|Home | About | Journals | Submit | Contact Us | Français|
Bio-oil, produced by the destructive distillation of cheap and renewable lignocellulosic biomass, contains high energy density oligomers in the water-insoluble fraction that can be utilized for diesel and valuable fine chemicals productions. Here, we show an efficient hydrodeoxygenation catalyst that combines highly dispersed palladium and ultrafine molybdenum phosphate nanoparticles on silica. Using phenol as a model substrate this catalyst is 100% effective and 97.5% selective for hydrodeoxygenation to cyclohexane under mild conditions in a batch reaction; this catalyst also demonstrates regeneration ability in long-term continuous flow tests. Detailed investigations into the nature of the catalyst show that it combines hydrogenation activity of Pd and high density of both Brønsted and Lewis acid sites; we believe these are key features for efficient catalytic hydrodeoxygenation behavior. Using a wood and bark-derived feedstock, this catalyst performs hydrodeoxygenation of lignin, cellulose, and hemicellulose-derived oligomers into liquid alkanes with high efficiency and yield.
Decreasing petroleum deposits combined with conversely climbing global demand and environmental concern makes it critical to develop fuel and chemical production processes based on renewable resources1. Lignocellulosic biomass, including cellulose, hemicellulose, and lignin, is the most abundantly available and cost-effective carbon reservoir that can be used to produce renewable fuel and high added-value chemicals2. Compared with biomass fractionation, which requires extensive pre-treatment and following hydrolysis, flash pyrolysis is a cheaper process for the production of upgradable bio-oil from lignocellulose, and offers potential for lignin utilization3, 4. However, the bio-oil also contains high oxygen content (up to 60wt%) and necessitates further catalytic upgrading, preferably via hydrodeoxygenation (HDO), before use in hydrocarbon fuel or chemical production5. The water soluble fraction of bio-oil mainly involves carbohydrates that could be catalytically converted into hydroxymethylfurfural6, 7, levulinic acid8, and monofunctional hydrocarbons9, 10; these platform chemicals can be further converted to hydrocarbons via HDO. By comparison, the water-insoluble bio-oil (WIBO) often contains a large variety of products derived from lignin, cellulose, and hemicellulose in the form of oligomers with molecular weight up to 5000Da5. The complexity of WIBO composition and the chemical inertness of the oligomers present therefore pose a great challenge for any HDO catalyst.
In recent years, bifunctional metal-acid catalysts have been developed, showing high efficiency in HDO reaction for hydrocarbon fuel production from lignin11–17. Lercher and co-workers have made important contributions on the design of an efficient combination of metal-catalyzed hydrogenation and acid-catalyzed deoxygenation for highly selective HDO of phenols to cyclohexanes11, 15. Both the Dyson and Leitner groups have made impressive progress on the development of high performance catalyst for the HDO of lignin-derived phenols using combinations of metal nanoparticles and acidic ionic liquids8, 17. In spite of the growing number of metal-acid combinations, the successful productions of alkanes using these bifunctional metal-acid catalysts have been mostly used for single-component lignin-derived monomers. The conversion of lignin into alkanes has been realized in a rare case via a two-step process, involving catalytic cleavage of the C–O–C bonds and subsequent hydrogenolysis/hydrogenation18. Very recently, the production of alkanes, with mass yield up to 28.1wt%, was reported via HDO from raw lignocellulose containing cellulose, hemicellulose, and lignin by using a Pd/NbOPO4 bifunctional catalyst. However, the mass yield of alkyl cyclohexanes converted from lignin was only up to 5.1wt%19. Efficient HDO requires the design and synthesis of a bifunctional catalyst containing active metal and both Brønsted and Lewis acid sites. Recent reports showed that atomically dispersed metal may demonstrate higher activity and efficiency compared with nanoparticles in heterogeneous reactions20–22. It is also believed increased Brønsted acid sites favor increasing alkane selectivities16, 23, and additional Lewis acid sites may help catalyze the reaction through initial binding of the oxygenated substrates and subsequent cleavage of the C–O linkages19.
Herein, we present a highly dispersed Pd–Mo catalyst that encompasses a combination of highly dispersed Pd and ultrafine Mo phosphate nanoparticles on silica (Pd/m–MoO3–P2O5/SiO2), in which m represents mixed-valent Mo. In batch reaction, this catalyst exhibits high activity (99%) for the catalytic HDO of phenol to cyclohexane in very mild conditions (383K, 1MPa). Furthermore, it shows good re-usability in continuous flow reaction conditions for 419h. Characterization studies show that the efficient HDO ability may originate from the cooperative effects between highly dispersed Pd, high Brønsted and Lewis acid sites. By feeding wood and bark-derived WIBO containing large amounts of lignocellulosic oligomers, the catalyst not only completely converts phenolic monomers to cyclohexanes, but also achieves HDO of lignin, cellulose, and hemicellulose-derived oligomers into cyclohexanes, hexane, and pentane in 13.4, 5.1, and 7.4wt%, respectively, with total mass yield up to 29.6wt% and carbon yield up to 46.3%.
The solid catalyst, Pd/m–MoO3–P2O5/SiO2, was prepared using a wet-impregnation method (see “Methods” for preparation). Transmission electron microscopy (TEM) image of the Pd/m–MoO3–P2O5/SiO2 (Supplementary Fig. 1) shows ultrafine, electron-rich m–MoO3–P2O5 nanoparticles supported on an amorphous SiO2 support. A high-angle annular dark-field scanning transmission electron microscopy (HAADF-STEM) image (Fig. 1a) shows these ultrafine m–MoO3–P2O5 nanoparticles with high coverage and size of 2.8±1.0nm (inset, Fig. 1a). Inductively coupled plasma mass spectrometry (ICP-AES) analysis showed a Mo:P molar ratio of ca. 1:1 (Supplementary Table 1), which is consistent with Energy Dispersive X-ray spectroscopy (EDXS) data (Supplementary Fig. 2). Aberration-corrected annular bright-field (ABF) scanning transmission electron microscope image (Fig. 1b) and corresponding HAADF-STEM image (Fig. 1c) clearly show that the m–MoO3–P2O5 nanoparticles do not form a crystalline lattice. Together with powder X-ray diffractogram (PXRD) data (Supplementary Fig. 3), these results suggest that the m–MoO3–P2O5 nanoparticles are crystallographically disordered. Elemental mapping revealed that Mo and P are well dispersed throughout individual m–MoO3–P2O5 nanoparticles. The structure of the m–MoO3–P2O5 nanoparticles was further investigated with the help of multinuclear solid-state nuclear magnetic resonance (NMR) spectroscopy. The 31P solid-state magic angle spinning (MAS) NMR spectrum of a sample of the Pd/m–MoO3–P2O5/SiO2 exhibits a single resonance centered at δ=–5.5ppm (Supplementary Fig. 4), the 31P chemical shift matches that of a MoO3–P2O5 glass with Mo:P molar ratio ca. 124, 25. A strong absorption band at ca. 966cm−1 in the Raman spectrum (Supplementary Fig. 5) can be assigned to a terminal ν(Mo=O) in a MoO6 octahedra25. The data suggest that this glassy structure consists of chains of inter-connected PO4 tetrahedra and MoO6 octahedra linked via Mo–O–P bridges.
In addition to Mo and P elements, ICP-AES (Supplementary Table 1) and EDXS (Supplementary Fig. 2) data also show the presence of Pd in Pd/m–MoO3–P2O5/SiO2 (0.48wt% from ICP-AES). The Pd signal is highly dispersed according to elemental mapping studies (Fig. 1d), and that the absence of Bragg diffraction of Pd from PXRD (Supplementary Fig. 3), these suggest Pd is presented in very small size. Due to the similar atomic number of Pd and Mo, it is difficult to distinguish Pd using the intensity information in the HAADF-STEM image. To understand the coordination environment and chemical states of Pd in Pd/m–MoO3–P2O5/SiO2, X-ray absorption near-edge structure (XANES) and extended X-ray absorption fine structure (EXAFS) spectroscopies were carried out. For clear comparison, Pd/m–MoO3–P2O5/SiO2 is labeled as reduced catalyst, and corresponding H2 treating precursor is labeled as oxidized catalyst. The Mo K-edge XANES spectrum for reduced catalyst shows a notable shoulder peak (denoted as a) in the pre-edge region (Fig. 2a) as MoO3, suggesting the formation of strongly distorted MoO6 octahedra having Mo=O bond26, in good agreement with the Raman results. The average valence of the Mo ions in the reduced catalyst is estimated to be ca. 5.5 using the linear relation between the Mo oxidation state and the energy position of feature b27, indicating the possible existence of mixed-valent Mo5/6+ oxide states. The Pd K-edge XANES spectrum is found to exhibit two well-defined features including a shoulder peak c and a doublet peak (d and e) at the white line (Fig. 2d); our XANES simulation reveals that the shoulder peak c can be considered to be the fingerprint of Pd2+ in square planar coordination, while the intensity of feature e is higher for Pd ions in distorted octahedral coordination than for the square planar coordination (Fig. 2g), consistent with XANES results in the literature28. Therefore, this implies that the Pd is highly dispersed in the framework of m–MoO3–P2O5/SiO2 via a combination of square planar and distorted octahedral configurations, with an averaged oxidation state between Pd2+ and Pd3+. This is in agreement with recently reported strategy of stabilizing single Pt atoms on a support interface by adopting a proper coordination22. The presence of m–MoO3–P2O5/SiO2 matrix may offer an ideal interface to stabilize Pd with high dispersion.
The structural change of Pd/m–MoO3–P2O5/SiO2 during H2 treatment was also investigated. Interestingly, except for the peak intensity decrease for the Pd–O nearest-neighbor (NN) coordination shells, the two higher shell peaks at distances of 2.5 and 3.5Å. The Pd K-edge EXAFS Fourier transform of reduced catalyst (Pd/m–MoO3–P2O5/SiO2) display a dramatic intensity increase and a radial shift to lower R direction, as compared to oxidized catalyst (Fig. 2e); similar phenomenon is shown for the Mo K-edge EXAFS results (Fig. 2b). Meanwhile, complementary wavelet transform (WT) EXAFS analysis29 reveals strong WT maxima at about 7.0 and 9.5Å–1, which can be, respectively, associated with the Pd–P/Si and Pd–Pd/Mo scatterings (see Supplementary Fig. 6), at a distance of around 2.55Å surrounding Pd atoms in reduced catalyst, in marked contrast to the WT maxima distribution in oxidized catalyst (Fig. 2f); slight change is observed for the Mo K-edge WT (Fig. 2c). Those results suggest that more oxygen vacancies are produced by H2 reduction in the oxide network of reduced catalyst, which may cause the collapse of the PdO6 octahedra by breaking the bridging oxygen atoms with the neighbor polyhedrons. A least-squares curve-fitting analysis was carried out for the multiple coordination shells of Mo and Pd (Supplementary Figs. 7–10 and Supplementary Tables 2 and 3). The coordination numbers of the Pd–O bonding in the first NN coordination sphere for reduced catalyst (oxidized catalyst) is estimated to be 4.9 (5.6) at a distance of 2.02 (2.03)Å, evidencing the transformation of the Pd–O bonding from a distorted octahedral to a square planer coordination. Simultaneously, the number of Pd/Mo atoms in the second NN coordination sphere of Pd in the reduced catalyst is estimated to be 1.5 at a distance of 2.80Å, much larger than that of 0.4 in the oxidized catalyst, strongly signaling the presence of Pd–Pd/Mo atoms. By EXAFS analysis it is difficult to distinguish Mo from Pd; the data suggest a probability of both Pd–Mo and Pd–Pd neighbors after reduction, as schematically shown in Fig. 2h using density functional theory (DFT) calculations. The data do allow us to conclude there is an absence of large Pd particles, implying a high dispersion for the Pd atoms (further discussions are shown in Supplementary Note 1 and Supplementary Figs. 11–13). Thus, XAFS data indicate that there are significant oxygen vacancy sites present in close vicinity of Mo and highly dispersed Pd in Pd/m–MoO3–P2O5/SiO2.
Phenol HDO in a batch reactor was initially chosen as a model reaction to evaluate the catalytic activity of Pd/m–MoO3–P2O5/SiO2. The total mass of Pd was kept constant in order to benchmark other catalyst systems (Supplementary Table 1). The reaction process using Pd/m–MoO3–P2O5/SiO2 catalyst (Fig. 3a and Supplementary Fig. 14) shows an increase in cyclohexane selectivity and a decrease in cyclohexanol selectivity. In comparison, Pd/SiO2 catalyst only delivered hydrogenated products (cyclohexanone and cyclohexanol) and subsequent dehydration was not observed (Supplementary Fig. 14). These results suggest that in the case of Pd/m–MoO3–P2O5/SiO2 catalyst, phenol is initially hydrogenated to cyclohexanol catalyzed by Pd, which is followed by dehydration of cyclohexanol catalyzed by m–MoO3–P2O5 and further hydrogenation of cyclohexene to yield cyclohexane catalyzed by Pd. Via a comparison of product selectivity at near identical conversion level (Fig. 3b), the Pd/m–MoO3–P2O5/SiO2 catalyst exhibited a much higher cyclohexane selectivity (81%) compared with catalysts without m–MoO3–P2O5, indicating its high dehydration ability is due to the presence of m–MoO3–P2O5. We have compared the performance of Pd/m–MoO3–P2O5/SiO2 for the HDO of phenol to a series of state-of-art bifunctional metal-acid catalysts (Supplementary Table 4). Overall Pd/m–MoO3–P2O5/SiO2 demonstrates higher phenol conversions, with better cyclohexane selectivity at a lower temperature and lower H2 pressure than any of other catalyst systems. The catalyst was re-used and showed no obvious deactivation after five cycles (Fig. 3c), showing its stability under batch reaction conditions. In order to investigate stability under more industrially related conditions, Pd/m–MoO3–P2O5/SiO2 was evaluated under continuous flow condition at 433K, 1MPa H2 with flow rate of 10cm3 (STP)min–1, weight hour space velocity (WHSV)=0.18h–1. The results show that there is some degree of deactivation (phenol conversion from 84.2 to 77.2%) but stable cyclohexane selectivity (from 68.2 to 68.8%) in the 32-h test (Fig. 3d), exhibiting higher cyclohexanol selectivity and stability than those in recent report under similar conditions (cyclohexane yield decrease from 52 to 28% in 4h)17. The spent catalyst shows no obvious change (Supplementary Figs. 15, 16). The Mo and Pd K-edge XANES and EXAFS spectra are also given for the spent catalyst, exhibiting no obvious changes in the local structure around Mo and Pd atoms (Supplementary Fig. 17). The deactivation was mainly due to carbonaceous deposition (Supplementary Fig. 18). To evaluate re-use ability, Pd/m–MoO3–P2O5/SiO2 was evaluated under continuous flow conditions (453K, 1MPa H2 with WHSV of 0.085h−1). As shown in Supplementary Fig. 19, after it was deactivated (phenol conversion from >99 to 69.0%, cyclohexanol selectivity from 98.3 to 90.4%) in 135h, the spent catalyst was calcined at 673K for 5h to remove the carbonaceous deposition and re-used for another two runs. The result shows that the initial performance could be substantially recovered, with phenol conversion of >99% (1st run), 94.5% (2nd run), and 94.0% (3rd run), respectively, and cyclohexane selectivity of 98.3% (1st run), 98.4% (2nd run), and 96.7% (3rd run), respectively, and a similar deactivation trend was observed. This good re-usability ensures a long-term reaction efficiency (maintaining phenol conversion >60% and cyclohexane selectivity >90%) in three runs for over 419h without obvious catalyst leaching (Supplementary Table 1).
31P MAS NMR spectrum were recorded following adsorption of the trimethylphosphine oxide (TMPO) probe molecule, giving information about the acid type (Brønsted vs. Lewis), strength and concentration in the materials30. The 31P MAS NMR spectrum (Fig. 4a) of the TMPO-adsorbed on reduced catalyst (Pd/m–MoO3–P2O5/SiO2) exhibits two distinct resonances at 58.0 and 73ppm, both signals are attributed to the protonation of TMPO by surface Brønsted acid sites with exceptionally high density reaching 0.0138mmol/m2 (Supplementary Table 5 and Supplementary Fig. 20), which is 2.6-fold higher than a strong acidic zeolite HZSM-5 (Si to Al of 15)31. The reduced catalyst exhibits another new resonance at 45ppm, which was not observed in oxidized catalyst and could be attributed to Lewis acid with density of 0.0075mmol/m2. A series of catalysts with different Brønsted acid sites and Lewis acid sites were prepared by the same procedure as the Pd/m–MoO3–P2O5/SiO2 catalyst by varying the molar ratio of P:Mo (Supplementary Table 5). The existence of the Lewis acid site is observed only on catalysts containing Mo species, thus it is most likely linked to oxygen vacancy associated with Mo centers32, which is well consistent with abovementioned XAFS analysis. The formation of oxygen vacancy is observed by temperature programmed reduction (TPR, Supplementary Fig. 21), and could be linked with the reduction of surface oxidation state evidenced by multi-frequency electron paramagnetic resonance (EPR, see Methods, Supplementary Figs. 22–24) and by solid-state magnetic susceptibility measurements (Supplementary Fig. 25).
We believe that the excellent performance of Pd/m–MoO3–P2O5/SiO2 for HDO may be attributed to cooperative effects between highly dispersed Pd, strong Brønsted acid sites, and Lewis acid sites. The highly dispersed Pd allows for higher hydrogenation activity (Fig. 3a) and enhanced metal atom efficiency compared with Pd/SiO2 (Supplementary Fig. 14), by increasing the number of exposed surface atoms by using very low Pd loading amount of 0.48wt%. The catalysts with different Brønsted acid sites (Supplementary Table 5) were compared in the dehydration of cyclohexanol (Supplementary Table 6), which is the most important intermediate in the HDO of phenol. Since most of the samples show comparable Brønsted acid strength, respectively, revealed by resonance position of 31P MAS NMR spectrum (except Pd/P2O5/SiO2 (P:Mo=1:0) which shows weaker Brønsted acid strength), one can look into the correlation between Brønsted acid and dehydration activity. The result shows that only samples containing Brønsted acid sites have activity, suggesting the dehydration is a Brønsted acid-catalyzed process. Specifically, Pd/m–MoO3–P2O5/SiO2 (P:Mo=1:1, the reduced catalyst shown in the main work) with the most Brønsted acid sites is a more active dehydration catalyst, and also Pd/m–MoO3–P2O5/SiO2 (P:Mo=2:1) with higher Brønsted acid sites compared with Pd/MoO3/SiO2 (P:Mo=0:1) has higher dehydration activity (conversion: 25.2 vs. 19.8%), which are consistent with the correlation between the Brønsted acid sites and the dehydration activity16, 23. It is also observed that Pd/P2O5/SiO2 has lower dehydration activity compared with Pd/MoO3/SiO2 (12.8 vs. 19.8%), although its Brønsted acid sites is higher. This could be due to its relatively weaker Brønsted acid strength13. The role of Lewis acid site on phenol conversion was further studied by performing diffuse-reflectance infrared Fourier transform spectroscopy (DRIFTS) upon exposure of the oxidized and reduced catalyst (Pd/m–MoO3–P2O5/SiO2) to a phenol/He mixture at reaction temperature (383K). In the measurements, the reduced catalyst was first treated with 5% H2/Ar at 383K before exposed to phenol/He mixture. The DRIFTS spectra of both reduced catalyst and oxidized catalyst (Fig. 4b) show bands at 1597, 1497, 1475, 1352 and 1267cm–1, which correspond to different vibrational modes of phenoxy species33. Compared with oxidized catalyst, the reduced catalyst shows more pronounced bands at 1684, 2951 and 2874cm–1 (Fig. 4b, Supplementary Fig. 26), which could be assigned to the v(C=O) and v(C–H) stretching modes, respectively, of adsorbed 2,4-cyclohexadien-1-one tautomer of phenol33, 34. Considering that both the DRIFTS of spectra of oxidized Pd/SiO2 and reduced Pd/SiO2 do not show obvious bands for 2,4-cyclohexadien-1-one (Supplementary Fig. 27), the formation of this compound is more likely promoted by Lewis acid sites (oxygen vacancy associated with Mo centers) on reduced catalyst via tautomerization of phenol. This compound is an important intermediate in the HDO of phenol supported by Priscillat et al. studies33, which could be first hydrogenated to 2-cyclohexen-1-one, then to cyclohexanone and further to cyclohexanol.
Based on the above understanding, efficiency test using Pd/m–MoO3–P2O5/SiO2 for upgrading a WIBO derived from fast pyrolysis of grounded eucalyptus wood chip and bark were carried out. This WIBO sample was a dark red, treacle-like, viscous liquid (Supplementary Fig. 28) with a 47.2wt% carbon content and a 42.8wt% oxygen content (Supplementary Table 7). The WIBO contains 13wt% moisture determined by Karl Fisher method, which mainly derived from the feedstock and also the dehydration reactions during flash pyrolysis treatment35. The content of the WIBO was studied using gas chromatograph-mass spectrometry (GC-MS) (Fig. 5a and Supplementary Fig. 29). The results show the WIBO contains a series of oxygenated monomers involving phenolic compounds such as guaiacol, syringol, and furans (4.7wt% of total WIBO mass). High-performance liquid chromatography (HPLC) of the water extract of the WIBO (Supplementary Fig. 30) showed no sugar-type monomers before it was hydrolyzed. The remaining mass can be attributed to lignin and sugar oligomers that are undetectable in both GC-MS and HPLC. Gel permeation chromatography (GPC) analysis (Supplementary Fig. 31) verified the presence of oligomers in the WIBO with molecular weight up to 1000Da and weight average molecular weight (M w) of 401. This M w value is consistent with those reported for wood pyrolysis oils36.
HDO of this WIBO was carried out in a batch reactor at 453K and 1MPa H2 in the presence of Pd/m–MoO3–P2O5/SiO2 catalyst. Total conversion of the oxygenated monomers was achieved after 4h (Fig. 5b), with the production of 9.4wt% total mass yield of liquid alkanes involving 0.9wt% pentane, 1.2wt% hexane, and 5.6wt% C6–C9 cycloalkanes (cyclohexane, methylcyclohexane, ethylcyclohexane, and propylcyclohexane; see Table 1 and Supplementary Fig. 32). Overall a 10.2% carbon yield in the form of cyclohexanes was obtained. This is higher than the maximum carbon yield (5.7%) possible from the lignin-derived monomers present in the WIBO. This indicates that some oligomers in the WIBO were converted into cyclohexanes. The production of pentane and hexane suggests that HDO of cellulose and hemicellulose into alkanes are achieved, which is supported by the earlier literature19. The color of the mixture became lighter after the reaction, serving as another qualitative indicator of the HDO efficiency (inset in Fig. 5a, b). To further increase the mass yield of liquid alkanes, the reaction was carried out at an increased temperature. At 523K and 1MPa H2, 29.6wt% total mass yield and 46.3% total carbon yield of liquid alkanes was achieved after 15h, the mass yield of cyclohexanes, pentane, and hexane reached 13.4, 7.4 and 5.1wt%, respectively (see Methods, Table 1, Supplementary Fig. 33). It should be noted that the total carbon yield is comparable to the state-of-art in catalytic performance19, 37, 38. GPC analysis (Supplementary Fig. 34) shows the decrease of oligomers. Although reaction mechanisms for HDO of different classes of oxygenates in bio-oil could be different, these results clearly indicate that the catalyst has a high performance for simultaneous HDO of lignin, cellulose, and hemicellulose-derived oligomers into liquid alkanes.
We report a highly active HDO catalyst comprising highly dispersed Pd and ultrafine Mo phosphate nanoparticles supported on SiO2. This catalyst is 100% effective and is 97.5% selective for HDO of phenol to cyclohexane under mild conditions, showing regeneration ability in long-term continuous flow tests for 419h with some decrease in catalytic performance after 32h in a continuous flow reaction. There are cooperative effects between highly dispersed Pd, Brønsted and Lewis acid sites on Pd/m–MoO3–P2O5/SiO2, which shows to be important for the efficient catalytic HDO behavior. Using water-insoluble fraction of wood and bark-derived WIBO, the catalyst showed state-of-art performance for the simultaneous HDO of lignin, cellulose, and hemicellulose-derived oligomers into liquid alkanes with total mass yield of 29.6wt% under mild condition (523K, 1MPa H2). Further work will include mechanistic investigations for HDOs of different classes of oxygenates in bio-oil to fully understand the high efficiency of this catalyst.
Pd/m–MoO3–P2O5/SiO2 synthesis: (NH4)6Mo7O24•4H2O (0.021mmol), (NH4)2HPO4 (0.150mmol), citric acid (0.30mmol) were dissolved in 0.3mL of HCl aqueous solution containing PdCl2 (1.5mg). The mixture was stirred for 1h and then added dropwise to SiO2 (0.12g). The obtained glue-like sample was dried in an oven at 397K overnight and calcined at 773K in a muffle furnace for 5h at a rising rate of 5K/min. The calcined sample, named as Pd/MoO3–P2O5/SiO2, was reduced in H2 at 323K for 3h with a flowing rate of 75cm3/min. The calcined sample changed to a greenish-blue color following reduction. This reduced sample was named Pd/m–MoO3–P2O5/SiO2. The synthesis of Pd/P2O5/SiO2 was similar as for Pd/m–MoO3–P2O5/SiO2, except with no addition of (NH4)6Mo7O24•4H2O. Pd/SiO2 was prepared by dropwise addition of 0.3mL of prepared PdCl2 solution into SiO2 under strong stirring. The following operations are the same as the one for Pd/m–MoO3–P2O5/SiO2. The fresh catalyst was protected in N2 and quickly prepared as samples for further characterizations and catalytic tests.
HRTEM images were acquired with JEOL JEM-2100F field-emission TEM with an accelerating voltage of 200kV.
Aberration-corrected TEM was performed on JEOL 200F TEM operated at 200keV. The attainable spatial resolution of the microscope is 78pm with a probe spherical-aberration corrector. Both ABF and HAADF were obtained with the illumination semi-angle of 25mrad and probe current of 100pA. The dwell time for imaging was set at 10ms per pixel to ensure desirable signal to noise ratio. The collection angles for the ABF and HAADF images were fixed at 12–25 and 90–250mrad, respectively. EDXS was performed to locate elemental distribution of Mo, P, and Pd with an SDD-type EDX detector. The attainable energy-resolution of the EDX detector is 130eV. A higher beam current of 300pA was used with a longer dwell time of 0.1ms per pixel repeated for 200 times. Spatial drift was corrected with a simultaneous image collector. Mo L-edge, P K-edge and Pd L-edge were used for elemental mapping.
PXRD data were obtained on a PANAnalytical X’Pert Pro diffractometer in reflection mode at 40kV and 40mA using Cu Kα radiation.
Raman spectra were recorded on a Renishaw Raman spectrometer at a laser excitation wavelength of 633nm.
Mo K-edge and Pd K-edge X-ray absorption spectra were acquired under ambient condition at beamline BL14W1 of Shanghai Synchrotron Radiation Facility (SSRF) using a Si(311) double-crystal monochromator in transmission and fluorescence modes, respectively. The storage ring of SSRF was operated at 3.5GeV with a maximum current of 250mA. The energy was calibrated using Mo/Pd foil. While the incident and transmitted X-ray intensities were monitored by using standard ion chambers, the fluorescence signal was detected by using a 7 element Ge detector. To prevent air oxidation, the samples were prepared in a glove box, and they were uniformly mixed with BN powder and pressed to a pellet, which was sealed in a cell holder with Kapton windows for XAFS measurement. The XAFS raw data were background-subtracted, normalized, and Fourier transformed by the standard procedures with the ATHENA program39, 40. Least-squares curve-fitting analysis of the EXAFS χ(k) data was carried out using the ARTEMIS program39 with the theoretical scattering amplitudes, phase shifts, and the photoelectron mean-free path for all paths calculated by ab-initio code FEFF9.0541. The details of curve fitting are discussed in Supplementary Figs. 7–10 and Supplementary Tables 2 and 3. The Pd K-edge theoretical XANES calculations were carried out with the FDMNES code in the framework of real-space full multiple-scattering scheme using Muffin-tin approximation for the potential41–43. The energy-dependent exchange-correlation potential was calculated in the real Hedin-Lundqvist scheme, and then the spectra are convoluted using a Lorentzian function with an energy-dependent width to account for the broadening due both to the core-hole width and to the final state width. Satisfactory convergence for cluster sizes had been achieved.
The electronic structure calculations were performed using the density functional theory and a plane-wave basis set as implemented in the VASP code44. The electron–ion interaction was treated by the projector-augmented wave method45 with the outmost s, p, and d states as valence orbital. The electron exchange-correlation energy was described by the modified Perdew-Burke-Ernzerhof generalized gradient approximation for solids (PBEsol)46. The kinetic energy cutoff was set to 350eV, and the Brillouin zone was sampled in Monkhorst-Pack k point meshes with an interpolation grid spacing of 0.04Å–1 for the 240 atoms supercell, to achieve the total energy convergence of less than 0.001eV/atom. Structural relaxation was allowed until the force on each atom was <0.01eV/Å.
Phenol (18.4mg, 0.195mmol) was added into a stainless steel Parr autoclave (reactor volume, 50mL) with decalin (7mL), together with Pd/m–MoO3–P2O5/SiO2 (10.0mg) and was stirred for 10min. The autoclave was then sealed, purged with H2 for three times, and then placed under 1MPa H2 at room temperature. The reaction was carried out at different temperature for a certain time with a stirring speed of 800rpm. After reaction, the organic product was collected and analyzed by GC-MS. For the re-usability experiments using the Pd/m–MoO3–P2O5/SiO2 catalyst; after each reaction cycle, the spent catalyst was separated by centrifugation and used for the next run.
The fixed bed reaction was carried out in a HEL made continuous trickle bed reactor (mode FlowCAT) with both the liquid feed and hydrogen gas passing in downward direction. The Pd/m–MoO3–P2O5/SiO2 catalyst (127mg) was located in the middle of the tubular reactor with quartz wool plugs on both the sides. Liquid feed was prepared by dissolving phenol in decalin to form a solution of 3.6mg/mL. The reaction was carried out at 433–453K (Fig. 3d and Supplementary Fig. 19), 1Mpa with H2 flow rate of 10cm3(STP)min–1 and liquid flow rate of 0.05–0.1mLmin–1 (Fig. 3d and Supplementary Fig. 16). The liquid was preheated at the desired reaction temperature before being fed into the reactor. The products were periodically collected from the outlet stream throughout the reaction and were analyzed by GC-MS. WHSV was calculated by dividing the feed flow rate per hour by weight of catalyst. The spent catalyst (32h-test) was collected and protected in N2, and then quickly prepared as samples for further characterizations. For the re-usability of the Pd/m–MoO3–P2O5/SiO2 catalyst, the spent catalyst was not removed from the reactor but calcined on-site in air at 673K for 5h with a flowing rate of 20cm3/min and then reduced in H2 at 323K for 3h with a flowing rate of 75cm3/min.
GC-MS analysis was conducted by using an Agilent gas chromatograph equipped with an Agilent 19091N-133 column of mode HP-INNOWax with high polarity, 30m×250μm×0.25μm connected online to mass spectrometer. The GC oven was programmed as: hold at initial temperature of 313K for 5min, ramp at 15Kmin–1 to 523K and hold at 523K for 5min. 2-isopropylphenol was used as the internal standard. The peaks were analyzed by comparing the corresponding spectra with those of the NIST 2011 MS library.
The WIBO was derived from fast pyrolysis of grounded eucalyptus wood chip and bark. Wood log (wood chip+bark) was fed directly into pyrolysis reactor in forms of grounded wood and bark. Pyrolysis was carried out at 773K in absence of oxygen. The produced bio-oil was separated into water-soluble phase and water-insoluble phase. For the WIBO part, the volatile species, mainly lignin-derived monomers, were determined by dissolving 84.0mg WIBO and 4.03mg internal standard, 2-isopropylphenol, in 1.5mL methanol or only WIBO in decalin, and analyzing by GC-MS, as presented in Supplementary Fig. 29 and Fig. 5a. For the analysis of sugar component, 1.038g WIBO was washed with 30mL water while stirring at room temperature for 12h, and the resulting water extraction was analyzed by HPLC, as presented in Supplementary Fig. 30. HPLC analysis was conducted by using a Laserchrom HPLC system equipped with a Aminex HPX 87H 300×7.8 column and Refractive Index Detector at 308K using 0.025M H2SO4 as the mobile phase with flow rate of 0.6mLmin–1. To hydrolyze oligomers into monomeric sugars, 200μL concentrated H2SO4 was added into 2mL upper water solution and heated at 393K for 1h. The resultant was also analyzed by HPLC, as presented in Supplementary Fig. 30.
For reaction at 453K, WIBO (86.2mg), decalin (7mL), and Pd/m–MoO3–P2O5/SiO2 (170.8mg) were added into a stainless steel Parr autoclave (reactor volume of 50mL) and stirred for 10min. Then, the autoclave was sealed, purged three times with H2, and placed under 1MPa H2 at room temperature. The reaction was carried out at 453K for 4h with a stirring speed of 800rpm. After reaction, the organic product was collected and analyzed by GC-MS using an added internal standard, as presented in Supplementary Fig. 32. For reaction at 523K, 87.2mg WIBO was used and the other parameters were kept identically. The reaction was carried out at 523K for 15h. The reaction products were analyzed by GC-MS, as presented in Supplementary Fig. 33.
The mass of lignin-derived phenols and furans in the WIBO and alkanes in the product were determined by internal standard. The corresponding response factors were determined by analyzing mixtures of those pure, commercial compounds of phenols, furans and alkanes, and internal standard with given weights. Mass yields of alkanes were calculated by the equation: mass of alkanes/mass of dried WIBO input. Carbon yields were calculated by the equation: mass of carbon in alkanes/mass of carbon in the WIBO input.
GPC measurements were performed on a Shimadzu LC-20AD instrument using MALLS detector. Two Mixed Bed PSS SDV linear S columns were used in series with tetrahydrofuran (THF) as mobile phase with flowing rate of 1.0cm3min–1 at 303K. The MALLS detector was calibrated using a polystyrene standard. The sample was prepared by dissolving WIBO in THF with 1wt% concentration and then filtering with 0.20μm pore size microfilter.
TGA measurements were performed on a Mettler Toledo TGA/DSC 1 system. 16.76mg of sample was heated in a corundum crucible between 423 and 973K at a heating rate of 5K/min in a flowing 50% O2/N2 mixture.
100mg of the catalyst was placed in a home-made glass tube, activated at 298K for 2h under vacuum (10–1Pa), mixed with a CH2Cl2 solution containing 0.1M TMPO under nitrogen, treated by ultrasound for 1h (for equilibrium), and then the solvent was evacuated under vacuum. Finally, the sample tube was sealed for storage and transferred to Bruker 4 mm ZrO2 rotor with a Kel-F endcap in a glove box under nitrogen atmosphere before NMR spectroscopy measurement.
Solid-state MAS NMR spectroscopy experiments were carried out using a Bruker Avance III 400WB spectrometer at room temperature. To remove the effect of proton spins on quantitative 31P spectra (i.e., Fig. 4a), a strong radio frequency field (B) is usually applied at the resonance frequency of the non-observed abundant spins (1H herein), which contribute to the coupling of both spin species. The high power decoupling was thus used for the quantitative 31P analysis. Considering the long relaxation time of 31P nuclei in NMR spectroscopy experiment, 30° pulse with 1.20μs width and 15s delay time. The radiofrequency for decoupling was 59kHz. The spectral width was 400ppm, from 200 to −200ppm. The number of scanning was 800 and spinning frequency was 10kHz. The 31P chemical shifts were reported relative to 85% aqueous solution of H3PO4, with NH4H2PO4 as a secondary standard (0.81ppm). The quantitative analysis of adsorbed TMPO molecules was then calculated according to the calibration line established by running standard samples with various adsorbed TMPO concentration47, 48. On the other hand, for samples without adsorbing TMPO (Supplementary Fig. 4), the 1H- 31P cross-polarization was used.
Specific surface areas were analyzed by Micromeritics Tristar II. The samples were measured for the N2 adsorption and desorption at 77K. Before each measurement, the samples were first in-situ degassed overnight at 110°C for 12h.
Continuous-wave electron paramagnetic resonance (CW-EPR) was carried out in the Centre for Advanced Electron Spin Resonance (CAESR), on two spectrometers and frequencies. X-band measurements (ca. 9.4GHz) were performed on a Bruker BioSpin EMXmicro spectrometer with a Bruker SHQE-W cylindrical TE011-mode resonator. The temperature was controlled with an Oxford Instruments ITC-503 cryostat and ESR-900 cryostat. W-band measurements (ca. 94GHz) were performed on a Bruker BioSpin EleXsys E680 spectrometer with a Bruker EN600-1021H cylindrical TE011-mode resonator. The temperature was controlled with an Oxford Instruments ITC-503 cryostat and CF-935O cryostat. For W-band, the microwave frequency was 93.9373GHz for W-band, microwave power 40μW, modulation amplitude 10G, with a sweep of 84s over 450mT. For X-band, the microwave frequency was 9.38914GHz, microwave power 4mW, modulation amplitude 6G, with a sweep of 41s over 80mT. The field modulation frequency was 100kHz at both microwave frequencies. Data analysis was performed in Matlab, and EPR simulations employed the EasySpin toolbox49.
In temperature-dependent X-band EPR measurements the sample does not undergo microwave power saturation down to 20K, while measurements present here were acquired at 85K. Quantitative X-band CW-EPR measurements determined a total spin concentration of 9.1×1016 spins/mg, and double-integration of best-fit W-band simulations shows that the Mo5+ accounts for 80% of EPR-active species. W-band simulations employed a g-strain linewidth model of σ =0.0054 and σ ||=0.0079, which the same as at X-band. Simulations uncertainties are estimated in g-values as +/− 0.001 and hyperfine as +/− 5MHz. Hyperfine values reported as A(95Mo) are those of 95Mo (15.9% natural abundance), while the 97Mo isotope (9.56% natural abundance) is related to 95Mo by the ratio of nuclear g-factors (−0.3734(97Mo)/−0.3657(95Mo)).
For EPR results, at 94GHz, W-band (Supplementary Fig. 22), a signal consistent with six-coordinate Mo5+ is resolved from oxygen vacancies at g=2 and donor defects at g=1.977 with an axial g-matrix of g =1.933 and g ||=1.88450. Natural abundance 95/97Mo5+ hyperfine in X-band (Supplementary Fig. 23), 9.4GHz, point to at least two species in the precursor, which are similar to two of three components in the H2-reduced form, but with about one-fourth overall EPR intensity in the precursor. A third component of reduced catalyst arises as an additional g || feature at g~1.90 and an additional component of hyperfine in the high field A ||(95/97Mo) feature, as demarcated in Supplementary Fig. 24. The W-band sample, aged under air, results in a relative reduction of the g=1.90 feature, indicating that the corresponding site is likely surface-exposed. Multi-component simulations in Supplementary Fig. 24 also show an increase in g || from the precursor value by Δg of ca. 0.0025 upon reduction, on the trend of increasing ionic ligand character.
Magnetic susceptibility measurements of powdered samples of catalyst before and after H2 reduction were carried out using a Quantum Design MPMS-5 SQUID magnetometer at a field of 0.1T and different temperatures (2–300K). Accurately weighed samples (ca. 60mg) were placed into gelatine capsules and then loaded into nonmagnetic plastic straws before being lowered into the cryostat. No corrections were made for the diamagnetic contribution to the magnetic moment due to the excess silica, and it was assumed the paramagnetic contribution originated solely from Mo5+ present in the sample, as determined by ICP-MS. Hence, a total magnetic susceptibility per Mo is reported. For quick comparison the magnetic susceptibility (per Mo) at 2K, for catalyst before and after reduction is 6.73×10–4 and 1.19×10–3cm3mol−1, respectively, demonstrating the formation of more Mo5+ in the catalyst after H2 reduction.
TPR was performed on Micromeritics AutoChem II 2920 using a flow of H2 in N2 (10%, 50mLmin–1) ramping from 309 to 1073K at a rate of 5.0°Kmin–1.
DRIFT spectroscopy studies were performed using a Bruker Tensor 27 spectrometer fitted with a high-sensitivity MCT detector (with resolution of 4cm−1) and a diffuse IR heated chamber equipped with ZnS window. The reduced catalyst (Pd/m–MoO3–P2O5/SiO2) and reduced Pd/SiO2 with mass of 20mg were pre-treated in the presence of 5% H2 in Ar at 383K for 1h. The oxidized catalyst and oxidized Pd/SiO2 (PdO/SiO2) with a mass of 20mg were pre-treated in air at 573K for 1h and then cooled down to 383K. The ZnS window was kept being heated to avoid phenol condensation. After the pre-treatments, the background was scanned in the presence of He at 383K and then the spectra were collected in the presence of phenol/helium mixture at 383K. The background and spectra were recorded at a resolution of 4cm−1 with 128 accumulation scans.
Data that support the findings of this study are available within the article (and its Supplementary Information files) and from the corresponding authors on reasonable request.
H.D. thanks SCG Chemicals Co., Ltd and SCG Packaging Co., Ltd for funding. J.D. acknowledges support from the National Natural Science Foundation of China (Grant No. 11605225) and the Jianlin Xie Foundation of the Institute of High Energy Physics, Chinese Academy of Sciences. Q.W. and J.F. thank State Key Laboratory of Chemical Resource Engineering for funding. W.K.M. is supported by the UK EPSRC 281 (EP/L011972/1, grant to CAESR, the Centre for Advanced Spin Resonance). A.F.R.K. and J.-C.B. thank SCG Chemicals Co., Ltd for funding. A.F.R.K. thanks Wadham College Oxford for a RJP Williams Junior Research Fellowship. We also thank Hefei Light Source and Shanghai Light Source for use of the instruments. We thank Dr Nicholas H. Rees for performing the solid-state NMR spectroscopy; Drs Ashley Shepherd and Robert Jacobs for their help with TPR and BET measurements, Dr Yang He, Jinglu Huang, and Qu Chen for their comments on TEM results, Drs Yufei Song and Wei Chen for their valuable discussion on sample characterizations.
H.D. conceived the idea, designed and carried out the synthesis, characterizations and catalytic reactions, analyzed the data and wrote the manuscript. J.D. and D.C. performed XAFS data analysis and DFT calculation. X.G. performed DRIFTS measurement. Y.-K.P. prepared the samples for TMPO-adsorbed 31P NMR measurement and analyzed the data. W.C. performed the XAFS measurement and data analysis. T.I. initiated the idea, provided WIBO and discussion. W.K.M. performed the EPR measurement and data analysis. M.-J.L. analyzed the XAFS data. N.Y. performed GPC measurement. A.F.R.K. performed the SQUID measurement and data analysis. Y.W. performed XAFS measurement. X.Z. performed the XAFS measurement. S.J. assisted with TEM characterization. Q.W. and J.F. helped with DRIFTS measurement. J.-C.B. assisted in regulating the experiments and discussion. H.C. designed DRIFTS measurement and discussed results. Y.L. provided the resources for TEM characterization and XAFS measurement. S.C.E.T. and D.O.H. supervised the project, helped design the experiments, analyzed the data and wrote the manuscript. All the authors commented on the manuscript and have given approval to the final version of the manuscript.
The authors declare no competing financial interests.
Haohong Duan and Juncai Dong contributed equally to this work.
Electronic supplementary material
Supplementary Information accompanies this paper at doi:10.1038/s41467-017-00596-3.
Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Shik Chi Edman Tsang, Email: firstname.lastname@example.org.
Dermot O’Hare, Email: email@example.com.