Preparation of highly dispersed catalytic Cu from rod-like CuO–CeO₂ mixed metal oxides: suitable for applications in high performance methanol steam reforming

Abstract

A space-confined synthesis method, using a mesoporous template to produce nano-sized copper particles (∼2 nm diameter) by the thermal hydrogen reduction (THR) of high surface area (∼105 m² g⁻¹) rod-like CuO–CeO₂ calcined at 700 °C, is described. X-Ray diffraction (XRD) and temperature program reduction (TPR) results showed the CeO₂ to be doped with Cu, thereby creating more oxygen vacancies. The excess Cu has a high degree of dispersion around 60%. Here we have used the steam reforming of methanol as a model reaction to demonstrate the capability of the materials formed and have used this as a basis of comparison with the impregnation method (carried out at 400 °C) using the same weight loading of Cu. Turnover frequency (TOF) analysis showed the rod-like CuO–CeO₂ mixed oxides to have a 3 fold better catalytic performance compared to analogous materials made by the impregnation method. Additionally, we show that the incorporation of Cu atom into the surface, i.e. non-particulate, structure of CeO₂ provides additional activity sites that are useful in augmenting catalytic performance.

---

1. Introduction

As an energy source hydrogen may eventually serve as a substitute for petroleum. To generate hydrogen from fossil fuels a catalytic reformer—the efficiency, cost, and lifespan of which is strongly dependent on the selection of a suitable catalyst—is often used. CuO–CeO₂ catalysts heterogeneously catalyze various reactions including the water gas shift reaction (WGS),¹ the preferential CO oxidation reaction (PROX),² NO abatement from the auto exhaust gasses³ and the steam reforming of hydrocarbons.^4–6 Some of these reactions are highly efficient with regard to the hydrogen production. The activity of CuO–CeO₂ catalysts in the above-mentioned reactions stems from interactions between the CuO and CeO₂ phases that influence the creation of oxygen vacancies by the facile interplay between Cu^1+/2+ and Ce^3+/4+ redox couples.⁷ One direct approach to enhance the interactions between the Cu–CeO₂ interface areas is to increase the calcination temperature. However, catalysts calcined at high temperatures usually have low surface areas, low (Cu) dispersions and large particle sizes, all of which are factors that lead to a reduction in hydrogen production activity.

Mesoporous silica has nano-sized pores and has been widely used as a molecular sieve in diverse applications, such as drug delivery, chemistry, and biosensors. Some scientists have used mesoporous silica as a template to confine and replicate the structure of metal oxides.^8–11 In a typical protocol a dry metal salt and mesoporous silica are well-mixed and then heated to the melting point of the salt. Due to capillary forces, liquid salt will flow or diffuse into silica's mesopores. We used this method to manufacture rod-like CuO–CeO₂ catalysts to increase the surface area and achieve a high Cu dispersion (60%) at a high calcination temperature (700 °C). Here we report that not only small Cu nano-particles (2 nm) but also other related activity sites promote catalytic performance for the steam reforming of methanol (SRM). This effect occurs by incorporating and reducing the Cu atoms into CeO₂'s surface structure, thereby producing more oxygen vacancies and an enhancement in the SRM reaction rate.

2. Experiment

2.1. Catalyst preparation

Mesoporous silica SBA-15 was synthesized in accordance with an existing literature method¹² except that the surfactants were removed by heating the synthesized materials at 550 °C for 10 h. Cu/Ce atomic ratios of 3 : 7 and 0 : 10, obtained by mixing Cu and Ce nitric salts together with 0.4 ml DI-water, to form a solution, which was placed in a vacuum oven at room temperature until a mucilage-like light blue salt was obtained. The salt was mixed with 1.5 g of SBA-15 and the mixture ground in an agate mortar and pestle. The mortar was placed on a hot plate controlled at 70 °C, until the light blue salt dissolved again, and was then placed in a vacuum oven at room temperature overnight. The mixture was put into a crucible and placed in a muffle furnace. The temperature was increased from room temperature to 700 °C at a heating rate of 1 °C min⁻¹ for 12 h. The specimen was then allowed to cool to room temperature. The silica template was removed by washing with a 2 M aqueous NaOH solution (75 ml) three times at 70 °C. The porous metal-oxide products were recovered by centrifugation and were washed with DI water three times. Herein, the specimens are denoted as MCuCe and MCe. Quantitative Cu determination by inductively coupled plasma (ICP) showed the Cu loading in the MCuCe to be 8.7 wt%. The same amount of Cu was loaded on samples MCe (0 : 10) and ACe (commercial cerium oxide CeO₂ 99.9% from ACROS) by the impregnation method for comparison; these samples were dried at 100 °C for 4 h and calcined at 400 °C for 2 h and designated ImpCu-MCe and ImpCu-ACe respectively.

2.2. Characterization techniques

Specimens were characterized using X-ray powder diffraction (XRD) with a D2 Phaser XRD. The pattern was recorded using a wavelength of 1.5406 Å as the energy of Cu K_α1 for limited angular regions at room temperature and the adopted Sherrer's constant is 0.9. The spectra were obtained at a scan rate of 5° min⁻¹, with steps of 0.05° in the 2θ scans from 20° to 75°. The peak positions were made with reference to JCPDS cards #75-0076 (CeO₂), #80-1268 (CuO) and #85-1326 (Cu). Inductively coupled plasma (ICP-MS) analysis of these samples was used for the quantitative determination of Cu cation. High-resolution transmission electron microscopy (HR-TEM) was conducted using a Philips Tecnai F30 FEI-TEM to determine the catalyst's morphology. The surface area and pore volume were determined using a physisorption analyzer—‘Accelerated Surface Area and Porosimetry’ (Micromeritics ASAP2010). Temperature program reduction (TPR) was controlled by an AutoChem II set at 7 °C min⁻¹ with a 10% H₂/Ar mixture from 50 to 800 °C. The dispersion of Cu, the surface area of reduced Cu and the average particle size were calculated by the N₂O oxidation method.¹³ The experiment was first set with a heating rate of 7 °C min⁻¹ from 50 to 300 °C in an atmosphere comprising a 10% H₂/Ar mixture. Secondly, 10% N₂O/Ar was absorbed on the sample by the pulse method to achieve saturation at 50 °C. Finally, in order to calculate the surface area the copper layer was reduced by employing the same TPR protocol as in the first step (N₂O + 2Cu_surface → N₂ + Cu₂O_surface and Cu₂O_surface + H₂ → 2Cu_surface + H₂O).

2.3. Catalytic test

Steam reforming of methanol (SRM) was conducted under the following conditions: a weight hourly space velocity (WHSV) of 1.5 h⁻¹, a methanol partial pressure of 9.7 kPa, and a water/methanol molar ratio of 1.5. The products were characterized by gas chromatography (GC) on an Acme 6100 equipped with a pulsed discharge helium ionization detector (PDHID). All catalysts were first reduced in a H₂ atmosphere at 300 °C before application to catalytic steam reforming. The methanol conversion X_MeOH is defined as the consumed methanol (molar) amount divided by the input methanol (molar) amount following the reaction equation: CH₃OH + H₂O → 3H₂ + CO₂. The turnover frequency (TOF) of the catalyst is determined by the consumed methanol molar rate divided by the surface molar amount of Cu.

3. Results and discussion

3.1. Morphology of catalyst

TEM images of empty mesoporous SBA-15 replica are shown in Fig. 1(a) and (b). Images of filled SBA-15 with Cu–Ce oxides are shown in Fig. 1(c) and (d) with lateral and axial views. TEM images of rod-like and commercial CeO₂ are shown in Fig. 2(a) and (b), where two samples have distinct appearance. The rod-like morphology of MCe was made by the space confined method, compared with the large particle size of ACe (commercially available CeO₂). The MCuCe represents Cu–Ce oxides made by the space confined replica, as shown in Fig. 2(c) and (d). It should be noted that there are pores between the rods for samples made by the space confined method, in which the rod diameter of MCuCe was about 6–9 nm (TEM), equivalent to mesopore size of the template SBA-15. With this small pore diameter, the metal salts were drawn into the SBA-15's inside space due to capillary force. The template effectively confined the mixed metal oxides, thereby avoiding adverse effects caused by high calcination temperatures. Fig. 2(e) and (f) show that the rod-like structure of MCuCe is maintained after hydrogen reduction. The high resolution TEM image in Fig. 2(f) shows the reduced Cu (111) lattice on the CeO₂ (111) lattice. The Cu particle size diameter was about 2 nm. Due to difficult recognition of Cu as shown in Fig. 2(e) and (f), EDX-mapping of the sample on a Ni grid by HAADF (High Angle Annular Dark Field) was applied to show the dispersion of Cu after the reduction of MCuCe at 300 °C. The presence and distribution of elements Cu, Ce and O are shown in Fig. 3.

Fig. 1 (a) and (b) The lateral and axial views of the template SBA-15, (c) and (d) the lateral and axial views of filled SBA-15 with Cu–Ce oxides.

Fig. 2 (a) MCe (rod-like CeO₂); (b) ACe (commercial CeO₂); (c) and (d) are M-CuCe before H₂ reduction, (e) and (f) are after reduction.

Fig. 3 Highly dispersed Cu on rod-like CeO₂ calcined at 700 °C and reduced at 300 °C by hydrogen.

3.2. Materials characterizations

The XRD patterns of the rod-like MCe and MCuCe, together with samples of impregnated Cu on MCe and commercial CeO₂ (ACe), are shown in Fig. 4. Commercial CeO₂ has a large crystal size [∼50 nm, determined by X-ray diffraction (XRD)] and a small surface area (∼5 m² g⁻¹), so there were sharp peaks for each position in the XRD pattern. The rod-like MCe and MCuCe all have small crystal sizes (∼8 nm). The characteristic peaks of CeO₂ in MCe, MCuCe and ImpCu-MCe are very similar, demonstrating that by using the space confined method with a mesoporous template is an ideal way to produce CeO₂ nanoparticles, while avoiding rapid growth in the crystal's size at higher calcination temperatures (>700 °C). Compared with samples of Cu impregnated on MCe and ACe (ImpCu-MCe and ImpCu-ACe), the CuO peaks in the XRD pattern of ImpCu-ACe are sharper than those of ImpCu-MCe. The CuO on ACe tended to form larger particles in comparison with the same amount of copper contained in CuO on MCe. Peaks of CuO in the MCuCe were hardly noticeable possibly due to its nanoparticle size. Doping Cu²⁺ ion into CeO₂ could result in the lattice cell contraction due to smaller copper ion. In addition, it is also possible that the insertion of CuO in the lattice of CeO₂ causes the lattice expansion. The overall effect was the shift of MCuCe peaks towards lower angle. The estimated CuO crystal sizes are shown in Table 1.

Table 1 ICP, BET and crystal size data of various CeO₂ and Cu/CeO₂ samples

Samples	ICP Cu (wt%)	BET		Crystal size (111) for XRD
		Surface area^a/m^2 g^−1	Pore volume/cm^3 g^−1	CeO2/nm	CuO/nm	300 °C reduction Cu/nm
a Before the hydrogen reduction.
MCe	0	109	0.24	8.8	–	–
ACe	0	5	0.02	50	–	–
MCuCe	8.7	105	0.25	8.2	NA	NA
ImpCu-MCe	8.3	82	0.18	8.9	9.5	13
ImpCu-ACe	7.5	5	0.02	50	24	48

---

Fig. 4 XRD patterns of ACe, ImpCu-ACe, ImpCu-ACe, MCuCe and MCe.

As demonstrated by TPR measurements (Fig. 6), copper oxide is able to be reduced at a temperature below 300 °C. The XRD patterns in Fig. 5 show the effect of the hydrogen reduction at 300 °C on the crystal structures of the rod-like MCuCe, ImpCu-MCe and ImpCu-ACe. The reduced Cu (111) peaks became recognizable in the impregnated samples, but not in MCuCe (sample denoted as MCuCe-Re). The reason is assumed to be that small Cu particles cannot be detected by the in-house XRD. This assumption was supported by elevating the reduction temperature to 800 °C, in which case, the reduced Cu (111) peaks in the MCuCe appeared (sample denoted as MCuCe-Re800). As a result, a finer dispersion of Cu on the CeO₂ surface was confirmed and could be achieved by the space confined method followed by the reduction procedure. The MCuCe's CeO₂ (111) peak position before and after hydrogen reduction shifted from 28.60° to 28.51°, ImpCu-MCe from 28.72° to 28.69° and ImpCu-ACe from 28.75° to 28.70°. The significant difference of the (111) peak of MCuCe to other two samples is attributed to the incorporation of smaller Cu into the CeO₂ lattice, when the sample was prepared by the space confined method. The shifts of XRD peaks in all samples after hydrogen reduction were also observed. Ce⁴⁺ was partially reduced to Ce³⁺ and Ce³⁺ has a larger ionic radius than Ce⁴⁺. Hence, the shifts were all towards lower angles.

Fig. 5 XRD patterns for samples after the hydrogen reduction. The -Re indicates the reduction at 300 °C, and -Re800 the reduction at 800 °C.

The reduction also creates oxygen vacancies near the surface, due to diffusion of oxygen to the surface. The oxygen vacancy formation process is essentially facilitated by a simultaneous condensation of two electrons into the localized f-level traps on two cerium atoms. Thus, when an oxygen atom moves diffusively towards the surface, e.g., oxygen vacancy moves into the crystal, these electrons localize on cerium atoms in the immediate surrounding of the vacancy.¹⁴ The formation of reduced oxides can be understood as a formation, migration and ordering of virtual Ce³⁺-vacancy complexes.¹⁴ Thus, the results indicate that the generation of oxygen vacancies was enhanced by the Cu doping and the reduction.

3.3. TPR analysis

Temperature programmed reduction (TPR) was used to analyze the degree of Cu reduction. The hydrogen consumption signals for MCuCe, ImpCu-MCe and ImpCu-ACe show different degrees of reduction at temperatures below 300 °C. The Cu in MCuCe, ImpCu-MCe and ImpCu-ACe was able to be reduced at 195 °C, 245 °C and 280 °C, respectively. The wide peak representative of ImpCu-ACe, which has the largest CuO particle size, indicates that a higher reduction temperature was required to completely reduce the Cu, as shown in Fig. 6. Below 300 °C, there was no hydrogen consumption signal for MCe. When the temperature exceeded 350 °C, the oxygen in the CeO₂ lattice started to be released, as we observed in TPR profiles. When compared with MCe and ACe, the oxygen in MCe is more easily released, because MCe has a smaller crystal size and is more porous as reflected by the MCe surface area measurement, which is 20-fold larger (as measured by BET) than ACe (Table 1). ImpCu-MCe exhibits a similar temperature signal with respect to oxygen release as MCe; but, there is no such peak for MCuCe at 550 °C. These results for MCe and ImpCu-MCe suggest that the oxygen is released from the CeO₂ lattice, in other words, the oxygen vacancies were created when Cu was doped into CeO₂ at a high calcination temperature (700 °C), and that some of the Ce atoms are replaced by Cu atoms.^7,15 Especially, when nano-sized Cu is generated by reduction, it is believed that more oxygen vacancies are also produced. In contrast when the ImpCu-MCe was calcined at a lower temperature (400 °C), Cu or CuO were found only on the surface of CeO_2, rather than in the CeO₂ lattice. Therefore, it is believed that it requires a reduction temperature higher than 700 °C to be able to release oxygen from the CeO₂ lattice as seen in ACe and ImpCu-ACe. Overall, the doping of Cu in CeO₂ and the reduction in MCuCe help create more oxygen vacancies. The porous and rod-like MCe enables to release oxygen at a lower reduction temperature than the commercial CeO₂ support.

Fig. 6 Temperature programmed reduction profiles for MCe, MCuCe, ImpCu-MCe and ImpCu-ACe.

3.4. Cu dispersion analysis

Copper dispersion (D_Cu), defined as the ratio of Cu exposed at the surface to the total Cu present, was calculated from the amount of H₂ consumed in the TPR analysis. The first TPR results reflected the total amount of Cu available. The second TPR was conducted after the material was exposed to N₂O for surface oxidation. Since a monolayer of Cu₂O formed on the surface due to N₂O,¹⁶ the actual surface Cu equals the double of the H₂ consumption amount in the second TPR, following the stoichiometric relationship. Thus the dispersion of copper can be derived accordingly. The dispersion degree of the catalyst is known to play a critical role in evaluating the catalytic performance under the same loadings. H₂ consumption (from TPR) can also be used to calculate the copper metal surface area (S_Cu) and the average particle size (φ_av) by making the following assumptions. The area per copper surface atom in the (100), (110), and (111) planes is 0.065, 0.092, and 0.056 nm², respectively.¹⁶ An equal abundance of these three planes gives an average copper surface atom area of 0.071 nm², equivalent to 1.4 × 10¹⁹ copper atoms per square metre. By assuming a spherical shape of the copper metal particles, S_Cu and φ_av can be expressed as eqn (1) and (2), respectively.

S_Cu (m² g⁻¹_Cu) = Mol_H2 × SF × N_A/(10⁴ × C_M × W_Cu) (1)

φ_av (nm) = 6000/(S_Cu × ρ_Cu) (2)

where Mol_H2, SF, N_A, C_M, W_Cu, and ρ_Cu are moles of hydrogen experimentally consumed per unit mass of catalyst (μmol_H2 g⁻¹_cat), a stoichiometric factor equal to 2, Avogadro's number (6.022 × 10²³ per mol), number of surface Cu atoms per unit surface area, Cu content (wt%), and the density of copper (8.92 g cm⁻³).¹⁶ The degree of dispersion of Cu, the metal surface area, and the average particle size are given in Table 2. It was observed that with respect to dispersion, the surface area of reduced copper and the average particle size, MCuCe shows the best results. The small diameter (6–9 nm) of the rod-like mixed metal oxides after hydrogen reduction can generate Cu nanoparticles with an average particle size around 1.7 nm. Both the surface area and the dispersion of these generated Cu nanoparticles are much higher than the counterparts loaded by the impregnation, as summarized in Table 2. These features of reduced MCuCe ensure a high TOF in catalytic steam reforming of methanol.

Table 2 Consumption H₂, dispersion, copper metal area, average particle size and turnover frequency of steam reforming of methanol at 210 °C of various Cu/CeO₂ catalysts

Samples	Consumption of H2		Information of Cu by N2O method			SRM
	First TPR/μmolH2 g^−1cat.	Second TPR^a/μmolH2 g^−1cat.	Dispersion DCu (%)	Surface area SCu/m^2 g^−1Cu	Average particle size φav/nm	TOF/h^−1 at 210 °C
a Second TPR performed after N2O surface oxidation.
MCuCe	1329	403	60	397	1.7	94
ImpCu-MCe	1293	114	18	118	5.7	33
ImpCu-ACe	1220	17	3	20	33	11

---

3.5. Reaction test

Steam reforming of methanol (SRM) was used to evaluate the catalyst's performance as shown in Fig. 7. In general, high methanol conversions at low reaction temperatures are desired. MCe is pure CeO₂ without a Cu content, so there is no methanol conversion even at 300 °C. At 240 °C, the MCuCe, ImpCu-MCe and ImpCu-ACe's conversion values were 100%, 16% and 2%, respectively. The turnover frequency (TOF) analysis results at 210 °C are shown in Table 2. The turnover frequency can be understood as the reactant consumed or product generated per (surface) catalyst weight per unit time. Under the same support, the MCuCe's TOF value is 3-fold greater than ImpCu-MCe, indicating that nanosized Cu with higher surface area and better dispersion in MCuCe contributed to higher activity of the catalyst. The other reason that the improvement might result from other active sites such as oxygen vacancies¹⁷ and the interface between copper and ceria support, in addition to the catalytic effect of reduced Cu particles.¹⁸ ImpCu-MCe and ImpCu-ACe are both prepared by the impregnation method and at the same calcination temperature (400 °C). The Cu doped in CeO₂ is limited at this calcination temperature. The ImpCu-MCe's TOF value is three times higher than ImpCu-ACe, because the dispersion, particle size and the surface area of reduced Cu in ImpCu-MCe are much different to ImpCu-ACe's.

Fig. 7 The steam reforming of methanol test used to determine the methanol conversion of various Cu/CeO₂ catalysts.

4. Conclusion

We used the space confined method to synthesize rod-like mixed copper and cerium oxides at calcination temperatures as high as 700 °C for 12 h that maintained their crystal size (∼8 nm) and high surface areas (105 m² g⁻¹) without sintering or crystal enlargement. Additionally, a high dispersion of reduced Cu particle can be obtained at a hydrogen reduction temperature below 200 °C and shows nearly 100% catalytic performance of steam reforming of methanol at 240 °C. Under the similar loading of Cu, the space confined method provides a better preparation of catalyst than the impregnation method. First, a finer dispersion and easier control over the particle size can be achieved. Nano-sized Cu can be produced by the hydrogen reduction. Second, the enhancement of catalytic activity not only confirms the effectiveness of fine copper nanoparticles, but also the role of oxygen vacancies, which was promoted by the ordered, mesoporous and high surface area of ceria structure. The doping of Cu into the lattice and its generation by reduction also benefits the formation of oxygen vacancies, which contribute to the interaction between the catalyst and support and the catalytic activity. This can be highlighted by the fairly low reduction temperature less than 200 °C to achieve a better catalytic effect than catalyst deposited by impregnation or using a commercially available cerium oxide support. Overall, the synthesis method using a space confined template overcomes the adverse effects associated with high calcination temperature and it offers a new approach to prepare various mixed solid oxides and heterogeneous catalysts.

Acknowledgements

The authors gratefully acknowledge financial support from the National Science Council of Taiwan (NSC-100-2221-E-011-105-MY3), the National Taiwan University of Science and Technology (NTUST) and the National Synchrotron Radiation Research Center (NSRRC).

References

1. H. Yahiro, K. Murawaki, K. Saiki, T. Yamamoto and H. Yamaura, Catal. Today, 2007, 126, 436–440.
2. G. Avgouropoulos, J. Papavasiliou and T. Ioannides, Catal. Commun., 2008, 9, 1656–1660.
3. P. Bera, K. R. Priolkar, P. R. Sarode, M. S. Hegde, S. Emura, R. Kumashiro and N. P. Lalla, Chem. Mater., 2002, 14, 3591–3601.
4. F. Auprêtre, C. Descorme and D. Duprez, Catal. Commun., 2002, 3, 263–267.
5. H. Oguchi, T. Nishiguchi, T. Matsumoto, H. Kanai, K. Utani, Y. Matsumura and S. Imamura, Appl. Catal., A, 2005, 281, 69–73.
6. T. Matsumoto, T. Nishiguchi, H. Kanai, K. Utani, Y. Matsumura and S. Imamura, Appl. Catal., A, 2004, 276, 267–273.
7. W. Shan, W. Shen and C. Li, Chem. Mater., 2003, 15, 4761–4767.
8. B. Z. Tian, X. Y. Liu, H. F. Yang, S. H. Xie, C. Z. Yu, B. Tu and D. Y. Zhao, Adv. Mater., 2003, 15, 1370–1374.
9. W. Yue and W. Z. Zhou, Chem. Mater., 2007, 19, 2359–2363.
10. M. B. Zheng, J. Cao, S. T. Liao, J. S. Liu, H. Q. Chen, Y. Zhao, W. J. Dai, G. B. Ji, J. M. Cao and J. Tao, J. Phys. Chem. C, 2009, 113, 3887–3894.
11. S. C. Yang, W. N. Su, S. D. Lin, J. Rick, J. H. Cheng, J. Y. Liu, C. J. Pan, D. G. Liu, J. F. Lee, T. S. Chan, H. S. Sheu and B. J. Hwang, Appl. Catal., B, 2011, 106, 650–656.
12. D. Y. Zhao, J. L. Feng, Q. S. Huo, N. Melosh, G. H. Fredrickson, B. F. Chmelka and G. D. Stucky, Science, 1998, 279, 548–552.
13. G. C. Chinchen, C. M. Hay, H. D. Vandervell and K. C. Waugh, J. Catal., 1987, 103, 79–86.
14. N. V. Skorodumova, S. I. Simak, B. I. Lundqvist, I. A. Abrikosov and B. Johansson, Phys. Rev. Lett., 2002, 89, 1666011–1666014.
15. Z. X. Yang, B. L. He, Z. S. Lu and K. Hermansson, J. Phys. Chem. C, 2010, 114, 4486–4494.
16. J. W. Evans, M. S. Wainwright, A. J. Bridgewater and D. J. Young, Appl. Catal., 1983, 7, 75–82.
17. Z. L. Wu, M. J. Li, J. Howe, H. M. Meyer and S. H. Overbury, Langmuir, 2010, 26, 16595–16606.
18. X. Q. Wang, J. A. Rodriguez, J. C. Hanson, D. Gamarra, A. Martinez-Arias and M. Fernandez-Garcia, J. Phys. Chem. B, 2006, 110, 428–434.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c2cy00330a

---