Manometric real-time studies of the mechanochemical synthesis of zeolitic imidazolate frameworks†

We demonstrate a simple method for real-time monitoring of mechanochemical synthesis of metal–organic frameworks, by measuring changes in pressure of gas produced in the reaction. Using this manometric method to monitor the mechanosynthesis of the zeolitic imidazolate framework ZIF-8 from basic zinc carbonate reveals an intriguing feedback mechanism in which the initially formed ZIF-8 reacts with the CO2 byproduct to produce a complex metal carbonate phase, the structure of which is determined directly from powder X-ray diffraction data. We also show that the formation of the carbonate phase may be prevented by addition of excess ligand. The excess ligand can subsequently be removed by sublimation, and reused. This enables not only the synthesis but also the purification, as well as the activation of the MOF to be performed entirely without solvent.

Molybdenum (VI) oxide (99.5%) was purchased from Alfa Aesar. Calcium carbonate (99.5%) was purchased from BDH Chemicals. Ethanol (95%) was purchased from Commercial Alcohols. Methanol and isopropanol were purchased from ACP. All chemicals were used without further purification.

Instrumental Details
Time-dependent pressure and temperature profiles in large-scale milling reactions were collected using 250 mL PM GrindControl TM jars supplied by Retsch GmbH. The milling was performed in a Retsch PM 400 planetary mill operating at 300-350 rpm, with the addition of either 4-7 medium-size steel balls (m ≈ 32 g, V = 4 mL, d = 17 mm) in the model reaction experiments, or with 7 large steel balls (m ≈ 44 g, V = 6 mL, d = 20 mm) in ZIF synthesis experiments. To ensure the jars were sealed gas-tight, vacuum-grade silicon grease was used on the seal, and the jar ring clamps were wrenchtightened. To avoid cross-contamination, the milling balls and jars were cleaned by milling a mixture of sodium carbonate and laboratory solid detergent (Sparkleen) with added ethanol for 15 min after every use. Analysis of data was conducted using the PM GrindControl TM software and Microsoft Excel.
Small scale milling reactions were conducted in a 10 mL stainless steel jar with one 7 mm (1.4 g) and one 9 mm (3.5 g) stainless steel ball. The samples were milled at 30 Hz for 5-30 min using a Retsch MM400 ball mill.
Powder X-ray diffraction (PXRD) patterns were collected using a Bruker D2 powder diffractometer equipped with a Cu-Kα (λ=1.54060 Å) source and Lynxeye detector set at a discriminant range of 0.110 V to 0.250 V. The patterns were collected in the range of 3° to 40°. Analysis of PXRD patterns was conducted using Panalytical X'Pert Highscore Plus software. Experimental patterns were compared to simulated patterns calculated from published crystal structures using Mercury crystal structure viewing software. Crystallographic Information Files containing published crystal structures were obtained from the Cambridge Structural Database (CSD) and Crystallography Open Database (COD).
High-quality PXRD data for use in structure determination calculations were recorded on a Bruker D8 instrument using Ge-monochromated CuKα1 radiation. The powder XRD data were recorded in transmission mode (2 range, 2 -70°; step size, 0.017°; total data collection time, 16 h 51 m) with the sample held between two pieces of tape (i.e., foil-type sample holder).

S3
Fourier-transform infrared attenuated total reflection (FTIR-ATR) spectra were collected using a Bruker Vertex 70 FTIR-ATR spectrometer in the range 400 cm-1 to 4000 cm-1. FTIR spectra were analysed using Bruker OPUS software and Microsoft Excel.
Thermogravimetric analysis (TGA) analyses were conducted on a Mettler Toledo TGA/DSC 1 STARe System. All samples were heated at a rate of 5°C/min from 25°C to 800°C under dynamic atmosphere of air with a flow rate of 60 mL/min. The flow rate of the protective gas (N2) was 40 mL/min. TGA curves were analysed using Mettler Toledo TGA analysis software.
Solid-state 13 C CP-MAS NMR spectra were collected on a 400 MHz Varian VNMR equipped with a 7.5 mm CPMAS probe at a spin rate of 5 KHz. All spectra were collected with a contact time of 2 ms and recycle delay of 2 s.

Pressure yield calculation
To calculate reaction yield from the pressure measurements, the difference in pressures, Δp was measured, along with the temperature at the start (Tstart) and end of milling (Tend). The starting pressure was corrected to the final temperature using the Gay-Lussac law: The difference in pressure was then used to calculate the number of moles of CO2 produced in the reaction (n) using the ideal gas law: where Δp is the corrected pressure difference between the start and end of milling at the end-of-milling temperature, V is the empty volume of the reaction vessel, R is the gas constant (8.314 J/Kmol), and T is the temperature at the end of milling. Use of the more exact Van der Waals gas law was attempted instead of the ideal gas law, but the difference in result was negligible for the range of temperatures and pressures used (less than 0.5 % difference in all cases).
The volume of the reactants (Vreact) was estimated from their density and included in the calculation of the empty volume of vessel (V) along with the volume taken up by milling balls. The volume of the vessel alone was measured to be 288 mL, based on the volume of water it can accommodate.
In later ZIF synthesis experiments, the vapor pressures of water [1] (formed in the reaction), and ethanol [2] (added in LAG experiments) were calculated for the given temperatures using the Antoine equation, and then subtracted from the corresponding total pressures before the rest of the calculations were performed. In addition, the amount of CO2 gas dissolved in 3 mL of ethanol was calculated using Henry's law [3] and subtracted from the theoretical number of moles of produced CO2 to get an accurate yield.

Model reaction of molybdenum (vi) oxide and calcium carbonate
To independently validate reaction yields obtained by pressure measurements, we took as a model reaction the solid state reaction of calcium carbonate and molybdenum(VI) oxide which releases CO2 gas according to the following equation: In a typical reaction, 5.9 g of MoO3 (0.04 mol) and 4.1 g of CaCO3 (0.04 mol) were milled in a 250 mL steel jar with 4-7 medium-size balls, using a PM 400 planetary mill at a frequency of 350 rpm for 90-270 min. The number of balls and milling time were varied in order to obtain a range of yields adequate for building a calibration curve. The pressure and temperature in the jars were measured during milling, and the products were analyzed via PXRD and TGA after standing in a desiccator for 1-2 h to remove any adsorbed CO2.
An example of the real-time graphical output for the model reaction Mod-4 is shown in Figure   S1. The temperature of the gas inside the vessel rises during milling, more rapidly at first, then slower as milling goes on. At end of milling the temperature drops rapidly, then continues to slowly fall off.
The pressure rises in a quasi-sigmoidal fashion, possibly indicating an induction period, followed by reaction progress, then as more of the reagents are spent, the reaction slows down. At end of milling, the pressure rapidly falls, following the fall in temperature of the gas inside the vessel. Near identical temperature behavior is seen in all experiments (model reactions, as well as later ZIF syntheses), but the pressure behavior can change drastically depending on the system in question.

S6
Vessel pressure yield was calculated for all model reactions according to section S2.1. The yield was also calculated from TGA analysis by measuring the overall mass loss between room temperature and 650 °C (Step 1 and 2, see example TGA curve for Mod-3 in Figure S5). As both MoO3 and CaMoO4 decompose above 650 °C, any mass loss below that temperature corresponds to CO2 release; either due to thermally induced reaction of leftover reagents, or due to calcium carbonate decomposition, and can be used to calculate the number of moles of CO2 released thermally, and thus the amount of unreacted CaCO3. Table S1 shows the range of reaction conditions and corresponding pressure and TGA yields, while the pressure and temperature profiles, and the model reaction curves are shown in Figure S2. The pressure and TGA yields show very good agreement, with a linear plot, R 2 = 0.9821 ( Figure S3). Figure S4 shows PXRD patterns of samples Mod-1 -Mod-5. Interestingly, the yield cannot be accurately determined from PXRD data, most likely due to amorphization of the reagents.

Large-scale milling reaction with temperature and pressure monitoring
In a typical large-scale reaction, 6.0 g of basic zinc carbonate (0.01 mol, 0.05 mol of Zn) and washing of products, to confirm that the conversion is not due to the washing procedure.
All syntheses resulted in the formation of pure zni-ZnIm2 as the only product ( Figure S6).
There were no signs of remaining reagents in any of the cases, indicating that the conversion based on PXRD is 100%. The products before and after washing are also identical based on PXRD, as shown in Figure S6, meaning that washing with ethanol doesn't change the reaction outcome, purely removes potential excess imidazole.
The NG synthesis reaction vessel pressure showed, similar to the model reactions, a quasisigmoidal curve with an induction period, a quick rise in product formation, and then a tapering off of the reaction progress as the reagents were spent ( Figure S7, solid line). Surprisingly, the reaction vessel pressure appeared to reach its maximum after only 15 minutes! To ensure reproducibility, several measurements were repeated, and are shown to be in good correspondence ( Figure S8). Conversely, the LAG and ILAG syntheses ( Figure S7, dashed and dotted lines, respectively) showed almost no induction period, and a quasi-exponential growth, reaching near-maximum pressure after less than 10 minutes. The maximum pressures (and calculated conversions) of the LAG and ILAG reactions were very similar and higher than that of the NG reaction.
Interestingly, the yields as calculated by TGA ( Figure S9) are higher than the vessel pressure yields, as seen in Table 2. We hypothesize that this is due to absorption of CO2 by the newly formed ZIF. This is in line with the fact that pressure yields of LAG and ILAG preparations are significantly higher than that of the NG synthesis, possibly due to the additives blocking access to the product voids and preventing CO2 absorption. S10

Small-scale milling reaction
In

Large-scale milling reaction with temperature and pressure monitoring
In Additional experiments were conducted without immediate washing of products, to confirm that the conversion is not due to the washing procedure. PXRD and SSNMR data was collected for the products of these syntheses immediately after opening the milling jars.

Neat grinding experiments
The vessel gas pressure curves of neat grinding experiments in a Zn:HMeIm ratio of 1:2 ( Figure   S13) showed a short induction period followed by rapid pressure growth, similar to that in HIm milling experiments. Unlike the previous HIm milling experiments, however, this rapid increase was followed by a steady decline in vessel gas pressure, evening out only after milling stops. The PXRD analysis of the 1:2 ratio large-scale reaction product after 15 min shows the formation of SOD-ZnIm2 product (ZIF8, Figure S12). Closer inspection of the washed 15 min 1:2 ratio synthesis product shows a small peak at ≈11°2 , corresponding to a previously reported complex zinc carbonate methylimidazolate, which is formed by exposing ZIF8 to wet carbon dioxide. The zinc methylimidazolium carbonate byproduct is formed in very small amounts after 15 min, so it is visible in PXRD only after washing away the amorphous background, but CP-MAS SSNMR without washing clearly shows its presence ( Figure S20). The same peak can be seen in the PXRDs of both the washed, and unwashed 1:2 largescale 1h milling product, in a much larger amount, showing that as the reaction proceeds, more of the byproduct is formed ( Figure S12). This formation of byproduct is mirrored by a drop in vessel pressure ( Figure S13), indicating absorption of CO2. We therefore hypothesize that the different behavior of S16 vessel pressure (compared to HIm reactions) is due to absorption of newly formed CO2 by the porous ZIF8 product, and subsequent formation of the unwanted complex carbonate byproduct 1.
It follows that if the absorption of CO2 could be prevented, the formation of carbonate could also be stopped. To that end, an additional equivalent of 2-methylimidazole was added into the milling reaction, in hope that it would help block the pores of newly formed ZIF8. PXRD of the products of this reaction ( Figure S12) showed only the formation of ZIF8, with no complex carbonate byproduct peaks. The vessel pressure shape during milling was much more similar to the original HIm reaction ( Figure S7), with a starting induction period, followed by a steady rise in pressure (albeit much slower than with the HIm reaction). Despite this similarity in shape, the pressure yields obtained were severely underestimated compared to the TGA yields (45% pressure yield vs. 99% TGA yield for the 1h NG reaction in a 1:3 ratio), much more so than in the HIm reactions. We hypothesize that some absorption of CO2 happens even when the pores are partially blocked, but not enough to facilitate formation of the byproduct.

Liquid assisted grinding experiments
The LAG milling experiments' vessel pressure profiles ( Figure S16) follow a similar trend to the NG experiments, in that 1:2 reactions show a drop in pressure after the initial rise, and show peaks of carbonate byproduct 1 in the PXRD pattern of the products after 15, and 30 minutes (both washed and unwashed, Figure S15). The LAG experiments reach an overall higher pressure before the pressure drop ( Figure S17), which could be ascribed to the dual effect of the ZIF8 product being formed faster in the LAG reaction, as well as pore-filling of ZIF8 by solvent molecules, which slows down the byproduct formation. Raising the Zn:HMeIm ratio to 1:3 results in a vessel pressure profile that has no drops in pressure, as well as pure ZIF 8 products, based on PXRD. Expectedly, the maximum reaction vessel pressure in the LAG experiments is reached sooner than in the NG experiments, after 10-15 min.
In these cases as well, the yield is greatly underestimated compared to the TGA yield.  In all cases, FT-IR and PXRD measurements were performed without washing the samples.
As in large-scale milling reactions, the small-scale LAG reactions in a Zn:HMeIm ratio of 1:2 using methanol or ethanol initially show only the formation of the desired ZIF8 product (5, and 10 min millings, Figure S21). However, after 30 min milling, both sets of reactions show the formation of the zinc carbonate methylimidazolate byproduct, with the ethanol-assisted reaction seemingly resulting in the formation of a smaller amount of byproduct. This is in line with the hypothesis that blocking the pores of the newly formed ZIF8 would also block the byproduct formation; ethanol molecules have a larger volume and diameter, so they could block the pores more effectively. The IR experiments conducted on experiments with added ethanol show that it is indeed present in the reaction mixture ( Figure S24), though it is impossible to say if it is adsorbed on the surface, or absorbed inside the pores.
Furthermore, as the ZIF8 framework is hydrophobic, it is likely that larger, less polar liquid molecules are better guests for the framework, and hence more efficient in blocking the framework pores.
To further test the limits of the hypothesis, experiments with water, a very small and very polar potential guest, and isopropanol, a much larger (and much less polar) potential pore-blocker, were conducted ( Figure S22). Based on the PXRD results, isopropanol indeed prevented the byproduct formation much better than water (only a very small amount of 1 was formed, even after 30 min milling), further reinforcing the hypothesis.
Another important factor in byproduct formation is the speed of the reaction. It appears that the byproduct is not formed simultaneously with the desired product, ZIF8. Instead, ZIF8 is formed first, and then transforms to the byproduct over time. It follows that slower reactions will produce less of the byproduct in a giver timeframe, while faster reactions will produce more of it. Indeed, the small-scale NG reaction doesn't seem to produce any 1 after 30 min, LAG reactions produce mostly ZIF8, with some amount of 1 dependent on the milling liquid, while the fast ING and ILAG reactions provide full conversion into the byproduct 1 with no signs of ZIF8 in only 30 min ( Figure S22).

S24
Finally, performing the LAG synthesis with basic zinc carbonate and 2-methylimidazole in a 1:3 ratio provided only ZIF8 after 30 min milling, with no carbonate byproduct formation, no matter if methanol, ethanol, or isopropanol were used as LAG liquids ( Figure S23).   Finally, low temperature N2 adsorption isotherms ( Figure S27) were used to elucidate the porosity of the ZIF-8 material prepared and purified without using any solvent. A BET surface area of 1785 m 2 /g was obtained for the material after sublimation; comparable to the 1758 m 2 /g surface area recorded for the commercial Porolite Z8 kindly supplied by MOF Technologies Ltd.

Structure determination of compound 1, Zn2(MeIm)2CO3, directly from PXRD data
A sample of 1, Zn2(MeIm)2CO3), was prepared according to a literature procedure 4 , and a highquality laboratory PXRD pattern suitable for structure determination was recorded as described in Section 1.2. In addition to the major contribution due to 1, the PXRD data also contained low-intensity peaks due to an impurity amount of ZIF-8.
The peaks due to 1 in the PXRD data were indexed using the ITO code in the program   S31 carbon and oxygen atoms with half occupancy) was located on a 2-fold rotation axis parallel to the caxis, and the fragment was allowed to translate along this axis and to rotate around this axis.
In total, 16 independent GA structure solution calculations were carried out. Each calculation involved the evolution of 100 generations for a population of 100 structures, with 10 mating operations and 50 mutation operations per generation. The same structure solution of highest quality (corresponding to the lowest value of Rwp) was obtained in all 16 cases.
The structure solution was then used as the starting model for Rietveld refinement, carried out using the GSAS program, with the ZIF-8 impurity included as a second phase in the refinement. In the Rietveld refinement, standard restraints were applied to bond lengths and bond angles, and planar restraints were applied to the methylimidazolate group. The final Rietveld refinement gave a good quality of fit to the PXRD data (Rwp = 4.97%, Rp = 3.37%; Figure S28

Cobalt(II) carbonate and imidazole experiments
In a typical large-scale reaction, 7.0 g of cobalt(II) carbonate (0.059 mol) and 8.0 g of imidazole (0.118 mol) were milled in a 250 mL steel jar with 7 large balls, using a PM 400 planetary mill at a frequency of 300 rpm for 90 min. For liquid assisted grinding (LAG) experiments 2 mL of methanol were added to the reaction mixture. The pressure and temperature in the jars were measured during milling ( Fig S28). The samples were analyzed via PXRD (Fig S29) Both the NG and LAG syntheses appeared to be much slower than the analogous reactions using basic zinc carbonate, and were not equilibrated even after 90 min (Fig S29). The NG reaction showed a maximum in pressure around 85 min, followed by a pressure decrease, similar to the reactions involving basic zinc carbonate and 2-methylimidazole. The LAG reaction showed a continuous increase in pressure and achieved a much higher final pressure than the NG reaction The NG synthesis resulted in the formation of a predominantly amorphous product ( Figure S30) with a pressure yield of 42.4%. The LAG synthesis resulted in the formation of a mixture of products