2 L β 2 + 1 − L + γ β 2 2 + L − 1 + γ β 2 − β 2 D e . t R p 2 (1)

A linear plot of ln(C/C₀) against time in the long-time region yields a gradient of β²D_e/R_p ². The dimensionless ZLC parameter, L, which should be greater than 1, can be found via the intercept and confirmed using Eq. 2 (Brandani, 2016): L = 1 3 U V s R p 2 K . D e (2) where, V_s is the solid volume, U is the mobile phase flow rate, K is the adsorption equilibrium factor, D_e is the effective diffusivity and R_p is the pellet radius.

Pyridine adsorption was carried out using a frontal analysis method, a break-through type uptake method, in cyclohexane (.1–10 mM), with a flow rate of .25 mL min^-1. Comparisons were made to a non-porous glass bead column for dead volume correction.

Mass transfer coefficients were obtained for the substrate LA using the same experimental set-up as in the ZLC measurements. Following the method laid out by Brandani and Mangano (Brandani and Mangano, 2022), larger concentration changes were used, with adsorptive column saturation at 100 mM LA in IPA, followed by a switch to pure IPA. The adsorbed concentration was calculated via ZLC/frontal analysis using the same experimental set-up as ZLC, with dead volume measured using a non-porous glass bead column of identical particle size. Where the flow rate is low, adsorbed concentrations at any time may be found using, Eq. 3: M s n ¯ − n ¯ 0 = ∫ 0 t F c A d s o r b e n t d t − ∫ 0 t F c B l a n k d t (3) where, n₀ = 0 in an adsorption experiment.

With the change in adsorbed concentration over time provided by a high flow rate experiment and Eq. 4: M s d n ¯ d t = F C I N − F C o u t − V F d c c t (4)

The mass transfer coefficient was found using Eq. 5: k . a = d n ¯ d t n e q − n ¯ (5) where, k is the mass transfer coefficient, a is the surface-to-volume ratio, and n_eq is the equilibrium adsorbed concentration, found from equilibrium adsorption isotherms, and n ¯ is the average adsorbed phase concentration.

During this investigation, it became apparent that overloading of the catalyst with LA substrate (>1%) prevented catalytic behaviour and lead to the formation of reaction side products. As this issue was ubiquitous for all catalysts, with a variety of pore distributions, it was assumed that this issue arose from the mass transfer of substrate to the catalyst, necessitating the need for the characterisation of mass transfer coefficients. This same decrease is observed by Wan et al. (2021) for a zirconium catalyst, but not to the same extent.

The catalytic conversion of levulinic acid is shown in Figures 4A–C. From these graphs, it can be observed that the catalyst pore size and surface area have little effect on the conversion, with ZIF-8 producing the highest conversion consistently. From the GC-MS yield, it is possible to infer that the smaller pore diameter imposes restrictions on side product formation, with the smaller pore aperture of ZIF-8 producing a product selectivity consistently above 80%. Where the pore apertures are larger, as in MIL-100 and MIL-88B, the selectivity decreases over time, which is due to the formation of local areas of high substrate concentration, or framework loss due to the basic conditions, producing the effect observed in substrate overloading. Surprisingly, the Fe-based MOFs performed comparably to the Pd catalyst but did not possess comparable material stability. For all materials, the conversion and subsequent turnover frequency (TOF) is low, especially when compared to like MOFs such as UiO-66(Zr) (Valekar et al., 2016), likely a result of low reaction temperature (Belguendouz et al., 2021).

All three catalysts were reused for five cycles. Catalytic recycling was performed after washing and drying. Over the five cycles, all the reused catalysts showed a marked drop in conversion, and a decrease in selectivity was found for the Fe-based materials. In the case of MIL-100 the drop in reactivity over five cycles is significant, with a permanent decrease in conversion seen after two cycles. This sudden loss of reactivity is due to the hydrolytic breakdown of the hydrophilic framework, via base hydrolysis as a result of the high pH (Bezverkhyy et al., 2016), reducing the surface area and access to catalytic centres (López et al., 2021); see Supplementary Information for post-reaction (5 cycles) XRD. The same process likely occurs in MIL-88B, but at a lower rate. During the batch reactions, it becomes clear that the monoliths would slowly break up into smaller particles, due to stirring actions. The larger surface area of these smaller particles may make up for any surface loss activity, leading to the variations in activity from batch-to-batch (Vasanthakumar et al., 2020). Alternatively, the hydroxylation of the metal centres, in the presence of -OH, may hinder substrate-catalyst interactions, also contributing to variations in batch-to-batch activity. It is expected that some of the Fe-metal leached due to the relatively high pH. However, due to the lack of maintained catalytic activity (which would be expected if homogenous catalysis was occurring) it is supposed that these nanoparticles do not contribute towards the selective catalysis of GVL.

In the case of ZIF-8@ Pd(NO₃)₂, there is a slow decrease in reactivity, followed by a sharp decrease; it is theorised that this may be the result of Pd nanoparticle agglomeration or more likely substrate build-up resulting in pore blockages. A comparison of ZIF-8@Pd(NO₃)₂ XRD patterns pre- and post-reaction, Supplementary Figure S4, reveals the retention of Pd(NO₃)₂ peaks, ruling out significant leaching, suggesting pore blockages are the main contribution to the reduction in catalytic conversion. Any conclusions of Pd leaching by XRD here are qualitative, not quantitative.

To assess the possible build-up of a substrate on the surface, IR measurements were performed on the catalysts after five cycles, Figure 5, post-drying under ambient conditions.

The effective diffusivities of LA and GVL in MIL-88B, MIL-100 and ZIF-8@Pd(NO₃)₂ were calculated using the ZLC method. This analysis allows the relationship to be determined between porosity and diffusion with catalyst lifetime and effectivity. Figures 7A, B shows the ZLC desorption curves for LA, with the calculated diffusivity values shown in Table 2. To ensure that the system noise did not contribute to the diffusion measurements, a range of C/C₀ .01–.05 was used as the long-time region of analysis; any curves below this range are within the range affected by noise and therefore were discounted.

The short-time region displays macropore diffusion contributions with: Z I F − 8 @ P d N O 3 2 < M I L − 100 < M I L − 88 B with MIL-88B having the largest macropore volume. The opposite may be said in the long-time region, between the dashed lines, where micropore control is observed as: Z I F − 8 @ P d N O 3 2 > M I L − 100 > M I L − 88 B

with ZIF-8@Pd(NO₃)₂ having a smaller pore aperture. Unexpectedly, from Figures 7C, D the large GVL molecule shows no diffusion limitation in any of the materials. Therefore, as GVL may have a slightly larger critical diameter than LA, it is likely that any LA diffusion limitations arise increasing surface diffusion effects, facilitated by the much stronger interactions between catalyst and substrate, compared to catalyst and product (Yuta et al., 2009; Jakob et al., 2021). Again, for a reaction to be efficient the removal of products from the surface and tortuosity is important as it drives the reaction forward. Though the MIL-88B material does show some tailing, and a slightly lower apparent diffusion constant than the ZIF-8 of MIL-100, this tailing is not the result of kinetic limitation as the diffusion time constant varies with both flow rate and particle diameter. In the case of MIL-88B, any tailing in the desorption curve is the result of equilibrium control effects not being fully resolved by the blank subtraction.

From these diffusion constants, the FT-IR data observed in Section 3.2.8 can be rationalised. Since the diffusion of LA is slower in ZIF-8 and MIL-100, there is a greater build-up of molecules around the surface, when compared to MIL-88B, leading to the post-reaction IR alkyl bands observed. The slower diffusion of LA over GVL may explain why substrate overloading is observed as the LA concentration is increased beyond .5 mmol, with 1% catalyst loading.

Adsorption equilibrium isotherms for levulinic acid in IPA were calculated from low flow rate experiments, using the same ZLC set-up and a frontal analysis method. Figure 8A shows the relative shape of the isotherms with both microporous materials ZIF-8 and MIL-100 displaying type 1 isotherms, with MIL-88B displaying a type 2 isotherm owing to the increased meso- and macroporosity. From these isotherms the bulk concentration in a high flowrate, high concentration swing ZLC experiment may be related to adsorbed concentration, n_eq. Further, the average adsorbed phase concentration, n ¯ , may be found over a set time, t, by determining the integral of the high flowrate experiments. Finally, dn/dt was found using Eq. 4, where the fluid phase volume, V_f, and column mass, M_s, is shown in Supplementary Table S1.

In Figure 8A, the isotherms were modelled using the Langmuir model in the case of both ZIF-8 Pd(NO₃)₂ and MIL-100; for MIL-88B, an adapted liquid phase BET model (Gocho et al., 1998; Girods et al., 2009) was applied due to the observable mesopore filling.

The resultant mass transfer coefficients, displayed in Figure 8B, show a concentration dependence which dramatically increases above n/n_q = .6. On average, the mass transfer coefficient for the ZIF-8 was larger than both MIL-100 and MIL-88B, owing to the increased surface area and framework hydrophobicity (Hunter-Sellars et al., 2021), see Table 3. The increased mesoporosity of MIL-88B will facilitate mass transfer more over the lower surface area of MIL-100. The barriers to mass transfer, such as surface defects and catalytic site-substrate interactions, may be found at the surface and have been cited as being influenced by surface heterogeneities (Hu et al., 2021). Therefore, it can be argued that although the ligand modulation increases the pore size distribution of a MOF, it can also produce surface defects that inhibit the entrance of molecules to the pores of crystals, increasing the contribution of surface diffusion toward the overall mass transfer process.

From this mass transfer data, it is clear that the overall mass transfer of reactants into the catalyst is the crucial factor for product selectivity, where there are side reactions occurring in the bulk. As frameworks collapse over repeating cycles (for MIL-100 and MIL-88B), this mass transfer decreases further, further reducing GVL selectivity.

The calculated mass transfer coefficients ka were of a similar magnitude to those found by Grant Glover et al. (2008) for water in silica gel and hexane in activated carbon. And similar to those determined by Štěpánek et al. (2000) also for water vapour in silica gel.

It was found by Shao et al. that the acid and base site concentrations were important for LA to GVL reaction selectivity, with Lewis Acid sites (LAS) causing increased GVL ring opening (Shao et al., 2022). For this reason, the frontal analysis adsorption uptake of pyridine was measured. Assuming a 1:1 stoichiometry between the adsorbed pyridine and acid site (Fe, Zn, or Pd), the number for the total acid sites may be expressed in equivalent number of acid sites per unit area (mol equiv. g^-1). The data from the pyridine uptake is shown in Figure 9, it is probable that the pyridine molecule is too large to enter the pores of the ZIF-8 monolith as the uptake is almost negligible. For MIL-100, the base uptake is initially higher than MIL-88B, due to the smaller pores improving uptake, however, the larger pore volume eventually leads to a higher observed uptake for MIL-88B. The larger number of LAS in MIL-88B was also reported by Yu et al. (2019). In all cases, it appears the isotherm plateau was approached, but not completed, and therefore the number of sites shown in Table 4 should be used for internal comparison, though it is unlikely the real value varies significantly.

From Table 4 it can be observed that the total equivalence of LAS calculated for the materials is also of a similar magnitude to those found Yu et al., in which the total LAS number was .24 and 4.93 mmol g^-1 for MIL-100 and MIL-88B respectively.

It is likely that under the basic conditions provided by the NaOH, the total number of available LAS is decreased, leading to preferential GVL selectivity over the hydrogenolysis product 1,4-PDO (1,4-pentanediol).

In terms of dictating selectivity, there was no trend attributed to the increased Lewis acid sites and GVL product selectivity.