id
stringlengths 24
24
| idx
int64 0
402
| paragraph
stringlengths 106
17.2k
|
|---|---|---|
657552c45bc9fcb5c97b17c9 | 2 | The Dewar isomer 1 was prepared as previously reported by photoisomerization of B-mesityl-N-(t-butyldimethylsilyl)azaborinine (A) as described in Figure . To preclude any thermal reversion to the azaborinine, the monomer was stored in a freezer at -28 ºC prior to use. We explored different generations of Grubbs and Hoveyda-Grubbs catalysts for the ROMP of monomer 1 (Figure ). The monomer conversion was determined by 1 H NMR integration and the results are summarized in Table . Grubbs 3 rd generation catalyst (G3) is one of the most widely used ruthenium catalysts for ROMP. In comparison to 2 nd generation catalysts (G2, HG2), G3 exhibits very fast initiation rates, which typically enables the formation of polymers with very narrow dispersities and excellent control over their molecular weights. However, G3 was found to be not very effective at converting Dewar isomer 1 to polymer 2. At 0.1 M monomer concentration in benzene or dichloromethane, G3 gave only low monomer conversion even after very long reaction times at either room temperature or 0 ºC (Table , entry 1, 2), and conversions were even lower when using THF as the solvent. This contrasts the successful controlled polymerization of other cyclobutene derivatives reported in the literature. A possible reason could be that the pyridine base that is liberated during the initiation step interferes by binding to the centers. However, no change in chemical shift was observed in the 1 H or B NMR spectra when monomer 1 was treated with an equimolar amount of pyridine. Another possibility is that the initiation step is very fast, as expected, but the stability of the initial ring-opening product and the propagating species in its resting state is insufficient. An indication that this is the case comes from an 1 H NMR experiment in which the monomer was mixed with G3 in a 10:1 molar ratio in C6D6. Upon mixing the color of the solution immediately changed from yellow-green to orange-brown. An NMR spectrum taken within <10 mins showed that the carbene proton of G3 disappeared completely and multiple new peaks arose in the region from -16 to -20 ppm (Figure ). These peaks gradually disappeared with longer reaction times. This is indicative of very rapid but unselective initiation and insufficient selectivity/stability of the propagating species. G2 and HG2, which do not contain unhindered basic pyridine ligands, are known to exhibit improved thermal stability, as well as oxygen-and moisture-tolerance. Although G2 and HG2 are usually not suitable for living polymerization due to the slow initiation and small initiation/propagation rate ratio (ki/kp), they have been successfully applied to the polymerization of cyclobutenes when other catalysts gave poor results. An 1 H NMR analysis after 10 mins of mixing 1 and G2/HG2 in a 10:1 molar ratio in C6D6 again revealed complete disappearance of the carbene proton of the catalyst with formation of several new peaks. The new carbene peaks were present almost exclusively in the region of 17.5 to 18.5 ppm, indicating more selective formation of isomeric propagating species with G2 and HG2 (Figure ). Indeed, performing the ROMP of 1 at room temperature with either G2 or HG2 in a 100:1 molar ratio proved to be more effective, resulting in higher monomer conversion (>90%) over a shorter time period of 7-9 hours (Table , entry 3-4). As seen in Table , entries 5-6, higher monomer concentration (0.3 M) promoted propagation, further shortening the reaction time to 4.5 h. HG2 consistently gave slightly higher conversions, indicating a faster rate of polymerization. The use of HGM2001, which is a Z-selective catalyst, yielded polymers with an increased number of cis-linkages than G2 and HG2 (vide infra). Remarkably, an 1 H NMR analysis after 10 mins of mixing 1 and HGM2001 in a 10:1 molar ratio showed the formation of only two broad overlapping peaks at ca. 15.5 ppm, suggesting far superior regio-and stereoselectivity in the initiation and propagation steps (Figure ). However, only 31% conversion was achieved over 24 hours at room temperature when using a 100:1 molar ratio of 1 to HGM2001 (Table , entry 7). An increase of the reaction temperature to 50 °C did not significantly increase the monomer conversion but led to competing cycloreversion to the azaborinine A (not observed at RT, see Table ). |
657552c45bc9fcb5c97b17c9 | 3 | The polymerizations were quenched with a large excess of vinylene carbonate (VC), the volatiles removed in vacuo, and the polymers isolated by repeated precipitation from benzene into MeCN. Polymers 2 and 2-Z showed good thermal stability as established by thermogravimetric analysis (TGA), revealing an onset of decomposition of ca. 200 °C (Figure ). DSC analysis of 2 did not reveal a clear glass transition (Tg) within the range of -20 to +180 °C (Figure ), most likely because the relatively rigid structure of the polymer results in a Tg above the accessible temperature range. The molecular weight distributions were analyzed by gel permeation chromatography (GPC) in tetrahydrofuran (THF) with a refractive index detector (GPC-RI) relative to narrow polystyrene (PS) standards. GPC-RI analysis for the polymer obtained with G2 (1 mol%) at 0.1 M monomer concentration (entry 3) gave a monomodal molecular weight (MW) distribution with a number average molecular weight of Mn = 8100 Da and a dispersity (Đ) of 1.76 (Figure ). Similar results were obtained for HG2 (entry 4) and HGM2001 (entry 7) (Figure ). The theoretically predicted molecular weights are significantly higher (31100 at 100% conversion), but the GPC-RI data are likely an underestimation because of the use of structurally different narrow PS standards. Indeed, molecular weights derived from GPC-LS detection were significantly higher (Figure ). Attempts to further verify the chain lengths and distributions by MALDI-TOF MS with various matrices and ionizing agents proved unsuccessful. To further study the controlled nature of the ROMP of 1 with G2 and HG2 as catalysts we carried out kinetic experiments at 0.3 M monomer concentration (Table , entries 5, 6 and Figure ). At room temperature in benzene the monomer conversion reached 80% within 4.5 hours for G2 and 93% within 1 hour for HG2. The conversion of Dewar isomer 1 with G2 followed first-order kinetics as illustrated by a linear plot of ln([M0]/[M]) vs time that gave a calculated kobsd,G2 = 0.38 M -1 s -1 . Meanwhile, HG2 showed firstorder kinetics but very fast conversion over the first 5 minutes, followed by faster first-order kinetics than for G2 with kobsd,HG2 = 1.54 M -1 s -1 (Figure ). This suggests that HG2 is more effective in promoting the polymerization than G2. However, G2 showed better first-order kinetic behavior, thus we chose G2 as catalyst in the subsequent studies. |
657552c45bc9fcb5c97b17c9 | 4 | The chemical structure of the new polymers was confirmed by 1 H, 11 B, and two-dimensional (2D) NMR spectroscopy. The disappearance of the characteristic olefinic group signals and pronounced peak broadening in the 1 H NMR spectra clearly indicated successful ROMP of the strained cyclobutene rings in Dewar isomer 1 (Figure ). In the 11 B NMR spectra of both 2 and 2-Z, a significant upfield shift from ca. 53 to 45 ppm and concomitant peak broadening provided further evidence for the successful polymerization (Figure ). An upfield shift is frequently observed upon polymerization of borane monomers as a result of shielding effects of the neighboring groups along the polymer chain. To further confirm the connectivity between the four-membered BN-heterocycles and vinylene groups in the polymer main chain, heteronuclear single-quantum correlation (HSQC), heteronuclear multiple-quantum correlation (HMBC), and nuclear Overhauser effect spectroscopy (NOESY) NMR data were acquired. The HSQC and HMBC data of 2 and 2-Z showed the expected cross peaks for the mesityl and tert-butyldimethylsilyl groups (Figures S10-11, S15-16). The backbone protons of polymers 2 and 2-Z in the range from 2.7-6.5 ppm could be assigned based on a NOESY NMR analysis (Figure ) and supported by DFT methods. To gain insights into the relative distances between protons we optimized the geometry of head-totail model dimers with vinyl end groups, having either a cis-and trans-vinylene linker. The methine protons H1 and H4 were assumed to retain the syn configuration that is seen in the Dewar-azaborinine precursor 1 and only one of the possible diastereomers is illustrated (relative stereochemistry of C atoms in the B-N heterocycles). The calculated Gibbs free energy was higher for the isomer with a cis-olefin linkage by 20.8 kJ mol -1 relative to that with a trans-olefin linkage, which suggests that formation of trans-linkages is thermodynamically more favorable (Table ). The structure of the trans-isomer is displayed in Figure , illustrating some of the closest intramolecular H … H distances. The distances in this model dimer were then used to assign the NOESY spectrum of the polymer (Figure and Table ). The allylic methine protons H1 and H4 are expected more upfield than H2 and H3. The upfield signal at 3.4 ppm was assigned to H1 as it shows the expected intense NOE peak at 3.4/2.6 ppm due to its proximity to the boron-bound mesityl group (H9). The orthomesityl protons (H9) are also in close proximity to H2 resulting in another dominant cross peak 5.8/2.6 ppm. A third strong cross-peak at 3.4/5.2 ppm is attributed to the NOE between H1 and H4 which are in adjacent positions and share the same orientation. Finally, the assignment of H3 at 6.0 ppm is based on an NOE peak with H2 at δ = 6.0/5.8 ppm. The separation of H2 and H3 within a single trans-vinylene unit is large, but H2 comes in close contact to H3' in the next vinylene repeating unit and vice versa. Additional weaker NOE peaks indicated the presence of a relatively smaller number of cis-vinylene linkages, or possibly some head-to-head defects (Figure ). For polymer 2-Z which was prepared with HGM2001 as a Z-selective catalyst, the 1 H NMR signals were much sharper than those of 2 and appeared at different chemical shifts (Figure and Table ). The presence of a dominant set of sharp signals in the region from 2.5-5.5 ppm allowed for unequivocal assignment of all the protons within the B-N heterocycles and vinylene linkers. The NOE peaks were in excellent agreement with the proposed head-to-tail structure with cis-linkages. Notably, the mesityl protons were split into two sets suggesting that rotation about the B-C bond is hindered. Collectively, these data suggest the presence of predominantly but not exclusively trans-linkages for polymer 2 (the exact E/Z ratio is difficult to determine because the 1 H NMR signals are heavily broadened) and a highly regioregular structure with dominant formation of cislinkages for polymer 2-Z (>90% Z). |
657552c45bc9fcb5c97b17c9 | 5 | The structural integrity of the four-membered BNheterocycles was further verified by comparison of the FT-IR spectra of the monomer and the polymers (Figure , for full spectra see Figure ). As expected, poly(1,2-azaborinine)s 2 and 2-Z displayed very similar spectral features. The individual IR bands were assigned based on comparisons with results from theoretical calculations (B3LYP/6-31g(d)) on monomer 1 and the trans-and cis-vinylenelinked head-to-tail model dimers (Table ), as well as previously reported experimental data for 1. Strong B-N stretching modes were observed for both the monomer and polymers (ca. 1354 / 1360-1361 cm -1 ). The characteristic C-H bending modes in the BN heterocycle of 1 (ca. 1179, 1136 and 979 cm -1 ), as well as the C-B (ca. 1038 cm -1 ) and C-N stretching modes (ca. 1252 cm -1 ) are also seen in the poly(1,2-azaborinine)s 2 (1250, 1041 cm -1 ) and 2-Z (1250, 1034 cm -1 ) with some peak broadening, further confirming the integrity of the BN-heterocycle. The C-B stretch for 2-Z is shifted slightly to lower wavenumbers as also predicted computationally. For monomer 1, an additional set of strong bands was found at 1275, 1228, 1158, 1120-1106 and 941-883 cm -1 and assigned to C-H bending modes in the cyclobutene ring. The disappearance of these bands in poly(1,2-azaborinines) 2 and 2-Z confirms the ring opening of the cyclobutene ring during polymerization. |
657552c45bc9fcb5c97b17c9 | 6 | While the NMR features for poly(1,2-azaborinine)s 2 and 2-Z are distinct, very similar FT-IR features were seen for the isomeric polymers as illustrated above. To further verify the preferential incorporation of cis-linkages for polymer 2-Z we also acquired Raman data for both polymers. Because of the different selection rules, Raman is uniquely able to distinguish between E-and Z-olefin incorporation into the backbone of polymers such as poly(butadiene) rubbers of polynorbornene. Raman spectra were acquired by excitation with a 785 nm laser. The region for the olefinic signals is illustrated in Figure . The bands for the C=C stretching modes are expected around 1640-1680 cm -1 and those for the olefinic C-H bending in the region of 1250-1320 cm -1 . While differences in the intensity of the bands in these regions are apparent for 2 and 2-Z, signal overlap prevented a quantitative analysis. Nevertheless, collectively, the NMR, IR and Raman data strongly support the unique structure with four-membered B-N heterocycles embedded in the backbone of a ROMP-derived polyolefin. In conclusion, we have successfully synthesized the first example of a Dewar-azaborine polymer (2) by ROMP of Dewar isomer 1 with Grubbs and Hoveya-Grubbs 2 nd generation catalysts, as well as a Z-selective catalyst. While several azaborinine-derived polymers have recently been reported as analogs of common carbon-based polymers such as polyphenylene, polystyrene, and polyvinyl naphthalene, 12b, 19a, 23 the structure of polymers 2 and 2-Z containing 4-membered B-N heterocycles is unprecedented. Successful formation of 2 was verified by GPC, 2D NMR, and FT-IR, and the differences in regiochemistry for the Zselective catalyst were further evaluated by 2D NOESY and Raman measurements. Recent studies by Klausen, Ouchi and others demonstrate the great promise of polymers with B substituents and B-N units in the pursuit of novel polyolefins with polar functional groups through post-modification methods. 12c, 23h, In a similar vein, we anticipate that the presence of the B-N four-membered rings may be exploited in the preparation of new functional polymers that contain both amine and hydroxyl side groups via chemoselective organoborane oxidation 8 and/or hydrogenation of the double bonds in the backbone. |
664ca8ab91aefa6ce1974d55 | 0 | We report herein our studies on the direct photoactivation of carbonyl cyclopropanes to give biradical intermediates, leading to selective cleavage of the more substituted carbon-carbon bond. Depending on the substrate structure, extended alkenes were isolated or directly reacted in a photo-Nazarov process to give bicyclic products. Based on these results, a unified reductive ring-opening reaction was developed by using diphenyl disulfide as a HAT reagent. By performing a sequential cyclopropanation/selective ring opening reaction, we achieved a CH2 insertion into the ,- bond of both acyclic and cyclic unsaturated carbonyl compounds. Our protocol therefore provides a further tool for framework-editing of carbocycles, complementing the recent progress in "skeletal editing" strategies. |
664ca8ab91aefa6ce1974d55 | 1 | Photochemistry has enabled numerous unique transformations using light as a green energy source. Carbonyl compounds have been known since a long time as one of the most important photoactive species in both synthetic and biological applications. They can be used as either photoactive substrates or as photocatalysts as exemplified by thioxanthone or benzophenone. The excitation of carbonyl compounds can be achieved via either energy transfer catalysis or direct excitation (Scheme 1A). Although energy transfer catalysis provides milder conditions to activate molecules with a low-energy light source, extensive optimization of the photocatalyst is usually required. Furthermore, electron transfer pathways are usually a competitive mechanism, leading to the formation of by-products. Alternatively, the direct excitation of carbonyl compounds enables several well-known transformations, such as the Paternò-Büchi 9 or Norrish-type reactions. 10 However, most studies focused on alkene-conjugated carbonyl compounds since the extended conjugated system makes them highly photoreactive. Excitation of the system of either the C=O or the C=C bond can occur, leading to Paternò-Büchi reactions 9 or [2+n] annulations. These transformations have been studied intensively since the start of photochemistry and have seen renewed interest in the past few decades. In contrast, the homolytic fission of C-C bonds in carbonyl compounds is underdeveloped due to two main reasons: i. The excitation of -carbonyl compounds requires higher energy than -conjugated systems, ii. The bond dissociation energy of bonds is much larger than the one of bonds, making the process thermodynamically challenging. |
664ca8ab91aefa6ce1974d55 | 2 | The heteroatom is stabilizing radical intermediates generated from a hydrogen atom transfer (HAT) process. As a result, the Norrish type II C-C bond cleavage, which involves a HAT process, became more favorable than the Norrish type I direct fragmentation. After C-C cleavage, an aldol-type reaction led then to formation of the ring-contracted product I. In addition, the competitive Norrish-Yang pathway to give strained ring II is also less favorable than in the case of acyl-substituted carbocycles. To achieve selective C-C bond cleavage on carbocycles, we envisioned the use of strained rings to lower the bond dissociation energy of the C-C bond (Scheme 1B). In the case of cyclic substrates, a ring expansion to give products III or IV could be achieved from the direct excitation of carbonyl compounds, complementing the ring-contraction strategy of Suarez and Sarpong. When considering that bicyclic structures containing carbonyl cyclopropanes are usually synthesized from the corresponding olefins, a sequential cyclopropanation/selective ring-opening reaction would provide a direct strategy for CH2 homologation of , unsaturated carbonyl compounds (Scheme 1C). The insertion would happen on the , bond of the carbonyl group, in contrast to most existing homologation reactions proceeding via insertion into the carbonyl -carbon bond. The key for success of this approach would rely on selective cleavage of the endocyclic C-C bond. Interestingly, most methods for reductive cyclopropane ring-opening are based on transition metals and favor cleavage of the exocyclic C-C bond. We envisioned however that selective endocyclic C-C bond cleavage could be achieved through formation of the most stable 1,3 biradical intermediate. In this work, we report first our mechanistic studies on the direct photoexcitation of cyclopropyl carbonyl compounds by experiments and quantum chemical computation. Our investigations resulted in the discovery of efficient photoisomerization processes in case of polycyclic substrates, as well as a cascade photoisomerization-photo Nazarov reaction for bicyclic compounds (Scheme 1D). By adding diphenyl disulfide as a hydrogen atom transfer (HAT) reagent, we were then able to achieve a unified ring opening reaction applicable to both linear and cyclic biradicals, paving the way for the development of a novel homologation methodology for , unsaturated ketones In our previous work, we reported preliminary results on the [3+2] annulation of cyclopropanes 1a and 1b with phenylacetylene to give products 2a and 2b (57% and 40% yield, Scheme 2A) in the absence of a photocatalyst. The reaction was successful for geminally disubstituted carbonyl cyclopropanes under irradiation at 390 nm with a UV Kessil lamp. A lower yield was obtained at 352 nm and no conversion was observed at 440 nm. In addition, Brown and coworkers reported the formation of biradicals from bicyclo[1,1,0] butane (BCB) and housane derivatives, which are more strained ring systems. We wondered if it would be possible to extend the annulation reaction to larger bicyclic ring systems such as bicyclo[3,1,0]hexane 3a or norbornene derivative 5a (Scheme 2B). Under our conditions, we however observed rearrangements leading to products 4a and 6a instead of the expected [3+2] annulations to give 4a1 and 6a1. While the mechanism of the [3+2] annulation has been investigated by our group 8 and others, the behavior of 1,3 biradicals in transformations beyond annulation has not been described thoroughly in the literature. 15e Therefore, we decided to initiate more systematic studies in the absence of alkynes as biradical trap (Scheme 2C). We first performed the irradiation of trans-cyclopropane 1c. After 48 hours of irradiation, epimerization was observed and 1c was recovered in 93% yield as a 4:1 mixture of trans and cis isomers, suggesting that homolysis of the C-C bond was occurring and was reversible, as had been observed previously. Performing the same control experiment with 1a resulted in a mixture of isomerization products 2a1 and 2a2 with 40% of 1a recovered. These products might arise from a Norrish type II reaction. The different reaction outcome between 1a and 1c suggested that there may be a competition between Norrish type II reaction and reversible C-C bond cleavage depending on the substrate. We then conducted the same experiments on more rigid polycyclic structures 3a and 5a. To our delight, products 4a and 6a were obtained in good yields (86% and 67% respectively). A trace amount of photoisomerization product 4a2 was also observed. This could indicate that both reactions proceeded via formation of the ring-opened alkene products, but in the case of 4a2 a fast photo-Nazarov reaction led to 4a. HAT process. However, this pathway is unlikely as the computed activation energy was 35.3 Kcal/mol. We hypothesize that 1a can undergo Norrish type I fragmentation, leading to generation of the allyl radical. It was reported that the radical-radical coupling can happen at a least steric hindered center between allyl radical and acyl radical, leading to olefin 7a2 (see SI for detail mechanism). We then performed similar computations on bicyclic cyclopropane 3a. In this case, the Norrish type II pathway A is energetically unfavorable for both 1,4 and 1,5 HAT with energy barriers of more than 20 Kcal/mol, while only 6.1 Kcal/mol is required for C-C bond cleavage. The reason might be due to the rigidity of structure 3a, making the HAT process geometrically less favored. Intermediate 4a1 could be obtained via 1,4 HAT from IB1-3aT to IB2-3aT then ISC and relaxation the singlet ground state IB2-3as. However, transition state TSB2-3aT was around 33 Kcal/mol higher in energy than IB1-3aT, which is challenging to reach under our conditions. When looking at the singlet state energy surface of intermediate IB1-3aT, we realized that after ISC, IB1-3aT would spontaneously transform to 4a2 via a 1,2 hydrogen shifts. In fact, it was demonstrated that the generation of 1,3 biradical often result in 1,2 hydrogen shifts in competition with radical-radical coupling to give cyclopropanes. 18b-f As described in literature, the photo-Nazarov reaction only happens if sufficient twisting of the double bond is possible (see SI). 17 This process is well established with cyclohexenyl phenyl ketones such as 4a2, furnishing product 4a under UV irradiation. In contrast, more rigid structures such as present in 5a disfavor double bond twisting, hence the intermediate olefin 6a can be isolated. We then explored synthetic applications of the photo-rearrangement of polycyclic cyclopropanes 3a-k and 5a-c (Scheme 4A). Starting from bicyclohexanes 3a-d, hydrofluorenones 4a-d bearing alkyl and fluoro substituents were formed in 71 -86% yield via the photo-isomerization-Nazarov cascade. Tetrahydropyran and thiophene derivatives 4e and 4f were also successfully obtained. Bicyclo[3,2,1]octene products 6a-c were obtained in 54 -67% yield via the photo-induced isomerization starting from norbornene derived cyclopropanes 5a-c. |
664ca8ab91aefa6ce1974d55 | 3 | During the investigation of the scope, we often observed a trace amount of reductive ring opening product 7. We speculated that 7 may be formed via a hydrogen atom transfer on the speculative biradical intermediate. As described in the quantum chemical computations (Scheme 3), a 1,3 biradical is always the most stable first intermediate generated from both bicyclic and linear cyclopropanes. We therefore speculated that a suitable HAT transfer reagent would allow to intercept the biradical intermediate in all cases, leading to a general homologation protocol. After optimization with several types of HAT reagents (See SI for detail), we were please to obtain reductive ring opening of product 7a starting from bicyclo [3,1,0] hexane derivative in 50% yield (Scheme 4B). The protocol only required the addition of 2 equivalents of diphenyl disulfide as a HAT reagent. Compared with previous studies on light-mediated ring-opening reduction of cyclopropanes, strongly reductive conditions were required, and the scope was limited to spiro cyclopropyl oxindoles 20a or aryl substituted cyclopropanes, 20b making these approaches not suitable for the development of a general homologation protocol. Several differently substituted-aryl cyclohexyl ketones 7b-h were obtained in yield ranging from 30 to 68%. The reaction is also possible for the formation of medium sized ring 7i in moderate yield. Starting from cyclopropanes 5a and 5c derived from the norbornene skeleton, products 7j and 7k were obtained in 54% and 45% yield respectively. For linear di-substituted carbonyl cyclopropanes 1a and 1b, only 1.2 equivalents of diphenyl disulfide are sufficient to furnish products 7l and 7m in good yields (69 and 89%). In contrast, 1.7 equivalents of diphenyl disulfide were required to reduce mono substituted or non-substituted carbonyl cyclopropanes, giving products 7n-q in 50 to 75% isolated yield. It is worth mentioning that a non-substituted cyclopropane can be used in this protocol, while no conversion was observed with the same substrate in the [3+2] annulation with alkynes. Starting from spirocyclic cyclopropane 1i, 60% of product 7r was obtained and the cyclobutane ring remained untouched, demonstrating the chemoselectivity of the reaction. Carbonyl cyclobutanes delivered products 7s and 7t. This strategy can therefore potentially be used as an alternative of the DeMayo reaction for the insertion of two carbon atoms. |
664ca8ab91aefa6ce1974d55 | 4 | Taking advantage of the simplicity of the reaction protocol, we performed a sequence of cyclopropanation/ring-expansion directly from norbornene derivative 8a (Scheme 4C). Without the need to purify the cyclopropane intermediate, we were delighted to observe the formation of product 6a in 55% and 7j in 45% overall yield. Overall, through only small changes in reaction conditions, we could access both saturated and unsaturated products resulting from a one carbon insertion. The same reductive telescoped process was also successful for both linear and bicyclic cyclopropanes, resulting in ring opening products 7p and 7e in 48% and 41% overall yield respectively (Scheme 4D). In some cases, we observed an over reduction of the carbonyl group to give the corresponding alcohol. Considering that the alcohol could be a suitable precursor for the removal of the acyl group via a further C-C bond cleavage step, we performed the reaction from cyclopropane 3a with 5 equivalents of phenyl disulfide (Scheme 4E). In this case, 51% of alcohol 9 was isolated. The conversion of 9 into ether 10 in 80% yield has been reported, demonstrating the possibility for acyl group removal, which can therefore be considered as a transient activating group for the cyclopropane. Based on our reported preliminary results on the [3+2] annulation under photocatalyst free conditions (Scheme 2A), we then further explored the scope of disubstituted cyclopropane substrates (Scheme 5). This protocol is especially attractive due to its simplicity, as no catalyst, Lewis or Brønsted acid or other additive is required, in contrast to other reported methods. The best yields were obtained with 5 equivalents of alkynes as trapping reagents under irradiation for 48 hours (see SI for optimization). |
664ca8ab91aefa6ce1974d55 | 5 | Cyclopropanes having vicinal di-substituents such as dimethyl, difluoro or diester gave the best results (products 2ad with 50-78% yield). Both naphthyl and mono substituted cyclopropanes can be used in this transformation, albeit moderate yields were obtained (2d -40% and 2e -55%). Spiro [4.5]decene 2f (54%) and spiro [4.3]octene 2g (44%) could be synthesized from the corresponding spirocyclopropanes. Substrates bearing methoxy or fluorine groups on the benzene ring of the carbonyl group gave similar yields (2h -66% and 2i -64%). We then studied the scope of alkyne partners. In general, electron rich aromatic alkynes gave better results than the electron poor counterparts (see SI for scope limitation). Apart from 3-fluoro and 3-chloro phenyl acetylene, there was no product observed from other electron poor aromatic alkynes. In contrast, [3+2] products were obtained for a wide range of electron donating groups on the benzene ring, such as alkyl, amide, methoxy, phenyl and alkylalcohol (2j -2r, 30 -82% yield). Unfortunately, aliphatic alkynes were not suitable for the reaction, which constituted a limitation when compared to our previous study on energy transfer catalysis. Alkynes bearing extended aromatic systems or heterocycles delivered products 2s-u in moderate yields (35 -45%). During completion of this manuscript, Zhang and co-workers reported the same [3+2] annulation with ethanol as a solvent. Nevertheless, our study presents a broader range of alkyne coupling partners. In fact, there are only two identical substrates present in both works. Therefore, we believe that adding our own results on this transformation will be still useful for the synthetic community. |
664ca8ab91aefa6ce1974d55 | 6 | In conclusion, we have presented in this work a detailed study of the reactivity of biradicals generated from the direct photoexcitation of carbonyl cyclopropanes. Depending on the substrate structure, intramolecular photoisomerization processes were favored or [3+2] annulations could be developed. Quantum chemical computations confirmed that 1,3 biradicals are viable key intermediates. By taking advantage of this common intermediate, a unified reductive strategy for the ring-opening of carbonyl cyclopropanes was achieved using diphenyl disulfide as a HAT reagent. These transformations pave the way for the development of new homologation strategies resulting in formal CH2 insertion onto the , c-C bond of unsaturated carbonyl compounds, extending the current toolbox 24 for the "skeletal editing" of organic compounds. |
67483870f9980725cf4fb11f | 0 | The driving key factor in engineering and thus in chemical engineering lies in the adaption of conventional knowledge to unconventional environments. Ideas can cook and ripe in this melting pot of ideas. A great unconventional area is applications beyond earth where technical ideas can emerge. We would be delighted if you could look at the article from this perspective and then confirm that polymer reactors are an answer that has so far gone unnoticed to the question: What construction material can be used in catalysis when metals are not suitable? Polytetrafluoroethylene (PTFE) and related compounds are thermoplastics and therefore in general suitable for processing via additive manufacturing for applications in space . Additive manufacturing is supported by institutions like ASTM international, is already prevalent in aerospace industry, and it has, while in its infancy, a massive potential in space applications . Here, several aspects of material science and material engineering have to be considered, being structure, properties, processing and performance, which need to be adapted and tuned for space missions . |
67483870f9980725cf4fb11f | 1 | Weight is usually assumed to be the governing factor when calculating the cost of space missions, although the launch cost per ton has decreased by a factor of forty in recent years ; for long or permanent missions weight is still prohibitive and in-situ resource utilization (ISRU) preferable. A reasonable and planned step is ISRU in the case of propellant production on Mars from CO2 of the Martian atmosphere and H2 from water electrolysis of water ice from the Martian surface . However, some high-performance materials cannot be produced on site. In these cases, the shipped-in resources should be recyclable in order to cease the costly transport as efficient as possible , which makes the choice of stable and unreactive Fluoropolymers reasonable. |
67483870f9980725cf4fb11f | 2 | Reaction engineering is crucial in designing highly integrated and efficient processes for optimal utilization of weight, resources and time. The unique challenges of extra-terrestrial sites like Mars or permanent space stations equally pose unique opportunities to reevaluate reaction engineering of already well understood reactions. For example, the production of Methane as propellant for the back transit is a mature process on earth , but conventional methanation reactors maximize yield with high temperatures and pressures, which require heavy periphery. In this context, the use of recyclable Fluoropolymers as an alternative reactor material, though never considered on earth, becomes advantageous for Martian exploration due to their reusability and lower density compared to typical steel reactors. Moreover, the concept of ISRU requires solid ground to start from. All technology contributing to a safe execution of space operations should be robust and somehow tolerant for the conditions at the landing site . Thus, the implementation of polymeric reactors for applications beyond earth comes with the challenge of suitable and robust catalysts. |
67483870f9980725cf4fb11f | 3 | We have recently demonstrated the viability of a polymeric reactor for use in the low-temperature conversion of CO2 to Methane, as well as identified Ru/TiO2 as the most active catalyst at mild process conditions . In this study, we expand our previous work by using a Ru/TiO2 catalyst in both a PTFE and PFA based reactor and measure the activity and temperature throughout a wide range of process conditions. After the reaction the reactors are tested with light microscopy and Raman confocal microscopy to evaluate material robustness and draw conclusions for the particular case of applications beyond earth and more general applications on earth. |
67483870f9980725cf4fb11f | 4 | TiO2 is used as catalyst support. The spherical support (Accu ® Sphere, Saint-Gobain) is used as received from the manufacturer and their tests show a narrow size distribution with diameters of D10 = 0.52 mm and D90=0.58 mm. The catalyst is prepared by impregnation with the excess solvent method. Ruthenium(III) chloride hydrate (38.0-42.0 % Ru basis, Sigma Aldrich), is dissolved in water and placed in a thermostated oil bath and stirred. The TiO2 support is poured into the solution, the mixture is heated to 50 °C, and left until the solvent has evaporated. Afterwards the particles are transferred into an oven and dried at 105 °C to remove residual water. The dry particles are subsequently transferred to a muffle furnace and calcined in static air at up to 200 °C. After the particles are cooled down, they are washed three times with dilute ammonia solution (3 wt.-%) to remove remaining chlorides and afterwards three times with deionized water to remove the ammonia. The particles are again dried in an oven at 105 °C over night and stored in a desiccator until use. |
67483870f9980725cf4fb11f | 5 | The PTFE and PFA tubes with an outer diameter of 1/4 inch are both cut to a length of 30 cm and the ends reinforced with tubing inserts (Swagelok) to allow gas tight ferrule fittings. The upstream end of the reactor is capped with a tee-piece with an 1/16 inch sealed tube containing a moveable K-type thermocouple, passing through the whole length of the tube and the gas-inlet. The reactor tube is heated with an aluminum jacked with four heating cartridges placed in it, one of which is equipped with an internal K-type thermocouple to control the temperature. A second thermocouple is placed in a 1 mm kerf in the heating jacked as to be in contact with the outer tube wall. The third K-type thermocouple is placed in the 1/16 inch sealed tube at the height of the catalyst bed to be able to measure the inner reactor temperature. A sketch of the setup is shown in Figure ). From bottom (downstream) to top the tube is first filled with 400 mg of quartz beads in the size fraction of 0.5-0.75 mm framed with silica wool to stabilize the catalyst bed and distribute the gas flow. |
67483870f9980725cf4fb11f | 6 | Secondly, 200 mg of catalyst are diluted with 200 mg of unimpregnated support to prevent hot-spot formation, in accordance with the Koros-Nowak technique and filled into the tube. Lastly, another 400 mg of Silica beads stabilized with quartz wool is placed in the tube to fix the catalyst bed. A picture of a similar setup is shown in Figure ). |
67483870f9980725cf4fb11f | 7 | The identical experimental procedure is conducted for both polymeric reactors. A scheme of the experimental program is shown in Figure . The catalyst is reduced in-situ at 200 °C and ambient pressure (about 960 mbar). Herein, 10 mLSTP min -1 of H2 are diluted with 90 mLSTP min -1 N2. Every set of process condition, including the reduction step, is held for 8 h. After reduction, the feed gas stream is switched to synthesis gas with a H2:CO2 ratio of 4:1 and the first parameter variation is conducted. The first set of process conditions with a 4:1 syngas ratio at 200 °C is used as reference conditions (Reference I). The temperature is varied in 20 K steps from 160 up to 220 °C and again down to 160 °C. After the temperature variation, the process conditions are again set identical to Reference I as a second reference (Reference II) to evaluate the catalyst stability with respect to varying temperature conditions. Afterwards, the pressure is increased to 5 bar. The process conditions of Reference III are 5 bar pressure, 4:1 syngas ratio and 200 °C. Then the syngas ratio is varied from 3:1 to test under-stoichiometric ratios and to 5:1 for testing over stoichiometric ratios. Reference IV is used to evaluate deactivation compared to Reference III in regards to varying gas inlet compositions. |
67483870f9980725cf4fb11f | 8 | The temperature gradient through the reactor wall is evaluated by recording the outer wall temperature and the temperature in the inner tube center at the height of the catalyst bed. The thermocouple at the reactor wall has the same diameter as the kerf it is placed in, therefore it is in ideal contact with the outer reactor wall and the aluminum jacket. Visual confirmation of the correct positioning is assured due to the transparency of the reactor wall, where the catalyst bed is visible, and by measuring the insertion length of the thermocouple in the inner capillary. Since the inner tube is made of stainless steel with a thin wall, heat transfer limitations are negligible and the temperature measured in the inner tube is assumed identical to the mean temperature of the catalyst bed. |
67483870f9980725cf4fb11f | 9 | The catalysts are characterized (3Flex, Micromeritics) by N2 physisorption, H2 chemisorption, CO2 chemisorption and H2 temperature programmed reduction (H2-TPR) measurements. For the H2-TPR experiments samples are dried under flowing Ar at 120 °C for 30 min, cooled down to 50 °C and then the feed is switched to 10 % H2 in Ar. After the TCD baseline signal stabilizes, the temperature is increased to 1000 °C with a heating rate of 10 K/min. The chemisorption experiments are conducted with the volumetric method. Circa 100 mg of each sample are reduced with H2 at 200 °C for 4 h and degassed at 300 °C for 6 h. The adsorption isotherms are then recorded at 50 °C. After degassing for 1 h, a repeat measurement is taken. Both chemisorption isotherms are measured and the Sinfelt method is applied to calculate the adsorbed amount by conducting a linear fit through both isotherms and subtracting the y-intercepts. |
67483870f9980725cf4fb11f | 10 | After the reaction experiments, three sections of the polymer tubes upstream of the catalyst bed, at the height of the catalyst bed and downstream of the catalyst bed are cut out and studied in a light microscope (OZP558, Kern Optics) to visually assess deposition of material on the polymers as well as abrasion and deformation of the tube. The microscope is equipped with SWH10X objectives and the images are recorded by an ODC825 camera. Pictures are taken in light and dark conditions to distinguish changes at the surface level and inside the polymer. |
67483870f9980725cf4fb11f | 11 | Confocal Raman microscopy is performed with a WITec Alpha 500 Raman microscope. From each sample surface, a Raman map with 50 × 50 spectra with a resolution of 2 μm is recorded. A frequencydoubled Ne:YAG-Laser (532 nm) with a power of 10 mW and a 50 x lens with N.A. = 0.55 is used. The integration time is 0.5 s per spectrum. The dataset is then decomposed by non-negative matrix factorization (NMF) as described below to obtain the components and their spatial distribution (maps). |
67483870f9980725cf4fb11f | 12 | A multivariate analysis of the Raman spectrum is performed for dimensionality reduction of all experimental Raman spectra. Inherent features of the investigated samples are extracted from the data set using NMF as implemented in the Python package "Scikit-learn" . It decomposes every measured sample spectrum (𝑐 ) into a specified number 𝑚 of components (loading plots, 𝑥 ) and their corresponding weighting factors (scores, 𝑤 ) . The original data is then represented by the following matrix equation : |
67483870f9980725cf4fb11f | 13 | Preprocessing of the spectral data is done by vector normalization and asymmetric least squares (ALS) baseline correction . The NMF classifier with 𝑚 = 3 components is trained on the complete data sets of 7500 spectra with 1,600 channels each. The data sets are decomposed using the pre-trained NMF classifiers, and the resulting component spectra are matched with reference spectra from commercially available Raman data bases (STJapan, ~ 5,000 spectra) to assign them to polymers and deposits respectively. Transformation of the data set using the trained NMF classifiers provided scalar concentrations of every resulting component and their spatial distribution (maps). |
67483870f9980725cf4fb11f | 14 | Figure is showing the TPR results for the applied Ru-based catalyst, which is compared to a typical Ni/Al2O3 catalyst, which was found to be active at 240 °C in an earlier study . The reduction of Ru starts slightly above 50 °C and is completed within the operation window achievable in PFA and PTFE reactors. The reduction of the Ni/Al2O3 catalyst starts at 200 °C and exhibits a peak above the maximum operation temperature of the polymers. Ni or Co based catalysts are therefore not suitable for the application in the polymeric reactors used here due to their high reduction temperatures above the operating temperature of both PTFE and PFA. However, Ru-based catalysts are intentionally used for low temperature hydrogenation, because of their activity at temperatures as low as 200 °C . The sorption measurements of the Ru based catalyst are shown in Table . The surface area of the Accu ® Sphere support is lower than for a typical γ-alumina supported Ni-catalyst, which is given as comparison. Consequently, the active surface area per gram catalyst is also lower. Interestingly, the sorption capacity and thus the affinity of the gases CO2 and H2 is remarkably different for the two catalyst materials. While the Ni-based catalyst has a higher affinity to CO2 compared to H2, the Rubased catalyst displays a higher affinity to H2. The different specific sorption capacities of both catalyst materials are caused by the nominal metal loading (1 wt.-% Ru and 10 wt.-% Ni). In our previous study we investigated different support materials for Ru and found the H2 sorption capacity for Ru supported on TiO2 (P25/20, Evonik) to be 7.5 µmol g -1 , which is in line with the capacity determined here. |
67483870f9980725cf4fb11f | 15 | The results for 𝑟 are displayed together with the directly measured bed temperature 𝑇 in Figure and Figure . The figures are structured by white and grey background into variations (grey) and reference points (white). In Figure and Figure the activity and temperature are shown over the whole campaign for the PFA and the PTFA reactor, respectively. The comparison of the activity between reference I and II as well as between reference III and IV reveals stable catalyst performance during the whole duration of the campaign with marginal changes during variation II at higher pressure. This is in contrast to other recent studies investigating Ru-catalysts in low-temperature CO2 methanation . When comparing both reactors, higher temperatures and CO2 consumption rates were measured in the PTFE reactor, as can be seen when comparing the data at 220 °C in Figure and Figure . A simple estimate using the Arrhenius equation and power-law kinetics shows that the difference in activity is unlikely to be due to the increased temperature alone. |
67483870f9980725cf4fb11f | 16 | While not in the focus of this study, permeation through the polymeric reactor material is possible, in principle. This may cause that Oxygen is present in the reactor, which reacts with H2 or leads to partial catalyst oxidation. To trace penetration of air into the reactor, N2 was used as indicator. Indeed, N2 was detected at a level of a few hundred ppm, as shown in Figure as function of time. Furthermore, the bed temperature for the PFA reactor is plotted, as well. Further results are available in the supporting information. The N2 level is decreasing for about 16 h until an approximately stable concentration of 350 ppm stabilizes at 180 °C. During variation I, the N2 concentration correlates with the temperature, which is also observable for PTFE (see supporting information). The increase in pressure up to 5 bar is accompanied by an increase of N2 by about 200 ppm. The change in N2 concentration during variation I can be explained by the change in permeability with temperature, but the concentration step after the increase to 5 bar has to have another reason. The profanest reason would be a permeation through the membrane of the backpressure regulator, which is controlled by nitrogen-based pilot pressure. However, this reason is unlikely, as this effect is absent for the PTFE reactor. Hence, the reactor material itself is responsible for the observation, which obviously becomes more permeable under induced stress. It has to be noted that such effect is not yet reported in literature for fluoropolymers, while for glassy fluoropolymers an opposite effect is indicated, especially under the influence of CO2 . Hence, further elaboration is needed to understand the air penetration through PFA, even though this is of limited relevance for application as reactor material on Mars. Importantly, both experimental campaigns show a high sensitivity of the catalyst toward the temperature, but a much smaller effect of pressure and composition changes. A fivefold pressure leads to a doubling of the rate; moreover, a change in feed composition has neglectable impact on the rate. |
67483870f9980725cf4fb11f | 17 | The results are interesting for feed tolerant operation and are in-line with other recent studies by our group, where the same behavior was found for a composition variation at 12.5 bar for TiO2 and ZrO2 supported catalysts performed in conventional stainless steel reactors . Hence, our experimental results indicate performance characteristics, which are beneficial for scenarios where tolerance and robustness play important roles. Thus, our study underlines that not just the polymeric reactors, but also the Ru-based catalysts are highly attractive choices for applications beyond earth in general and space mission in particular. |
67483870f9980725cf4fb11f | 18 | In a first attempt, we still recognized the ignition of the catalyst during operation, which could be critical due to the high heat capacity of the aluminum jackets leading to a time delay of minutes when it is necessary to reduce the heating temperature. In order to ensure safe operation, the higher catalyst activity was therefore not just compensated by a lower specific reaction volume to surface ratio, but further decreased by dilution of the catalyst bed. These safety measures are necessary for the present initial proof-of-concept lab experiments, but can be omitted in later stages of the technology development. |
67483870f9980725cf4fb11f | 19 | The resulting temperatures are displayed in Figure . The bar graph shows the temperature at the outer wall and in the bed for both polymers. Please note that the bars are partly transparent to show that the temperatures did not change for the repeating temperatures. It can be seen that the temperatures are comparable for both polymers and always exhibit a higher temperature inside the bed. The differences are more prominent looking at the temperature differences 𝛥𝑇 on a separate scale as displayed by the points. Here, the temperature difference is clearly increasing between inside the reactor and at the outer wall with increasing operation temperature. This does of course correlate with the different temperaturedependent catalyst activities. Moreover, the depiction shows a higher temperature difference for PFA than for PTFE; although, the activity in the PFA reactor tended to be lower as described in the section above. The higher temperature difference in the PFA reactor may be explained by a lower heat transfer through the material, as the manufacturer states heat transfer coefficients of 0.24 and 0.19 W m -1 K -1 for PTFE and PFA, respectively. In general, the PTFE reactor (see Table ) had a higher catalyst activity than the PFA reactor (see Table ) and thus a higher heat generation throughout the investigation. While the bed temperature is slightly underestimated for 160 to 200 °C, the theoretical temperatures are equal with the measured ones for 220 °C. It can be seen that the measured bed temperatures of the PFA and PTFE reactors are generally in accordance with the theoretically expected values. However, the discrepancy between the theoretical temperature increase, and the measured ones have a small deviation, always below one Kelvin. Thus, the difference between PFA and PTFE depicted in Figure is unspecific being within the typical margin of a temperature measurement. |
67483870f9980725cf4fb11f | 20 | The post-reaction material was analyzed by optical methods. For the PFA reactor, some changes were already visible from the outside due to the excellent transparency of this material. The opened reactors showed some obvious changes at the inner tube wall being a dark powder and some embossments with the size of the catalyst particles (see Figure ). These observations were only made for the zone in which the catalyst and the inert glass beads were mounted. The microscopic images (Figure ) were recorded in different light modes to highlight different details. |
67483870f9980725cf4fb11f | 21 | These images show four different phenomena being the embossed particles, adhesive glass wool fibers, the dark powder and also some details from the material processing. The dark powder is mainly present in the catalyst zone, while the greyish details in the inlet and exit zone originate from refraction of light by colorless powder and crushed glass fibers. Especially in the catalyst zone, the dark powder is in the same positions as the embossed particles. All images can be found in the supporting information. The combination of the embossed particles with the dark powder lead to the question whether the found results are indicating a decomposition of polymeric material under influence of catalytic material. On the one hand, the temperature is reasonably below that required for thermal decomposition of the material; however, the catalyst particles could locally heat to higher temperatures than the measured values. On the other hand, it could be possible to catalyze the decomposition of the polymers by the active material, which is reported for fluoropolymers and Ruthenium complexes . While, our results provide no evidence for significant decomposition of the polymer, a more detailed investigation of this issue is matter of future work. |
67483870f9980725cf4fb11f | 22 | The post reaction polymeric tubing was analyzed by confocal Raman microscopy for chemical changes due to the exposition to catalyst material under the harsh conditions of the chemical reaction. The regions of interest were selected to include some visible features that are observed after the catalytic reaction i.e., deposits, fiber fragments and embossings (Figure and). The Raman image is created as a twodimensional information leading to pixels of ca. 2 µm edge length, with specific information about the composition. It is important to note that the Raman images were mapped in three planes (top, vertical and horizontal) respectively for all samples and regions analyzed. |
67483870f9980725cf4fb11f | 23 | As seen before, the exit zone of the reactor does not exhibit many of these optical features. Therefore, the exit zone was not investigated, but additionally the virgin material was analyzed as reference. The regions for the Raman analysis of the three different sample locations can be seen as microscopic images in Figure . Please note, these images are normal light microscopic images with two-dimensional information. The virgin materials (before use) have a characteristic surface structuring, which is originating from the different material processing during manufacturing of the tubing. The PTFE surface has some sort of striations, while the PFA surface is exhibiting a crater structure on the surface, likely due to the different extrusion methods as mentioned previously (see 2.1.1). The used material of the inlet and catalyst zone keep some of the characteristic surface structures, but additionally exhibit deposits in fiber form and some fragments, which might also originate from the fibers. |
67483870f9980725cf4fb11f | 24 | The light microscopic images in Figure show the different planes of Raman mapping. The view from the top plane (xy) is indicated by a red rectangle, and the view from the vertical (zy) and horizontal (zx) planes are blue and green, respectively. The inner surface of the used reactors looks jagged and rough and the PTFE tubing has some inorganic deposits on the surface, which can be catalyst material. The size, in the range of a few micrometer indicate that these deposits are most likely fragments of the originally fractionated catalyst material. The fibers and fragments thereof identified with optical microscopy are not detected with Raman microscopy. This can be due to negligible signal intensity compared to the main signal, glass fibers or fragments which have no Raman signal or due to optical effects. Interestingly, the embossed particles can be seen in the Raman spectra of both inlet zones. However, the depth of the embossed spheres is less prominent than expected from the light microscopic images, in the range of a few micrometers, in the vertical and horizontal planes of the Raman maps. |
67483870f9980725cf4fb11f | 25 | Figure : Raman maps of the inner reactor surface of both reactors before use and after reaction in the catalyst zone and the inlet zone; each map is 100 x 100 µm 2 ; Raman maps were taken from planes depicted in Figure ; red channel: PTFE/PFA, green channel: inorganic oxides like titanates (for PTFE) and no match for PFA, blue channel: no match for both polymers. |
67483870f9980725cf4fb11f | 26 | It is important to note that there are no carbon deposits and no indication of polymer decomposition at the inner surface. For both polymeric materials, no chemical degradation of the material is observable and no structural changes could be seen near the embossed catalyst particles. These results indicate that under the prevailing reaction conditions between 160 and 220 °C over 100 h, the catalytic material does not initiate chemical degradation reactions of the polymer. Certainly, the results should be interpreted in the way that the polymers are most probably not chemically decomposing, but the mechanical stress might lead to structural damage of the surface. This is potentially critical for the mechanical integrity of the reactor, since the small notches induced by glass fibers or catalyst particles result in stress concentration. Phenomena like these can finally lead to reactor failure and need further investigation, in order to consider these effects in the derivation of criteria for the design of polymeric reactors. |
67483870f9980725cf4fb11f | 27 | The propellant production on Mars needs to be completed within a tight timeframe given by the mission scenario. There are several mission scenarios under consideration, which are all depending on the transfer trajectories between Mars and Earth . The base calculation for this study will use the time of a Hohmann-transfer 𝑡 , which takes approximately 258 days . This is the time from start on earth to arrival on Mars and thus the minimum time between the confirmed start-up of a propellant production plant on Mars and the arrival of a manned mission. The fuel capacity of the Starship by SpaceX is used as reference for the production scenario with 240 t of CH4 and a total capacity of 1.200 t (fuel and oxidizer). |
67483870f9980725cf4fb11f | 28 | Note that the CO2 consumption rate, 𝑟 , is identical to the formation rate of methane, since we assume no by-product formation. The catalyst mass 𝑚 can now be calculated according to Eq. ( ) and a value of about 56 kg is derived. Improvements of the specific methane formation rate by an order of magnitude is realistic by increasing the Ru-loading from 1 wt.-% to above 10 wt.-% and by rising the operation temperature to 220 °C . |
67483870f9980725cf4fb11f | 29 | For comparison, Zubrin et al. developed the integrated Mars in situ propellant production system (IMISPPS) and reported the mass of the optimized system of 54 kg for a O2:CH4 production rate of 1 kg/d, which corresponds to 0.33 kg/d CH4 . For experimental tests, the authors employed 130 g of reverse water-gas shift catalyst and 0.5 g of 0.5 wt.-% Ru/Al2O3 methanation catalyst and operated the reactor at 400 °C. We achieve a specific methane production rate directly from CO2 of about 6 µmol g - 1 s -1 at 200 °C (see Table , Table ), which corresponds to 6.9 kg kg -1 d -1 (methane per catalyst basis) |
67483870f9980725cf4fb11f | 30 | and therefore a methane production capacity of ca. 0.90 kg/d assuming 130 g catalyst, as well. Hence, our concept in the conservative case provides the threefold production capacity as reported by Zubrin et al., but at significantly lower temperature. On top of that, the production rate can be doubled by operating at 220 °C, while higher Ru-loadings will increase the specific productions rates by another factor of 10 or more. Since the use of polymeric materials in chemical reactions system may lead to further mass reduction of the production unit, very attractive mass specific productivities are possible after further optimizations. |
67483870f9980725cf4fb11f | 31 | Our experimental results proof the theoretical concept of polymeric reactors for thermocatalytic reactions up to 220°C under elevated pressure for methanation as typical example. The successful proof is based on the fact that the used Ru/TiO2 can be activated and is stable within the temperature limits of the polymer. Our results successfully demonstrate a significant and stable production rate of methane under a wide range of process conditions with sufficient capacity for in-situ propellant production on Mars. |
67483870f9980725cf4fb11f | 32 | Furthermore, extensive investigations are performed to reveal information on mechanical strength and heat removal performance of the polymeric reactor material. We found a certain temperature increase during reaction, which agrees with calculations. Moreover, the polymeric material was deformed by the catalyst particles and inert glass beads resulting in embossments of the particles in the polymeric reactor wall. However, both mechanical strength and heat removal performance is sufficient, while Raman microscopy confirm the absence of significant chemical degradation of the polymer. Hence, this study establishes the feasibility of using polymeric reactors for thermocatalytic processes, which opens the door to novel applications both on Earth and in space. |
659fffab9138d23161b1e894 | 0 | Metathesis reactions are among the most important chemical processes in modern organic synthesis. They are used to efficiently break and rearrange carbon-carbon bonds obeying excellent atom economy, both in industrial-scale applications, as well as in chemical and pharmaceutical research. From a mechanistic perspective, olefin metathesis (OM) is based on a [2+2]cycloaddition/-reversion cycle that is catalyzed by proficient and well-defined metal-alkylidene complexes involving a metallacycle intermediate. Cyclopropanation can be an undesired side reaction in this process (Figure ). For their pioneering work in this very field, Chauvin, Schrock and Grubbs were awarded the Nobel Prize in chemistry in 2005. Consequently, the most widely used types of catalysts for metathesis (of any type and flavor) are high oxidation state Molybdenum complexes developed by Schrock and the Ruthenium based systems later developed by Grubbs. Important modifications to the existent systems and contributions to this field have been reported by Hoveyda and Copéret, as well as Buchmeiser, Blechert 10 and Nolan. Even though the current catalysts provide high yields under mild conditions, the development of sustainable, non-toxic and cheap base-metal catalysts is highly desirable. One area of research, that is currently of significant interest, is the substitution of ruthenium with iron, the corresponding 3d transition metal of the first series of the periodic table. Numerous research groups have been working on the design of operational iron systems for OM, both experimentally and theoretically. Within this context, the research groups of Bukhryakov and Milstein have recently reported on new low-valent iron complexes to be active in ring-opening metathesis polymerization (ROMP). Recent advances regarding the chemistry of iron carbene complexes and iron-based cycloaddition can further be attributed to Illuc and Meyer. While iron is well accepted as a surrogate for ruthenium in OM, it is somehow surprising not to find any investigations or at least considerations on the use of manganese for the same purpose. |
659fffab9138d23161b1e894 | 1 | In recent years, several catalytic transformations involving Mn(I) complexes have been reported to exhibit similar structure, reactivity and stability as compared to established Ru(II) and Fe(II) systems. Alike the similar chemical properties between elements of the second and third period, it can be assumed that Ru(II) and isoelectronic Mn(I) comprise a diagonal relationship in the periodic table. This relationship is strongly inspired by the successful application of vanadium alkylidenes in OM, mimicking the reactivity of related Mo complexes. Unfortunately, the total number of currently known and stable manganese carbene complexes is small and mostly comprised of Fischer-type complexes. In contrast to Mo and Ru based catalysts, these compounds readily undergo cyclopropanation reactions. Fischer-type carbenes are defined as M=CRX with X being a hetero atom and are thus electrophilic by nature, in contrast to Schrock-type carbenes (M=CR 2 ). The only Mn system that resembles elements of OM was reported by Braunschweig and coworkers in 2013 and involves the formation of [MnCp(CO) 2 (=CPh 2 )] via cycloreversion of a Mn borylene metallacycle. Consequently, manganese catalyzed olefin metathesis has (in contrast to iron) neither been theoretically studied nor experimentally attempted before. |
659fffab9138d23161b1e894 | 2 | All approaches to use elements of the first transition metal series for olefin metathesis face certain intrinsic challenges that need to be addressed in order to devise an active catalyst: (i) The metal-carbene (M=CR 2 ) bond tends to be weaker in 3d metal systems than in 4d/5d transition metal systems. This fact leads to cyclopropanation being the preferred reaction in first row transition metal (carbene) complexes. (ii) While heavier metals such as Ru and Mo favour low-spin closed shell configurations, first row metals usually exhibit a wide range of possible spin states, which are close in energy. (iii) Surface crossings between singlet and triplet states are likely to take place. (iv) Open shell systems hamper the coordination of an olefin and could very likely also undergo other radical reactions. However, in the absence of experimental data, theoretical studies can aid to guide experimental synthesis, predict trends and patterns, and exclude unsuitable systems. Due to the complex electronic structure and the multitude of accessible spin states in Mn (and Fe) systems, the computational chemist faces the problem of accurately describing these states and their relative energies. For these reasons, it is not surprising that the first Fe systems to show activity in OM were discovered years after their computational exploration. In this contribution, we explore the feasibility of low-valent manganese alkylidene species complexated by mono-, bi-and tridentate ligand-systems for potential use in OM by means of density functional theory. The focus is on the comparison of isostructural Fe/Mn systems, as well as on the conception of systems that are building upon already synthetically viable precursors. |
659fffab9138d23161b1e894 | 3 | Comparison of isoelectronic Fe and Mn systems In pursuit of an in silico operational manganese system for olefin metathesis, we developed a series of suitable model complexes that allow for a comparison with already known and investigated systems regarding spin state splittings, electronic structure and thermodynamic parameters. For a direct comparison with a literature known Fe system, we first adopted the simple model catalyst 1-Fe reported by Mauksch and Tsogoeva 22 and substituted the putative Fe(II) center with formally isoelectronic Mn(I). The main focus of this 2017 publication was to demonstrate the possibility of low-valent iron for OM based on orbital symmetry and aromaticity arguments. |
659fffab9138d23161b1e894 | 4 | For simplicity reasons, only one topological isomer of this system was considered, bearing the coordinated olefin, a CO ligand and the carbene fragment in the equatorial plane. Secondly, we adopted a tridentate pincer-type carbene system with a benzene backbone (2-Mn) as devised by Solans-Monfort and coworkers. The latter work concluded that Fe=CH 2 fragments and metallacyclobutanes are ideally stabilized in the singlet state utilizing ECE pincer ligands (E = NHC, PR 2 ) with a strong σ-donor present in the central position. |
659fffab9138d23161b1e894 | 5 | For these model systems, relevant stationary points along the OM reaction path were calculated. This includes the initial olefin adduct, the corresponding metallacyclobutane structure, as well as the transition states (TS) for [2+2] cycloaddition, cycloreversion and cyclopropanation for all relevant spin states. Since all model complexes exhibit a d 6 electron configuration, a low-spin (S = 0) singlet state, a triplet state (S = 1) and a quintet high-spin (S = 2) state can be relevant. For calculations of the electronic structure and thermodynamic parameters, we adopted the meta-hybrid exchange-correlation functional TPSSh (with 10% HF exchange), as it was found to accurately predict spin state splittings of manganese and iron based spin cross-over compounds. Nevertheless, other density functionals were also used, as well as DLPNO-CCSD(T) calculations were performed on selected structures to gauge the accuracy of the methodology. |
659fffab9138d23161b1e894 | 6 | In order to gain a better understanding of the M=CH 2 (M = Fe, Mn) bonds in our singlet model systems, we first carried out energy decomposition analysis (EDA-NOCV) as applied and advocated by Frenking and coworkers. The EDA decomposes the interaction energy ∆E int between two fragments within a molecule into three different components, namely ∆E elstat , ∆E Pauli and ∆E orb , in order to obtain a chemically meaningful bonding picture. Thereby, either two closedshell fragments that interact or two triplet fragments that interact in an anti-parallel fashion can be generated in the EDA. Whichever fragmentation yields values closer to zero for the orbital mixing term ∆E orb determines the chemical interpretation of the bonding. If the two singlet fragments yield a lower value, the bond is termed dative, whereas if it is the two anti-parallel triplet fragments, the bonding is referred to as electron sharing. Metal carbene complexes with an electron-sharing bonding are traditionally classified as Schrock-type (alkylidene), while those with a dative bonding as Fischer-type. Table shows a compilation of the energy values obtained by decomposition analysis with TPSSh/TZ2P/D3. Complex 1-Fe can be described as electron-sharing bearing two interacting triplet fragments, while the carbene bond in 1-Mn can rather be described with a dative interaction. However, the differences for the decisive ∆E orb values in the manganese and iron systems are small. To further investigate the nature of the M=CH 2 bond, Wiberg bond orders were calculated to be 1.70 in the Fe and 1.91 in the Mn complex, respectively. This analysis points to a more pronounced double bond character in the Mn system. According to EDA, the model catalysts 2-Fe and 2-Mn can both be described as having two interacting triplet fragments and can therefore be classified as Schrock-type carbenes. The associated bond order is 2.11 for the Fe system and 2.19 for the Mn system, respectively. Consequently the double bond character in these complexes is more pronounced than in 1-Fe and 1-Mn. In all cases investigated, the total interaction energy ∆E int of the closed-shell fragments was much lower (negative) than that of the triplet fragments. Based on this limited data set, no clear differentiation of Fe vs. Mn carbene complexes can be made. Within one pair of complexes Fe and Mn behave rather similar, while the differences attributed to the different ligand spheres are more pronounced between the two pairs, 1-Fe/1-Mn vs. 2-Fe/2Mn. A comparison with related Ru based carbene species, namely 2-Ru, the analogous structure to 2-Mn and a second generation Grubbs catalyst, revealed that these also favor an electron-sharing bonding mode. At least within the EDA, the carbenes can be considered as similar -highlighting once more the diagonal relationship of Ru and Mn. Details about these calculations can be found in the Supporting Information (see table ). |
659fffab9138d23161b1e894 | 7 | The reaction energy profile for the simple model catalyst 1-Mn with a Grubbs-type N-heterocyclic carbene (NHC) moiety is depicted in Figure . The starting point in this olefin metathesis cycle Table : Results of the energy decomposition analysis (EDA) of the model compounds 1-Fe, 1-Mn, 2-Fe and 2-Mn. Dative bonding results from interaction of two singlet fragments, whereas electron-sharing bonding results from triplet-triplet interactions. All energies are given in kcal/mol obtained using TPSSh/TZ2P/D3. |
659fffab9138d23161b1e894 | 8 | plays a barrier of 12.7 kcal/mol for cycloaddition of ethylene and a barrier of 37.7 kcal/mol for decomposition via cyclopropanation, hence, a larger separation of the two pathways. Unlike, Ru complexes, the singlet-triplet energy splitting for the metallacyclobutane intermediate is one of the most crucial parameters in first-row TM metathesis and for 1-Mn was calculated to be ∆G ST = -5.7 kcal/mol with TPSSh and ∆G ST = -5.4 kcal/mol employing DLPNO-CCSD(T) in favor of the triplet state. A survey on the performance of various exchange-correlation functionals on the singlet-triplet splitting for metallacycle 1-Mn can be found in the electronic Supporting Information (see Table ). Essentially all tested functionals yielded similar trends. Coming back to 1-Mn, the quintet state of the metallacyclobutane is located just 0.4 kcal/mol above the triplet state and could also play a role in surface crossing events and deactivation processes. Similar to the reported iron system (1-Fe), the transition state for the [2+2]-cycloaddition in the singlet state also exhibits aromatic character with a NICS(1) value (NICS = Nucleus Independent Chem- Next, we investigated the reactivity and reaction energetics of 2-Mn, where Mn is coordinated by an anionic PCP pincer-type ligand. Similar to the iron CCC, PCN and CCN pincer complexes investigated by Solans-Monfort, the M=CH 2 unit in the initial carbene complex 2-Mn (as depicted in Figure ) is directed away from the PCP plane at an obtuse angle. In the triplet state, 2-Fe and 2-Mn both have an almost ideal square planar geometry. In sharp contrast to the reported Fe systems, the initial carbene complex 2-Mn in the triplet state is 20.5 kcal/mol more stable than in the envisioned singlet state. This finding is also supported by OPBE calculations with a ∆G ST value of -10.3 kcal/mol. |
659fffab9138d23161b1e894 | 9 | The reaction energy profile for olefin metathesis with 2-Mn is shown in Figure . Again, the depicted reaction profile starts with the olefin adduct that can be generated by coordinating an olefin to 2-Mn. The formation of the olefin adduct from the separated singlet reactants is favored by ∆G = -20.4 kcal/mol (not displayed), while the olefin adduct is more stable in the triplet than in the singlet state. The energetic barrier for the initial [2+2]-cycloaddition on the singlet surface is 11.2 kcal/mol and leads to the formation of a singlet metallacycle with a trigonal bipyramidal geometry. Similar to the situation of 1-Mn, the metallacyclobutane intermediate is more stable in the triplet state than in singlet state with ∆G ST = -7.3 kcal/mol. This trend is again supported by OPBE calculations yielding a ∆G ST value of -12.9 kcal/mol. The barriers for cyclopropanation were calculated to be 30.4 kcal/mol in the singlet state and 18.9 kcal/mol in the triplet state, re- spectively, and are higher in energy than productive OM. In addition, the total cyclopropanation reaction is an endergonic process for both spin states. Note that cycloadditon and cycloreversion are symmetry related processes in this specific example and therefore starting points and endpoints, as well as corresponding transition states have the same energy (Figure : Cycloreversion for the S = 1 state is omitted as it is also 5.5 kcal/mol). We would like to point out that, despite of the shortcomings of the system, the metathesis process is preferred over cyclopropanation in both, the singlet and triplet state, which is an important necessity for OM. This also constitutes a major difference to iron-based systems such as 2-Fe, where the triplet state inevitably leads to decomposition. As with 1-Mn/1-Fe, also in the case of 2-Mn/2-Fe, the spin-state ordering in the Mn variant is less sensitive to the chosen density functional than in the Fe counterpart. |
659fffab9138d23161b1e894 | 10 | To further investigate the influence of the donor strength and design of the supporting pincer ligands, we considered three additional frameworks. As for OM, the metallacyclobutane structure is the key intermediate and should ideally be in the singlet state. Hence, we investigated the impact of these ligands on ∆G ST in the corresponding metallacyclobutane complexes (see Figure ). For 3-Mn a singlet-triplet splitting of -18.3 kcal/mol was found, for 4-Mn a splitting of -7.2 kcal/mol and for 5-Mn a value of -8.6 kcal/mol, respectively. In all these cases (including 2-Mn, vide supra) the triplet state remains the more favorable one, but a strong influence of the ligands can be observed. Similar to the computational studies of iron complexes, the effect of the central donor is pivotal, but the stabilization of the singlet state in Mn systems is much more difficult to achieve than in related Fe systems -at least based on the limited existing data. It is also particularly noteworthy at this point that the deviations from ideal <S 2 > in the calculated triplet states are significantly higher for the calculated Mn pincer species than for the corresponding Fe pincer species -indicating a more complex electronic structure of the Mn systems. Earlier investigations by Chirik and coworkers already revealed the complexity of the electronic structure of pincer coordinated Fe carbene complexes. It would certainly be of interest to examine the influence of various pincer ligands and substituent effects 53 in more detail, but this is beyond the scope of this current article. spectively. Applying this concept to the metallacyclobutane structure derived from 1-Mn, we computed δ(α) = 119 ppm and δ(β) = -11 ppm. A similar and even better result is obtained for 2-Mn with δ(α) = 115 ppm and δ(β) = 4 ppm. Even though these rules were primarily derived from Ru, Os, Mo, W and Ir complexes, the values obtained for both model complexes were found to match well and are in line with our assumption on an isodiagonal relationship. |
659fffab9138d23161b1e894 | 11 | Lastly, surface crossings between singlet and triplet surfaces must not be neglected for the investigated complexes and require a closer inspection. A well-known concept in this regard is the use of minimum energy crossing points (MECP) to approximate the adiabatic transition between two spin states. Such processes are formally spin-forbidden in the absence of spin-orbit coupling. |
659fffab9138d23161b1e894 | 12 | As suggested by Harvey, these points can be calculated using an optimization algorithm utilizing combined gradients for the singlet and triplet PES to eventually yield a geometry where ∆E ST = 0. As mentioned before, our simple test system 1-Mn (see Figure ) displays a metallacyclobutane intermediate with the triplet state being 5.7 kcal/mol more stable than the singlet state. |
659fffab9138d23161b1e894 | 13 | We could localize a MECP at ∆G MECP ≈ 2.3 kcal/mol above the singlet state, low enough to allow thermal deactivation of the catalyst by surface crossing. Similarly, for the PCP complex 2-Mn (see metallacycle in Figure ) a MECP could be localized at ∆G MECP ≈ 2.8 kcal/mol above the singlet state. These surface crossing points are well-accessible at reaction conditions and can compete with all other reaction steps. |
659fffab9138d23161b1e894 | 14 | Synthetically more feasible model systems While it is well-known that present OM catalysts in their active form are usually 4-fold coordinate species with a low d-electron count, even simple base-metal model systems such as 1-Mn or 2-Mn are very challenging to realize experimentally. The preparation of low-spin Mn(I) complexes is commonly based on the reaction of a suitable carbonyl precursor such as Mn(CO) 5 Br or Mn 2 (CO) 10 with bidentate or tridentate ligands and subsequent chemical modifications. Moreover, chemical experience dictates that the presence of carbon monoxide is an integral component of most low-spin manganese complexes and that a coordination number of five or six is usually encountered in these molecules. Even though CO is usually not found as a co-ligand in active OM catalysts, it is worth to explore these possibilities computationally since CO ligands can be helpful in maintaining a singlet ground state throughout a catalytic cycle. Figure depicts a conceptual extension of the presented complexes 1-Mn and 2-Mn. A possible starting point for a synthetically feasible compound such as 6-Mn could be a Mn(I) |
659fffab9138d23161b1e894 | 15 | dicarbonyl complex [Mn(PCP)(CO) 2 ], similar to as described by Tonzetich and coworkers. Depending on the stability of the depicted system under experimental conditions, either a direct reaction and formation of an olefin adduct or metallacycle (7-Mn) or alternatively loss of remaining CO to eventually form a 4-fold coordinate complex such as 2-Mn (vide supra) seem possible. The electronic effect of including a CO ligand in the carbene complex 6-Mn is clearly expressed in the stabilization of the singlet state and a ∆G ST value of 8.1 kcal/mol. Note that the corresponding complex without CO is favored in the triplet state. However, an adverse effect is observed for the singlet-triplet splitting in the metallacycle 7-Mn: A 6-fold octahedral coordination imposes strong geometrical constraints on the metallacycle and forces the metallacycle to adopt a distorted square pyramidal geometry rather than a trigonal bipyramidal coordination as observed naturally for 1-Mn and 2-Mn in the singlet state. As a consequence, the singlet state is destabilized in comparison to the triplet state. Accordingly, a singlet-triplet splitting of -6.3 |
659fffab9138d23161b1e894 | 16 | In order to circumvent this problem and at the same time maintain the 6-fold coordination, bidentate ligands can be used. In this respect, 8-Mn represents an octahedral modification and extension of 1-Mn. In contrast to tridentate ligands, bidentate ligands can offer more structural flexibility and are less rigid. Experimentally, a compound such as 8-Mn could be derived from a carbonyl precursor [Mn(NHC-P)(CO) 3 R] (R = Alkyl, Aryl, OAc) or a putative dinuclear precatalyst. To further increase the donor strength of the ligand, a bis-NHC system could be employed, which is also well established in literature. The reaction energy profile for an olefin metathesis cycle with the octahedral species 8-Mn is shown in Figure . For simplicity reasons, again only one isomer was considered in the following and the starting point is the coordinated olefin adduct. The energetic barrier for the initial [2+2]cycloaddition of the simple test substrate ethylene is low (∆G ‡ = 3.5 kcal/mol) and quickly leads to formation of a stable singlet metallacycle. In the absence of surface crossings and isomerisation reactions, the most important side reaction is cyclopropanation and formation of methylcyclopropane. The barrier for this process was calculated to be 22.4 kcal/mol in the singlet state. |
659fffab9138d23161b1e894 | 17 | The cycloreversion process forming the propene adduct at 2.9 kcal/mol displays a barrier of ∆G ‡ = 15.5 kcal/mol. Note, that educts and products are not identical and no symmetry relations are present anymore. The most important observation is the inversion of the spin state ordering (∆G ST > 0) and the preference for the singlet state of the metallacycle compared to 7-Mn. This can be primarily attributed to the presence of the strong π-acceptor ligand CO as an additional ligand, but also to the changes in electronic structure induced by altering the ligand sphere. Consequently, the effect of the CO ligands is two-fold: It stabilizes the system in the singlet state and the octahedral coordination prevents structural relaxation that is observed as a deactivation channel in 5-fold coordinated species such as 1-Mn and 2-Mn. To achieve the desired stabilization of the singlet state, an intricate balance of ligands is required. The incorporation of carbon monoxide into the chemical structure has fundamental advantages but also potential downsides. Regarding the geometry of the olefin adduct 8-Mn in the triplet state, we observed that the obtained minimum is 22.1 kcal/mol higher in energy. However, small displacements led to significant structural rearrangements and CO migration to the carbene (see Figure ). A similar optimization behaviour was also found using other exchange-correlation functionals. Hence, it is not an artefact of the approximated DFT functional. This very stable species exhibits a Mn-C(H)(CH 3 )(C=O) motif and is located 2.0 kcal/mol above the singlet olefin adduct. |
659fffab9138d23161b1e894 | 18 | It should not go unmentioned that migratory insertion of highly nucleophilic ligands such as alkyl groups is also a well-known concept in manganese carbonyl chemistry in the ground state. The quintet states for the adduct and the metallacycle showed dissociative character with respect to ethylene or CO, and thus could not be fully converged. We assume that both observed behaviors, the migratory insertion-like reactivity of the nucleophilic alkylidene in the triplet state and the dissociation of CO or olefin in higher spin states could lead to early deactivation of the catalyst or pre-catalyst. Another potential side-reaction could be insertion of CO into the metallacycle forming a metallacyclopentanone species. However, we have not observed such a behaviour in the singlet or triplet states of our investigated Mn complexes. |
659fffab9138d23161b1e894 | 19 | The present work investigated the potential of olefin metathesis with a series of Mn based complexes by exploring thermodynamics and energetics of homo metathesis with DFT. In the course of our investigation, we first compared isoelectronic iron and manganese complexes and later extended our test set to potentially more synthetically accessible carbonyl complexes. Our study highlights challenges that have to be addressed in the design of functional catalysts. Based on our findings we conclude: (i) The overall thermodynamics and free energy profiles for the metathesis cycle with Mn based systems are feasible and comparable to literature known Fe systems. (ii) |
659fffab9138d23161b1e894 | 20 | Surface crossings are likely to take place if the triplet state is lower in energy. Together, this supports our assumptions on the isodiagonal relationship and the suitability of low-valent manganese systems as they are comparable to iron. Our investigated complexes 1-Mn and 2-Mn do fulfill several requirements for OM such as low barriers for cycloaddition and -reversion, as well as feature electronic descriptors that are indicative for functioning OM -given a singlet spin state. |
659fffab9138d23161b1e894 | 21 | One aim of our investigation was to design systems that could be prepared in a laboratory and are feasible from a synthetic point of view. Mn pincer-type complexes or bidentate systems are well known to literature and known to stabilize Mn(I). They offer great variability in terms of electronic properties and steric demand. Therefore, a very careful choice of the pincer ligand could be sufficient to maintain a singlet ground state. However, based on our findings bidentate systems with additional CO (such as 8-Mn) seem to better suited to stabilize the singlet ground state. |
659fffab9138d23161b1e894 | 22 | Despite these yet to be tackled issues and lack of experimental benchmark data, the high abundance of first-row transition metals provides an opportunity to develop inexpensive and non-toxic catalysts. Our research provides insights into the reactivity of Mn complexes and functions as a reference for the experimental chemist to pave the way for the design of manganese based catalysts for olefin metathesis. |
659fffab9138d23161b1e894 | 23 | program package utilizing the Vienna Scientific Cluster (VSC 4) in part. Electronic ground state calculations, including geometry optimizations, frequencies and transition-state searches were carried out with density functional theory (DFT) utilizing ORCA and the TPSSh meta-hybrid functional 66 together with Grimme's D3 dispersion correction and Ahlrichs' def2-SVP basis set. The resolution of identity (RI) approximation was used along with the corresponding auxiliary basis sets to accelerate the calculations. Gibbs free energies at 298 K were calculated using the rigid-rotor harmonic oscillator (RRHO) approximation as implemented in ORCA and real frequencies below 100 cm -1 were raised to 100 cm -1 to improve accuracy. Final single-point energies were calculated with TPSSh/def2-TZVP/D3 or OPBE/def2-TZVP/D3 70,71 on already obtained geometries, while Gibbs energies were obtained by adding zero point energies, thermal and entropic corrections at 298 K. For geometry optimizations and single-point calculations implicit solvation was included via the CPCM model and dichloromethane as solvent. Transition states were localized with the help of the nudged elastic band (NEB) method as implemented in ORCA and confirmed to be first-order saddle points by analysis of the Hessian. Intrinsic reaction coordinate (IRC) calculations have been employed to verify the correct assignment of transition states by following the eigenvector of the corresponding imaginary frequency, starting from the TS structures and connecting with reactants and products. Minimum Energy Crossing Points (MECP) were calculated using the algorithm implemented in ORCA and the SurfCrossOpt keyword. DLPNO-CCSD(T) calculation were performed on previously obtained TPSSh-geometries, using the def2-TZVP basis set and a converged BP86 UKS reference wave function. For the sake of completeness, it should also be mentioned that the wave functions of all Mn complexes were checked for internal stability: This revealed that the S = 0 species derived from 2-Mn exhibit a slight instability towards a unrestricted Kohn-Sham (UKS) wave function. However, the elec-tronic energy differences are negligible (< 1 kcal/mol). Energy Decomposition Analysis (EDA) was specifically performed in ADF using the TPSSh functional together with the TZ2P basis set and no frozen core orbitals. The numerical quality was set to "very good", and the scalar relativistic effects were accounted for using the Zeroth-Order Regular Approximation (ZORA). For the interaction between triplet fragments, the spin polarization was set to +2 for one fragment and -2 for the other to ensure anti-parallel spins during Energy Decomposition Analysis. Orbital plots and graphics were visualized and generated with ChemCraft. Associated Content |
677c09ac81d2151a0219b31b | 0 | Smell, the conversion of volatile molecules in the environment to information or behaviors by a creature, is foundational to life on Earth. Its inner workings in animals, however, long eluded a detailed description. Particularly mysterious has been the initial step: how odorant molecules are detected by the neurons that transmit their signals to the brain. At the close of the twentieth century, Buck and Axel discovered the family of G protein-coupled receptors expressed in the membranes of olfactory neurons that detect odorants. These olfactory receptors (ORs) are one of the largest multigene families in vertebrates. Humans have about 400 ORs, although mice have closer to 1000 ; they account for almost half of the known G protein-coupled receptors, themselves among the largest families of protein-coding genes . Axel and Buck shared the 2004 Nobel Prize in Physiology or Medicine for their discovery . |
677c09ac81d2151a0219b31b | 1 | Much has been learned about olfactory receptors since then. Each olfactory neuron is dispersed randomly throughout the olfactory epithelium , and expresses only one OR . However, the receptors are highly promiscuous: each OR responds to several different olfactant molecules, and the same olfactant may stimulate several ORs. It is the combination of activated ORs that encodes a particular smell; its combinatorial nature means that humans may be able to sense up to 10 12 different odors . But much, too, remains unclear. For one thing, olfactory receptors are difficult to isolate; the first experimental structure of a human OR, OR51E2, was not identified until 2023 . The details of the mechanism by which olfactants bind to ORs are also unclear. While makes clear that the usual explanation for protein-cofactor binding, a protein binding pocket accepting small molecules based on their shape, is an important component of olfactant recognition, it does not offer a complete explanation. Compounds with very similar structures may not bind to the same OR, and ones with very different structures may do so. What is more, olfactants with only slightly different structures often have vastly different odors; smells are sometimes-but not always-associated with particular functional groups . |
677c09ac81d2151a0219b31b | 2 | There is even some evidence that ORs can distinguish isotopologues: molecules with the same chemical formula and connectivity, differing in only their isotopic composition. Fruit flies and honeybees can be trained to distinguish between hydrogenated and deuterated odorants like acetophenone. However, these results do not extend directly to vertebrates because of the difference the two groups' olfactory systems; insect olfactory receptors are (likely) not even G protein-coupled receptors . Trials in humans, both in vitro and in living subjects , are inconclusive. The extreme sensitivity of the olfactory system to impurities and the subjective nature of smell tests complicates these experiments. |
677c09ac81d2151a0219b31b | 3 | The mechanism by which ORs detect odorants is also uncertain, but if isotopologues can be distinguished then vibrational modes (phonons) may play a role. Apart from their vibrational spectra, which depend sensitively on the mass of their nuclei, two isotopologues have nearly identical chemical properties. ORs, then, might detect olfactants by coupling to their vibrational modes, as proposed as long ago as 1938 . This has a certain intuitive appeal for explaining the dependence of odor on specific functional groups, which leave easily identifiable vibrational traces in infrared (IR) and Raman spectra . Turin proposed a mechanism, inelastic electron tunneling (IET), for vibrational detection of olfactants by receptors. In IET, an electron tunnels from a donor to an acceptor site in the active site of an OR. When there is no olfactant present, the donor and acceptor are separated in both distance and in active energy levels, and electron transfer is suppressed. The olfactant fills the space required for tunneling, and is also hypothesized to have a phonon whose energy approximates the donor-acceptor gap, suppressing the barrier to electron transfer . In biological systems, this mechanism is plausible only if the active site is well separated from the ambient water, implying a much smaller Marcus theoretic reorganization energy than is usual for an aqueous solution at 310 K . IET has been observed at physiological temperatures in artificial circuits , but it is not yet clear whether it occurs in living creatures . |
677c09ac81d2151a0219b31b | 4 | Computational evidence for vibrational effects in olfaction has also been mixed. A three-state model for inelastic electron transfer in fruit flies suggested that vibrational effects play a role in the detection of acetophenone, but these effects were not expected to generalize to other odorant molecules . A more detailed quantum chemical model predicted differential binding of deuterated octanoic acid, but not acetone, compared to its hydrogenated form ; the latter study also observed that a trained dog could partially distinguish between D 15and H 15 -octanoic acid. The experimental structure and associated molecular dynamics simulation of OR51E2 obtained in presents an opportunity to advance by simulation the understanding of vibrational isotope effects on this critical first step in olfaction. |
677c09ac81d2151a0219b31b | 5 | In this work, we present the first computational infrared spectra of an olfactant molecule bound to an OR, including all its possible isotopologues. The high-frequency vibrational modes, above 2000 cm -1 , of the olfactant in its binding site, differ based on isotopic composition. This difference is large relative both to the variance in the modes due to thermal noise and to the thermal energy available under physiological conditions. We do not address a specific mechanism by which vibrations couple to the OR. Indeed, the unprecedented structural detail observed by Billesbølle et al. demonstrates that noncovalent bonds to the ligand and the shape of the binding pocket play integral roles in olfactant recognition. However, we keep open a door for the impact of vibrational modes to be felt-that is, smelt-by the high-energy vibrations of olfactant molecules. |
677c09ac81d2151a0219b31b | 6 | OR51E2 binds to acetate, propionate (C 2 H 5 COO -), and lactate (2-hydroxylpropanate) . It is not found only in olfactory neurons: it is also expressed in kidneys , helping to regulate blood glucose concentration; carotid arteries , sensing blood oxygen content; and sperm cells , aiding in their movement. In this work, we focus on propionate, which is bound to OR51E2 in the single-particle cryogenic electron microscopy (cryo-EM) structure elucidated by Billesbølle et al. . Propionate is a short-chain fatty acid "with a pungent, disagreeable, rancid odor" ; it is, however, an important component of some cheeses, especially Emmental . |
677c09ac81d2151a0219b31b | 7 | In addition to the cryo-EM structure of OR51E2 and propionate, the authors of deposited a 500 ns molecular dynamics (MD) trajectory at 310 K to the GPCRmd repository (Record 1245) . The simulation did not include the associated G protein, but was placed in a simulated lipid bilayer surrounded by water. We sampled twenty-six coordinate snapshots at 20 ns intervals from the trajectory, including the starting structure, to simulate thermal fluctuations in the system, and computed infrared spectra via quantum mechanics/molecular mechanics (QM/MM) in the harmonic approximation. |
677c09ac81d2151a0219b31b | 8 | The propionate binding site is "fully occluded from the extracellular milieu" . We verified with the VMD software that the nearest water molecule is almost 10.9 Å distant from the olfactant; likewise, the nearest lipid molecule is more than 13 Å from C 2 H 5 COO -. Since we expect the relevant vibrational modes to be spatially localized, we remove the solvent, lipid, and counterions from the structure; leaving only OR51E2 and the propionate ligand. |
677c09ac81d2151a0219b31b | 9 | QM/MM calculations at the (point-charge) electrostatic embedding level were performed with the ASH software , using its interface to OpenMM (version 8.1.1) for the MM routines and to PySCF (version 2.6.2) for the QM routines. The ligand alone was defined as the QM region; thus, there were no covalent bonds crossing the QM/MM interface, and no link atoms were required. |
677c09ac81d2151a0219b31b | 10 | From the optimized geometries, we compute the partial Hessian of propionate with ASH numerically, using analytic gradients of the energy with respect to nuclear coordinates from B3LYP-D4. This Hessian was then weighted by the masses of the nuclei and diagonalized, yielding the vibrational normal modes. We did not project out the translational and vibrational modes, as they couple to the rest of the complex and are ill-defined for partial Hessian calculations ; however, we omit them from the following analysis. We additionally obtain the derivatives of the dipole moments along the normal modes, which provide infrared intensities. |
677c09ac81d2151a0219b31b | 11 | Since we compute IR spectra in the harmonic approximation, the geometry, Hessian, and dipole moments are all independent of isotopic composition. Thus, they are obtained only for the all-H isotopologue C 2 H 5 COO -. We then weight them by the nuclear masses of each of the 32 isotopologues to obtain its vibrational frequencies and IR intensities without requiring a repetition of the full calculation; this feature is available in the NEW branch of ASH. We additionally compute the infrared spectra of each isotopologue of propionate in the gas phase under the same computational workflow (except for QM/MM embedding). |
677c09ac81d2151a0219b31b | 12 | We obtain infrared spectra of each isotopologue of propionate, in the gas phase and from 25 snapshots in the MD trajectory, each 20 ns apart. We estimate the effect of the protein environment on the infrared spectra by averaging the wavenumber ν (in cm -1 ) of each vibrational peak and its infrared intensity (in arbitrary units) over the twenty-five spectral snapshots. Because the peaks are so far apart in time, we assume that their residuals are independent of one another, and model their thermal fluctuations as the standard deviation of the frequencies. We visualize these standard deviations as the width of each peak's Gaussian distribution. Time-averaged spectra of all 32 isotopologues are available in the Supplemental Material; the underlying data and scripts used to generate the spectra are available in the Duke Research Data Repository . |
677c09ac81d2151a0219b31b | 13 | Our key observation is that the high-frequency C -H stretches are far separated from the C -D counterparts in the receptor binding pocket environment and at physiological temperature. This is clearly visible when the all-hydrogen and all-deuterium isotopologues' computed IR spectra are overlaid (Fig. ). The C -H stretches lie between 3020 and 3120 cm -1 , while the C -D stretches range from 2200 to 2310 cm -1 . Each C -H mode is 800-840 cm -1 higher in energy than its corresponding C -D stretching mode, almost four times larger than kT = 215 cm -1 , the thermal energy available at 310 K. Note that the modes' fluctuations in the binding pocket are also much smaller than kT , roughly 10-20 cm -1 (Table ). The number of these stretching modes in intermediate isotopologues corresponds with their isotopic composition: C 2 H n D 5 -n COO - has n modes above 3000 cm -1 and 5 -n around 2200-2350 cm -1 . It is worth pointing out the two carbonyl/carboxyl (C --O) modes around 1400-1600 cm -1 , which are by far the most infrared-active peaks in the propionate spectrum. They are shifted to slightly lower frequencies in the olfactory receptor binding pocket relative to the gas phase. This peak differs slightly between the fully hydrogenated and fully deuterated isotopologues, with C 2 D 5 COO - having slightly lower frequencies (but higher intensities) than C 2 H 5 COO -. The differences are only about 10 cm -1 , however, on the order of the fluctuations each isotopologue sees over time (and substantially less than kT ). This is true even though the shorter C --O peak has substantial H/D elemental character, as Table : Details of the labeled peaks in Fig. . Frequency and intensity are given as mean ± standard deviation. C -D stretches are labeled C -H for economy of space. Elemental compositions are projections onto each atom type; they sum to 1, up to rounding error. can be seen in the last line of Table . It is therefore unlikely that the C --O mode plays a part in distinguishing the odors of propionate isotopologues, although it is possible that it has a role in detecting propionate in general. Some modes of lower wavenumber shift strongly upon isotopic subsitution, but the spectrum is fairly crowded below 1500 cm -1 and these low-energy vibrations are more variable over time than their high-energy counterparts; we do not expect them to allow olfactory discrimination between isotopologues. |
677c09ac81d2151a0219b31b | 14 | We have shown computationally that the difference between the high-frequency C -H and C -D modes of propionate substantially exceeds the fluctuations due to the olfactory receptor environment at 310 K. It is not known with certainty whether OR51E2 can distinguish between hydrogenated and deuterated propionate, although a null result for C 2 H 3 D 2 COOH was reported by . If they can be distinguished by vibration, however, we predict that the mechanism involves only the high-frequency modes. This idea is in harmony with the way propionate binds to OR51E2 in the experimental structure: the C -H and C -D stretches, localized on the alkane tail of the olfactant, are close enough to interact with the F155 and L158 residues of OR51E2 . It may be that strong noncovalent interactions with the carboxylate group lead to recognition of propionate in general, while weaker couplings to the alkyl vibrational modes allow distinguishing of isotopic composition. For a more complete understanding of olfactant recognition, both electronic and vibrational interactions should be modeled; for isotopic differentiation, our work suggests that simulating highfrequency modes be prioritized. |
64ee4a7479853bbd78b21dc7 | 0 | Advancements in chemistry and material science hinge on the availability of high-quality chemical reaction data, and the advent of machine learning (ML) for science has highlighted the value that data can bring to chemistry. One important application is in the pharmaceutical industry, where figuring out how to make novel molecules remains a significant bottleneck, causing delays in the "make" step of the "design, make, test" cycle . Making a molecule (product) includes predicting the reaction pathway (retrosynthesis) and suitable reaction conditions (e.g. solvents and reagents), and optimising for one or more outcomes such as reaction yield, selectivity, and conversion. ML is well suited to assist with these tasks, with a range of tools being developed for forward reaction prediction , retrosynthesis , condition prediction , yield prediction , and closed-loop optimisation . |
64ee4a7479853bbd78b21dc7 | 1 | Building reaction prediction tools requires access to large datasets for training. Historically, researchers have accessed proprietary in-house datasets or acquired the data through commercial databases such as Reaxys . The advantage of commercial databases is both the scale of the datasets available (often millions of reactions) and the annotation already completed by the publishers. Yet, these datasets are not freely available to ML practitioners, stymieing advances in reaction condition prediction in both academia and industry. |
64ee4a7479853bbd78b21dc7 | 2 | Recently, efforts have been made to create openly-accessible databases for chemical reaction data. In particular, the Open Reaction Database (ORD) is promising due to its exhaustive schema for describing chemical reaction data and breadth of data already incorporated. Yet, many of the datasets in ORD (license: Creative Commons Attribution Share Alike 4.0 International) require further processing before they can be used in ML pipelines, preventing practical use. This is especially true for the largest dataset in ORD extracted from the US patent literature (the "USPTO dataset" ). In this work, we endeavor to close this gap. |
64ee4a7479853bbd78b21dc7 | 3 | The remainder of the paper proceeds as follows. In section 2 we present a formulation of the key reaction prediction tasks considered in this work. This is followed by a brief review of related work in section 3. We then discuss the data extraction and cleaning methodology, and how ORDerly was used for dataset generation, in section 4. This is followed by experimental validation of these datasets with neural network and transformer architectures in section 5, demonstrating that missing key cleaning steps results in a dataset with contamination, which can inflate key performance metrics. We finally discuss the technical limitations of ORDerly in section 6 and then present our conclusions. |
64ee4a7479853bbd78b21dc7 | 4 | As noted by Meng et al. , reaction related tasks operate on molecules. There are numerous machine readable molecular representations , including molecular graphs and strings, and in this work molecules are represented as SMILES strings. Each character m i in a SMILES string represents an atom or a molecular feature (bond, branch, ring closure): M := m 1 , m 2 , m 3 , . . . , m L , where L is the total number of characters in the string. Molecules can take on one of three roles in a reaction: reactant, product, or agent. A reaction R transforms N reactant molecules (sometimes called educts) Figure : Overview of ORDerly. |
64ee4a7479853bbd78b21dc7 | 5 | Probabilistically, the task is to predict the distribution p(M P |{M E i } N i=1 ). While experimental evaluation in a wet lab requires expert chemists and is a time intense task, reaction outcome prediction can help as a tool to evaluate the quality of a predicted retrosynthetic route (i.e,. the probability that the reaction predicted by the single-step retrosynthesis model leads to the desired product) . |
64ee4a7479853bbd78b21dc7 | 6 | Existing tools for cleaning reaction data are primarily targeted at retrosynthesis and forward prediction tasks and have somewhat limited extensibility, given that they are built to take as inputs CSV files or the stationary XML files of the US patent (USPTO) dataset instead of the outputs of continuously updated databases such as ORD . Furthermore, in the original publications, there is little to no discussion of how decisions made during cleaning (e.g. restricting the number of components in a reaction or the minimum frequency of occurrence) impact the datasets being cleaned or performance of models trained on the datasets. We believe that this is in part due to data cleaning historically being viewed as a "low value" task, and therefore not adequately discussed and published on. USPTO, being the largest open-source chemical reaction dataset, has been cleaned a number of times for different learning tasks. For example, the USPTO-50K and USPTO-MIT datasets are commonly used for benchmarking single-step retrosynthesis and forward predictions models , and these benchmarks are available in aggregate benchmarking sets such as the Therapeutics Data Commons (TDC) . However, the code used to process the raw data to generate the aforementioned USPTO benchmarks was not published and, there is no publicly available benchmark for reaction condition prediction extracted from these datasets. |
64ee4a7479853bbd78b21dc7 | 7 | Forward prediction and single-step retrosynthesis models both need to predict how bonds might be broken and formed to produce new molecules. A common approach is to enumerate a set of templates for bond changes that happen in particular classes of reactions and use a classifier to predict the most likely template given a set of molecules . Alternatively, some models have been designed to explicitly predict bond changes . One promising approach is to directly predict the SMILES strings of the reactants (single-step retrosynthesis) or products (forward prediction) using a natural language processing model such as a transformer . In this work we use the transformer architecture of Schwaller et al. |
64ee4a7479853bbd78b21dc7 | 8 | Many reaction condition prediction models have focused on indirect prediction of conditions by learning to predict a measure of reaction performance such as yield . The downside of these models are that they are usually only scoped to a single class of reactions due to the lack of reliable yield measurements in benchmarks based on the USPTO dataset . However, Gao et al. built a model for reaction condition prediction agnostic of reaction class for sequential prediction of agents and temperature using approximately ten million reactions mined from a closed-source dataset, Reaxys . We train this model with minor modifications on our new open-source condition prediction benchmark. |
64ee4a7479853bbd78b21dc7 | 9 | ORDerly extracts data directly from ORD . Even though the data in ORD is stored in accordance with a structured schema, we found that further effort is required to transform the labeled data into ML-ready datasets. Therefore, ORDerly is centered around a data extraction script and a data cleaning script, both of which take numerous arguments that customize the operations being performed. |
64ee4a7479853bbd78b21dc7 | 10 | The extraction script allows the user to choose whether reaction roles should be assigned using the labeling in ORD or using chemically-informed logic on the atom-mapped reaction string (if available). It also enables specification of data source (e.g., USPTO or non-USPTO), allowing users to train models with data from one source and test the performance with data from another source. Creating test sets from different data sources is a robust way to evaluate generalization performance. |
64ee4a7479853bbd78b21dc7 | 11 | We chose cleaning operations motivated by first-principles understanding of chemistry. Cleaning operations on the chemical reaction data include: (1) Restricting the number of reactants and product, preventing multi-step reactions being included in the datset; Ensuring that all molecules can be sanitized by the cheminformatics package RDKit ; (3) Restricting the maximum number of unique catalysts, solvents, and reagents in a reaction based on commonly used experimental amounts; (4) Frequency filtering to remove outliers; (5) Sanity checking the yield (0% yield 100%), temperature, and pressure; (6) Removing duplicates, and finally; Applying a random split to create training/validation/test sets, carefully ensuring that any inputs present in the train set (i.e. reactants and products for reaction condition prediction) are not also present in the test set. |
64ee4a7479853bbd78b21dc7 | 12 | Computational details: All extraction/cleaning operations described in this section were performed using a 2022 Mac Studio with an Apple M1 Max chip and 32GB memory. In ORD there are roughly 1.7 million reactions from US patents (USPTO) and 91k reactions that are not from US patents. During handling of the USPTO data in ORD we found that extracting and sanitizing the reaction components using the ORD labeling of components was slightly faster than using our custom logic applied to the reaction string, taking 28 minutes and 48 minutes, respectively. The cleaning steps took 6-8 minutes. Due to the amount of non-patent data being much less, extraction and cleaning of non-USPTO data took only a few minutes. |
64ee4a7479853bbd78b21dc7 | 13 | We experimented with two approaches to assigning roles to the molecules found in a reaction (e.g., whether a molecule is a reactant or an agent): trusting the labeling of molecules in ORD (referred to as "labeling") or applying chemical reaction logic to identify the role of different molecules from the reaction string (referred to as "rxn string" or "reaction string"). Our reaction logic identified reactants (molecules that contribute atoms to the product(s)) and spectator molecules (molecules that do not contribute atoms to the product(s)) based on the atom-mapping and their position in the reaction SMILES string. Solvents were identified in the list of spectator molecules by cross checking Figure : We present two different approaches for handling rare molecules. Rare ! "other" is investigated as a strategy to avoid deleting reactions with rare molecules. against a list of solvents compiled from prior research (see Appendix B.1.1), while all other spectator molecules are marked as agents. |
64ee4a7479853bbd78b21dc7 | 14 | Removing rare molecules can increase the signal to noise ratio in a dataset by removing outliers. In this work, we investigated two different strategies for filtering spectator molecules based on their frequency: deleting reactions with rare spectator molecules (rare!delete rxn) or keeping the reactions but mapping the rare molecules to an "other" category (rare!"other") (see Figure ). We conducted experiments with both the rare!delete rxn and rare!"other" strategies for the task of condition prediction. The frequency threshold was set at 100 in line with previous research , though the sensitivity of dataset size to frequency threshold was still investigated (see Appendix D.2). Deleting reactions with rare molecules may create a more cohesive dataset by removing outliers, while renaming rare molecules "other" allows more reactions to be kept, offering more training data for the model. |
64ee4a7479853bbd78b21dc7 | 15 | We used ORDerly to create datasets for three tasks: forward, retrosynthesis, and condition prediction. Several different datasets were created for each task, and the impact of each cleaning step on the dataset size can be found in Table . The datasets are freely available and can be downloaded immediately from FigShare or regenerated using the code in the ORDerly Github repository. |
Subsets and Splits