How Topological Differences between Two Oxide Surfaces Determine Selectivity—The Case of the Dehydra-Decyclization of Tetrahydrofuran

Abstract

Production of butadiene from biomass-based tetrahydrofuran (THF) is explored as an alternative to the existing petroleum-based processes. Metal oxide catalysts have been shown to exhibit varying product selectivities when reacted with THF. Among those oxides, ZrO₂ showed the highest selectivity for butadiene. In contrast, Al₂O₃ showed the highest selectivity for the competing retro-Prins products, C₃H₆ and HCHO. The reasons behind the varying selectivity across oxides are unclear. In this work, we employ periodic density functional theory and mean-field microkinetic modeling to investigate the mechanism of the reaction of THF to butadiene and retro-Prins products on t-ZrO₂ (101) (dry and hydrous) and on γ-Al₂O₃ (110). Our simulations reproduce the experimental selectivity trends. High selectivity for butadiene is promoted by the presence of neighboring Lewis acid metal sites that facilitate E1cB hydroxyl elimination from a 3-butenoxide intermediate; on hydrous Al₂O₃; where such neighboring Lewis acid centers are not available, the butenoxide undergoes E2 elimination and retro-Prins products ensue. The THF ring opening is rate-determining on ZrO₂, whereas the γ-proton elimination that yields the 3-butenoxide intermediate is rate controlling on hydroxylated Al₂O₃. We conclude that the local topology around the active site greatly influences the mechanism and selectivity.

1. Introduction

1,3-Butadiene is an important, high-demand chemical used in the manufacture of styrene-butadiene rubber (SBR) and acrylonitrile-butadiene-styrene (ABS). Currently, butadiene and other C4 compounds are produced as side products in the production of ethylene by steam cracking of crude oil. However, the current revolution in shale gas (primarily light hydrocarbons) is resulting in diminished butadiene production, thereby increasing its demand-supply gap. Moreover, non-renewable feedstocks, high reaction temperatures and high C4 separation costs involved in the existing technologies further exacerbate the situation. Thus, it is highly desirable to establish greener alternative processes for ‘on-purpose’ butadiene production. Biomass-based butadiene production is a potential alternative as it involves renewable feedstocks and catalytic reactions typically run at lower temperatures than steam cracking.

Dehydra-decyclization of biomass-derived tetrahydrofuran (THF) [1,2,3,4] is a promising route to butadiene. The reaction can be catalyzed by Brønsted acid catalysts [5,6,7], and mechanistic studies on H-[Al]MFI [8] have elucidated the mechanism. However, Brønsted acids also catalyze the retro-Prins condensation to propene and formaldehyde [8], the major side reaction (ca. 30% selectivity at 220–270 °C) that suppresses the formation of butadiene. On H-[Al]MFI [8], dehydra-decyclization to butadiene (Scheme 1) proceeds through 2-buten-1-ol, which forms by 1,2-hydride shift in the primary 4-hydroxybutyl oxide intermediate that forms from the THF ring opening; the 2-buten-1-ol subsequently undergoes Brønsted-catalyzed E2 elimination to the diene [8]. If the protonated open ring (primary 4-hydroxyl butyl cation) is not effectively stabilized by the conjugate base, then it can undergo coordinated 1,2-hydride shift and C-C bond scission yielding propene and formaldehyde (Scheme 1). Although substantially less likely, this coordinated nuclear motion remains kinetically possible after the stabilization of the 4-hydroxyl butyl cation by the conjugate base [8]. These insights lend perspective to the work of Abdelrahman et al. [5] and Kumar et al. [7], who reported that P-SPP (phosphorus-self pillared pentasil) [5,9] and B-MWW (framework boron-containing MWW) are less selective than H-ZSM5 for retro-Prins products when reacted with THF, presumably on account of their weak Brønsted acidity. The lower selectivity for retro-Prins notwithstanding, the rate of butadiene formation over weak Brønsted acids is low.

Venturing into the Lewis catalyst space, and after testing several metal oxide catalysts (ZrO₂, CeO₂, TiO₂, Nb₂O₅, WO_x/ZrO₂), Ji et al. [10] reported that ZrO₂ consistently exhibited the highest selectivity for 1,3-butadiene (ca. 90%) when reacted with THF. Its catalytic performance was also structure-insensitive and the catalyst was relatively stable with time on stream—although the tetragonal phase of ZrO₂ was found to be the most stable one. Apart from Al₂O₃, which was ca. 90% selective for retro-Prins products, the rest of the oxides were more selective than H-ZSM5 for 1,3-butadiene. Adding to its high selectivity for dienes, ZrO₂ is also highly selective for a single diene isomer when multiple dienes can form (C5+) [11].

In this work, we undertake mechanistic studies aimed at understanding the different performance of ZrO₂ and Al₂O₃ when reacted with THF. We perform periodic density-functional theory (DFT) calculations and microkinetic simulations to investigate the reaction mechanisms of the conversion of THF to butadiene and retro-Prins products on the models of the predominant facets of t-ZrO₂ and γ-Al₂O₃ [10,12,13,14]. Although butanal can form and subsequently undergo hydrogenation to butene [10,15,16], this is not a reaction pathway we have investigated here, as Ji et al. reported only the presence of trace amounts of butene. The kinetic parameters and selectivity values obtained from the simulations are compared with available experimental data [10]. Last, we provide insights into the factors that bring about selectivity on ZrO₂ and Al₂O₃.

2. Materials and Methods

2.1. DFT Simulations

Periodic DFT calculations were performed using the Vienna ab-initio simulation package (VASP 5.4.1). The core electrons were modeled with the projector augmented wavefunction (PAW) pseudopotential, and the valence electrons were modeled with the Perdew−Burke−Ernzerhof (PBE) exchange-correlation functional and a plane-wave basis set with energy cut-off at 400 eV [17,18]. Dispersion interactions were modeled using the D3 method of Grimme with Becky-Johnson dampening [19]. Transition states were determined using the nudged-elastic band (NEB) + dimer [20] or the NEB + climb [21] method. Electronic energies and ionic forces were converged to 10⁻⁶ eV and 0.02 eV/Å, respectively. For the very few transition state calculations that could not converge to 0.02 eV/Å, we relaxed the cut-off criterion to a higher value (up to 0.05 eV/Å). The ionic degrees of freedom were optimized using the conjugate gradient algorithm. Vibrational frequencies were estimated by partial diagonalization of the Hessian matrix using nuclear displacements of 0.015 Å. For shallow, first-order saddle points, a displacement of 0.002 Å was deemed necessary to eliminate redundant imaginary frequencies. Thermal corrections to the electronic energies at 673.15 K were calculated using the quasi-rigid-rotor-harmonic-oscillator (quasi-RRHO) model proposed by Grimme and Li et al. [22,23]; the vibrational enthalpies and entropies corresponding to modes below 100 cm⁻¹ have been replaced by corresponding free-rotor values. Small imaginary frequencies appear in a few stationary and transition states, and they have been replaced by a wavenumber below 100 cm⁻¹ (50 cm⁻¹) [24]. The energies of gas phase molecules were referenced to the corresponding experimental values obtained from NIST and RMG-MIT databases using the referencing scheme in pMuTT [25].

The (101) termination of tetragonal ZrO₂ (t-ZrO₂) was represented by a (2 × 4 × 1) periodic slab (dimensions 12.89 Å × 14.58 Å × 28.86 Å) (Figure 1a). The model consists of 4 Zr-O layers with the bottom two layers frozen and a vacuum spacing of 20 Å along the z-direction between the periodic images (Figure 1a,b). The reciprocal space was sampled on a 3 × 3 × 1 k-point mesh. The 4s, 4p, 4d, and 5s electrons of Zr were treated as valence electrons. Hubbard’s U parameter with a U value of 4 eV [26] was added to the Hamiltonian to reduce the delocalization of the outermost d-electrons. This model consists of 64 Zr atoms and 128 O atoms. There are two types of surface Zr atoms (Zr_a and Zr_b) and two types of O atoms (O_3a and O_3b). The Zr in the bulk is coordinated to eight O atoms whereas the surface Zr atoms are hepta-coordinated to O atoms, three of them on the surface and the other four in the sub-surface layers. The bulk O atoms are coordinated to four Zr atoms whereas the surface O atoms are coordinated to three Zr atoms.

We model Al₂O₃ as a 2 × 1 unit cell (dimensions 8.40 Å × 16.14 Å × 27.40 Å) of (110) termination of γ-Al₂O₃ (Figure 1c) [12,27]. The model consists of 32 Al atoms and 48 O atoms. The bottom Al-O layer (8 Al and 12 O atoms) is frozen, and a vacuum spacing of 24 Å along the z-direction between the periodic images. The reciprocal space was sampled on a 3 × 2 × 1 k-point mesh. The 3s and 3p electrons of Al were treated as valence electrons. As shown in (Figure 1c), the surface consists of different types of under-coordinated Al and O sites. In the case of Al sites, there are tri-coordinated Al_III sites, and two types of tetra-coordinated Al sites, namely Al_IVa and Al_IVb. There are two types of di-coordinated O atoms (O_2a and O_2b) and two types of tri-coordinated O atoms (O_3a and O_3b). We model the hydroxylated surfaces of ZrO₂ (Figure 1b) and Al₂O₃ (Figure 1d) by chemisorbing a H₂O molecule on the most Lewis acid-base pairs, Zr_a-O_3a and Al_III-O_2a of ZrO₂ and Al₂O₃, respectively. The coordinate files of the ZrO₂ and Al₂O₃ models are provided in the Supplementary Materials.

2.2. Microkinetic Simulations

Microkinetic modeling was performed using CHEMKIN [28]. The simulations were run at a temperature of 673.15 K and total pressure of 1 atm. The mole fraction of THF in the feed is 0.05, with the rest being an inert gas, and these values are chosen based on the experiments performed by Ji et al. [10]. The PFR reactor was modeled as a series of 900 CSTR reactors. The values of the input parameters and details about evaluating them are provided in Section S1 in the Supplementary Materials. Sensitivity analysis was performed by perturbing the forward and backward rate constants by ±4%.

3. Results

3.1. Dry t-ZrO₂ (101)

The t-ZrO₂ (101) surface exposes two distinct Zr sites, denoted as Zr_a and Zr_b in Figure 1a. Surface-bound THF interacts with either of these two sites via its ring-O alone (ring in upright configuration) and the respective binding energies relative to gas-phase THF are −1.13 and −0.95 eV.

Pathway 1_Zrd in Scheme 2 depicts the dominant route to butadiene. Considering the salient elements of it (a detailed description is in Section S2.1 in the Supplementary Materials), we see that there is substantial departure from the mechanism on H-[Al]MFI outlined above—but there are also elements reminiscent of it. Following the ring opening, the O atom of THF is coordinated by Lewis acidic surface Zr and the oxide stabilizes the primary carbocation (2_Zrd) via coordination by a surface O. Intermediate 2_Zrd then undergoes γ-proton elimination by a surface oxygen to form the 3-butenoxide intermediate (3_Zrd). Butadiene (5_Zrd) forms through the 2-butenoxide intermediate 4_Zrd, which is reminiscent of the 2-buten-1-ol intermediate on H-[Al]MFI. However, on the oxide, the double bond shift from 3_Zrd to 4_Zrd is effected by elimination of the allylic proton by a surface oxygen. The THF O atom, which has been coordinated to a surface Zr all the while, attracts one of the two surface protons in the vicinity, and this sets off the C-O bond breaking along with rehybridization of the backbone carbons yielding butadiene.

Pathway 2_Zrd in Scheme 3 is the kinetically dominant mechanism to retro-Prins products. It is remarkably similar to the mechanism on H-[Al]MFI as the stabilized primary carbocation 2_Zrd undergoes coordinated 1,2-hydride shift from C4 to C5 and C2-C3 bond scission. The free energy and enthalpy profiles of all the pathways are shown in Figure S1. The adsorption of THF (ΔH_ads = −1.14 eV), H₂O (ΔH_ads = −1.5 eV), and HCHO (ΔH_ads = −0.63 eV) is exothermic (Figure S1b). From the adsorption enthalpies, we can conclude that THF and HCHO chemisorb on the surface—by strongly interacting with the Lewis-acidic Zr sites via the lone pairs of the O atoms in these molecules. When compared to the rest of the reactant and product species, H₂O binds the strongest to the surface via dissociative chemisorption, and this strong binding of H₂O is also reflected in the highest free energy-based stability of the intermediate (6_Zrd) (Figure S1a).

For completeness, two more retro-Prins pathways (pathways 3_Zrd and 4_Zrd) and the corresponding free energy profiles are described in the Supplementary Materials (Section S2.1) but they do not contribute in any significant way as microkinetic analysis shows that 99.9% of the reactive flux is on the butadiene Pathway 1_Zrd while the remaining 0.1% of it is on the retro-Prins Pathway 2_Zrd (Table S5). The product selectivities are in satisfactory agreement with the experimental values (Figure S2) [10].

Sensitivity analysis (Figure 2a,b) shows that the ring opening (TS1_Zrd) primarily and the γ-proton elimination (TS2_Zrd) to a lesser extent control the rate of butadiene formation. Since S_butadiene is nearly 100%, the selectivity for butadiene is insensitive to any of the elementary steps. On the other hand, the rate and selectivity for propene are enhanced by TS3_Zrd and the hydride-shift TS4_Zrd (for TS4_Zrd, see Scheme S1 in Supplementary Materials) and hampered by the γ-H abstraction. Since the ring opening transition state (TS1_Zrd) is higher than all the rest and is common to all the pathways, it also enhances the rate for propene formation. Surface coverage analysis shows that the chemisorbed H₂O (6_Zrd) is the most abundant reaction intermediate on account of its stability (see profile in Figure S1); water coverage increases with conversion (Figure 2c).

3.2. Hydroxylated t-ZrO₂ (101)

The hydroxylated t-ZrO₂ (101) model (Figure 1b) consists of chemisorbed H₂O on a Zr_a-O_3a pair, forming a terminal and a bridging OH group; this model corresponds to 1 H₂O molecule/supercell with the corresponding surface coverage being 8.82 × 10⁻¹⁰ mol/cm². To investigate the effect of the OH groups on the mechanism, we put THF on sites nearest to the OH groups, Zr_a1 and Zr_a2 sites. The THF binding energies on Zr_a1 and Zr_a2 are nearly the same, −1.06 and −1.05 eV, respectively. For the subsequent ring opening, we evaluate the free energy barriers of ring opening on the Zr_a1-OH, Zr_a2-OH-Zr_b and Zr_a2-O_3a2 sites. The O atoms of the OH groups participate actively by coordinating the C5 atom of THF. The ring opening on Zr_a2-OH-Zr_b is more facile than on the other two sites (free energy profiles and the corresponding analysis are provided in Section S3.1 in the Supplementary Materials). In the following, we consider catalytic pathways on Zr_a2-OH-Zr_b.

The salient elements of the dominant pathways are shown in Scheme 3 (detailed description in Section S3.2 in the Supplementary Materials). Pathway 1_Zrh to butadiene is similar to Pathway 1_Zrd (Scheme 2) on the dry-ZrO₂ surface. The minor differences between them lie in the type of the surface atoms participating in the proton-shuttling step. Specifically, on the hydroxylated surface, the atoms that participate in the reaction are Zr_a2, O_3a2, the bridging OH (O_b), and the terminal OH, whereas on the dry-ZrO₂ surface, the participating atoms are Zr_a, O_3a, and O_3b (Scheme 2). On Pathway 1_Zrh, after the γ-proton elimination step, a proton shuttles from the terminal OH to the bridging OH, whereas on the dry-ZrO₂ surface, the proton in 3_Zrd shuttles from O_3b to O_3a (TS6_Zrd in Section S3.2). More details are provided in Section S3.2. Regarding the retro-Prins reaction, Pathway 2_Zrh is the most dominant one and is mechanistically similar to Pathway 4_Zrd on the dry surface. In addition to the retro-Prins Pathway 2_Zrh, there is a less dominant one (Pathway 3_Zrh) described in Section S3.2.

Comparing the free energy profiles for the dry and hydrous ZrO₂ surfaces (Figure S4a), we make two observations that have opposite effects on the THF dehydra-decyclization rate. First, the THF ring-opening barrier (TS1_Zrh in Figure S4a) is 0.1 eV higher on the hydrous than on the dry surface, suggesting a slower reaction. Second, dissociated water (6_Zrh) is less stable on the hydroxylated surface than on the dry one, which should result in a higher fraction of empty sites and hence higher reaction rates. Figure S4 shows the free energy and the enthalpy profiles of all four pathways. Similar to the dry-ZrO₂ model, enthalpy profiles show that THF, HCHO, and H₂O strongly chemisorb on the surface, with H₂O binding the strongest via dissociative chemisorption (Figure S4b).

To analyze these competing effects, we performed microkinetic modeling for the Zr_a2-O_3a2 active site. The site density for Zr_a2-O_3a2 is 4 sites/supercell, which is equal to the maximum number of Zr_a2-O_3a2 sites in a 2 × 4 × 1 ZrO₂ supercell. We found that, although there is less surface water (6_Zrh) and more open sites on the hydrated surface (cf. Figure 2b and Figure S5a), the turnover frequency (TOF) on the hydroxylated surface is 2.4 times as low as on the dry one (TOF values in Table S7). Thus, between the increase in the ring-opening barrier and the increase in the fraction of empty sites, the former has more of an impact on the TOF. S_butadiene and S_propene are 99.8% and 0.2%, respectively (Figure S5b), not much different from those on the dry surface and in good agreement with experiments [10]. Additionally, sensitivity analysis (Figure S5c,d) shows that, similar to the mechanism on the dry surface, the ring opening (TS1_Zrh) predominantly controls the reaction rates to butadiene and propene. While S_butadiene is insensitive to the rate of any of the steps, S_propene is enhanced by the retro-Prins step (TS11_Zrh) and hampered by the competing C-O bond cleavage step (TS5_Zrh).

Table 1. Apparent activation energy values from experiment [10] and microkinetic simulations. The reactor parameters, reaction and feed conditions are shown in Table S8 in the Supplementary Materials.

Experiment/Model	Ea (kJ/mol)
Experiment	65.2 ± 12.1
Dry-ZrO2	94.2 ± 0.3
Hydroxylated-ZrO2	73.3 ± 0.2

contains the apparent activation energies, E_a, obtained from the microkinetic analysis of the dry and hydroxylated ZrO₂ surface models. On account of the experimental E_a being closer to the hydroxylated than the dry zirconia model, one could reasonably infer that the former is a better representation of the real system. This prediction creates an avenue for experimentalists to perform spectroscopic experiments such as FTIR under operando conditions to identify the concentration and type of surface hydroxyl group and various other intermediates.

3.3. Dry γ-Al₂O₃ (110)

We investigate the mechanism of THF conversion on both dry and hydroxylated surfaces of γ-Al₂O₃ (110). Although the Al_III site is hydroxylated at 400 °C [12,27,29,30,31,32], here we consider the dry surface as well in order to understand the effect of surface OH on the reaction mechanism.

THF coordinates to Al_III or Al_IVb sites with its oxygen and assumes an upright configuration (Figure 1c). The respective binding energies are −1.98 eV and −1.51 eV. Scheme 4 shows the salient features of the most dominant pathways that result from the binding of THF on the Al_III site (details in Section S4.1 in the Supplementary Materials). Pathway 1_Ald to butadiene is mechanistically similar to Pathway 1_Zrd, with some minor differences: first, and unlike Pathway 1_Zrd, on Pathway 1_Ald, the γ-proton elimination is not followed by proton hopping; and second, following butadiene desorption, the H₂O recombination is preceded by proton hopping. Pathway 2_Ald is the most dominant pathway to retro-Prins products and is mechanistically similar to Pathway 2_Zrd on dry-ZrO₂. There are two more retro-Prins pathways (Pathways 3_Ald and 4_Ald) described in Section S4.1, but they are associated with very low reaction fluxes and thereby are of minor catalytic interest. The free energy and enthalpy profiles of all the pathways are shown in Figure S6 in the Supplementary Materials. As seen in the cases of ZrO₂ models, enthalpy profiles (Figure S6b) show that both THF and HCHO chemisorb on the Al_III sites of the surface (ΔH_{ads_THF} = −2 eV and ΔH_{ads_HCHO} = −1.43 eV). H₂O can chemisorb on the surface both in undissociated (7_Ald) and dissociated forms (6_Ald and 5b_Ald). When compared to ZrO₂ models, the binding strength of THF, HCHO, and dissociated H₂O are much higher on the dry-Al₂O₃ surface, and the strong binding of dissociated H₂O is also reflected in the highest free energy-based stability of intermediates 6_Ald and 5b_Ald.

Microkinetic analysis of fluxes of pathways (Table S9) and product selectivities (Figure 3a) confirm that Pathway 1_Ald has nearly 100% flux and hence, S_butadiene is close to 100%, which does not agree with the experimental selectivity on Al₂O₃ [10]. Surface coverage analysis (Figure 3b) shows that at very low conversions (1%), the total fraction of chemisorbed H₂O is close to 1 (Figure 3b), a consequence of the hydrophilic nature of the Al_III site. This implies that once all the Al_III sites are blocked by chemisorbed H₂O, the surface becomes inactive. We also performed sensitivity analysis (Figure S7 in the Supplementary Materials) and the results were very similar to those on the dry t-ZrO₂ (101) surface. Considering the significant blocking of Al_III sites by water, we next turned our attention to the catalytic behavior of hydroxylated alumina.

3.4. Hydroxylated γ-Al₂O₃ (110)

The hydroxylated Al₂O₃ model (Figure 1d) consisted of a chemisorbed H₂O water molecule on the Al_III-O_2a site pair. Out of the three available sites, namely Al_IVb, Al_IVa1, and Al_IVa2, THF binds the strongest on Al_IVb (the respective binding energies are −1.86, −1.3, and −1.11 eV). The THF ring opening is more favorable on Al_IVb-O_3b than on Al_IVb-O_2b or Al_IVb-OH_terminal (Figure S8). Hence, we investigated the reaction on the Al_IVb-O_3b site.

Scheme 5 shows the important features of the most dominant pathways for butadiene (Pathway 1_Alh) and retro-Prins (Pathway 2_Alh) (details in Section S5 in the Supplementary Materials). Pathway 1_Alh presents some significant differences from the pathways on ZrO₂ (dry and hydrous) and on dry Al₂O₃ described so far. First, after the ring opening, it is more favorable for the γ-proton to be abstracted from the distal surface oxygen O_3b (the O-site coordinating the C5 carbon of 2_Alh) than from the neighboring O_3a2; on ZrO₂ (dry and hydrous) and dry Al₂O₃, the γ-proton is abstracted by a neighboring surface O-site. The distal O_3b is involved because when Al_III-O_2a is hydroxylated, Al_IVb-O_3b is the most favorable site pair to bind the butenoxide-H⁺ pair (see Table S10), and a similar observation is also found in the case of CH₄ dissociation to CH₃—H⁺, where Al_IVb-O_3b is the most favorable site pair [31]. Second, unlike ZrO₂ and dry Al₂O₃, on the hydroxylated Al₂O₃ surface the butenoxide intermediate (3_Alh) undergoes E2 elimination instead of E1cB elimination due to the absence of a vicinal undercoordinated Lewis-acidic metal site that would stabilize the excess charge on the resulting carbanion (vide infra). The E2 elimination in 3_Alh entails simultaneous C-O bond cleavage and β-proton elimination to form 4_Alh (physisorbed butadiene and chemisorbed H₂O). Pathway 2_Alh is the dominant retro-Prins pathway. The free energy profiles of pathways 1_Alh and 2_Alh are shown in Figure 4. There are two additional less important retro-Prins pathways (Pathways 3_Alh and 4_Alh) whose mechanisms and energy profiles are shown in Section S5. Some additional minor differences between Pathway 1_Alh and butadiene formation pathways on the other surfaces are also discussed in Section S5. Figure S9 shows the free energy and enthalpy profiles of all the four pathways. Similar to the dry Al₂O₃ model, THF and HCHO strongly chemisorb on the surface, and H₂O chemisorbs in both undissociated and dissociated forms. Additionally, the binding strengths of THF, HCHO, and dissociated H₂O species are higher than those on ZrO₂ models and lower than those on dry Al₂O₃.

Unlike the reaction free energy profiles on ZrO₂ (dry and hydrous) and dry Al₂O₃, where the ring opening presents the higher transition state (cf. Figures S1, S4 and S6), it is the γ-proton elimination that presents the higher transition state (TS3_Alh) on hydrous ZrO₂, suggesting a change in the rate-determining step. These observations are confirmed by microkinetic analysis, presented next.

The microkinetic analysis of reaction fluxes (Table S11) further confirms that Pathway 2_Alh is the most favorable pathway; the flux percentages through pathways 1_Alh, 2_Alh, and 3_Alh are 28%, 51.4%, and 20% respectively. At low conversions, propene is more selective than butadiene (Figure 5a), which is in qualitative agreement with the experimental selectivity on Al₂O₃ [10]. As the conversion increases, thermodynamics takes over and the selectivity for butadiene gradually increases. At ~15% conversion, the selectivities for butadiene and propene are the same. At conversion close to the equilibrium value at 673 K (99.96%), S_butadiene and S_propene approach their respective equilibrium values of 91.1% and 8.9%, respectively (Section S6 of Supplementary Materials). Surface coverage analysis (Figure 5b) shows that at low conversions, the ring opened intermediate (2_Alh) is the most abundant one due to the slowest rate of γ-proton elimination. As the conversion increases, S_butadiene and the fraction of chemisorbed H₂O (5_Alh) increase.

Sensitivity analysis (Figure 6a) further confirms that the γ-proton elimination (TS3_Alh) is rate-limiting for both butadiene and propene formation, as suggested by the free energy profiles. The slowest rate for γ-proton elimination is likely to be responsible for the high surface coverage of ring-opened intermediate (2_Alh) at low conversions (Figure 6b). Here, we should recall that when we compared the ring opening barriers on Al_IVb-O_3b, Al_IVb-OH_terminal, and Al_IVb-O_2b sites (to select the most favorable pathway for the mechanistic investigation (Figure S8)) we assumed that ring opening is the RDS. Since the γ-proton elimination was found to be the RDS, we revisit the comparison made in Figure S8 to make sure that the choice of the Al_IVb-O_3b site for the mechanistic investigation is still meaningful. In Figure S11, we compare the energy of TS3_Alh with the TS energies of ring opening on Al_IVb-OH_terminal and Al_IVb-O_2b (TS_OH and TS_O2b, respectively). Figure S11 confirms that the choice of Al_IVb-O_3b site for mechanistic investigation is meaningful since the energy of TS_3Alh is lower than the energies of TS_OH and TS_O2b.

Returning to the sensitivity analysis, in addition to TS3_Alh, the rate of propene formation is enhanced by the retro-Prins steps on pathways 2_Alh (TS9_Alh) and 3_Alh (TS4_Alh, see Figure S9) and hampered by the E2 elimination step (TS7_Alh). Figure 6b shows that S_propene is mainly enhanced and hampered by retro-Prins (TS9_Alh) and E2 elimination steps (TS7_Alh), respectively. In contrast, the rate and selectivity of butadiene formation are mainly enhanced and hampered by E2 elimination (TS7_Alh) and retro-Prins steps (TS9_Alh), respectively.

4. Discussion

Based on the results obtained so far, we observe that the difference in product selectivities between hydroxylated Al₂O₃ and the rest of the surfaces can mainly be attributed to the difference in the free energy barriers of the retro-Prins and dehydration/elimination steps that branch off at the butenoxide intermediate. Therefore, analyzing the variations in energy barriers of these specific dehydration and retro-Prins steps will help us explain the surface-dependent product selectivity.

On the surfaces (dry Al₂O₃, dry and hydroxylated ZrO₂) that favor the formation of butadiene, the dehydration of the butenoxide intermediate is E1cB, whereas on hydroxylated Al₂O₃, the dehydration is E2. In E1cB, the β-proton of the butenoxide intermediate is abstracted by a basic O-site on the surface and the corresponding carbanion is stabilized by the surface metal site (intermediates 4_Zrd, 4_Zrh, and 4_Ald in Scheme 2, Scheme 3 and Scheme 4 respectively); the position of the double bond shifts from C4-C5 to C3-C4, and the C5 carbon forms a bond with the surface metal site. This surface-stabilized intermediate further undergoes facile C-O bond cleavage (intrinsic barrier in the range of 0.3–0.6 eV) to butadiene and chemisorbed H₂O. However, on hydroxylated Al₂O₃, there is no proximal undercoordinated metal site to stabilize the E1cB intermediate, hence the elimination is E2—the β-C-H and C-O bonds are broken simultaneously. The free energy barriers for E1cB are in the range of 1–1.2 eV and lower than the E2 free energy barrier on hydroxylated Al₂O₃ (1.5 eV).

We believe that the E1cB barriers are lower than the E2 barriers because the former transition states are stabilized by the interaction between the C5 carbon and a surface metal site, which is not the case in E2. In the E1cB transition state, the negative charge on C3 delocalizes over the allylic C4-C5 double bond while the charge on C5 is stabilized by the proximal Lewis-acidic metal site. E2 is favored on hydroxylated Al₂O₃ because there is not a proximal surface metal site that could interact with the C5 atom. Charge density difference (CDD) and electron localization function (ELF) analyses indicate an ionic interaction between C5 and a proximal Lewis-acidic metal site, as discussed in Section S7 in the Supplementary Materials.

Despite the absence of a neighboring Lewis-acidic metal site on hydroxylated Al₂O₃, we tested the β-proton elimination of the E1cB mechanism by stabilizing the resulting carbanion on a surface proton (which originated from the γ-proton elimination) (Figure S12a). The barrier for this step is very high compared to the barrier for the E2 step (Figure S12b), implying that the TS of the β-proton elimination is likely to be charge stabilized only by a Lewis-acidic metal site and not by any other electrophilic site.

Regarding the retro-Prins products, the intrinsic barriers for propene formation from the butenoxide intermediate vary across the investigated surfaces which can be ranked as follows: hydroxylated Al₂O₃ < hydroxylated ZrO₂ < dry ZrO₂ < dry Al₂O₃. It is not straightforward to rationalize the differences in these energy barriers because the retro-Prins step from the butenoxide intermediate is a concerted mechanism and involves multiple chemical and geometric factors: C2-C3 bond cleavage, C5-H bond formation, O_surface-H bond cleavage, O1-M bond cleavage, C2=O1 bond formation, the dihedral angle between H, C5, C3, and C2, etc. By merely using the limited data we have in this study, it is challenging to relate those factors to the energy of the retro-Prins TS.

Therefore, a higher barrier for butadiene formation, due to the lack of a proximal Lewis-acid metal site, and a relatively lower barrier for retro-Prins reaction make hydroxylated Al₂O₃ more selective towards propene at low conversions. The local topology around the active site plays a crucial role in determining the product selectivities by deciding the dominant reaction channel.

5. Conclusions

We have used density functional theory (DFT) and microkinetic modeling to understand the differences in selectivity for products obtained from the dehydra-decyclization of THF on ZrO₂ and Al₂O₃. We use t-ZrO₂ (101) and γ-Al₂O₃ (110) models to investigate the reaction mechanism and investigate the effect of H₂O on the reaction mechanism.

On the dry surface of t-ZrO₂ (101), the selectivity of butadiene is close to 100% which is in good agreement with the experiment. Ring opening is the rate-determining step, and the retro-Prins products are predominantly formed by a concerted C-C bond cleavage and hydride shift of the ring opened intermediate. Chemisorbed H₂O is the most abundant surface species. The OH-groups on the hydroxylated ZrO₂ surface neither enhance the reaction rate nor significantly affect the product selectivity. Despite insignificant differences in the rates and product selectivities between dry and hydroxylated ZrO₂, the pathway that predominantly contributes to the formation of propene is different on both surfaces; when surface OH-groups are involved, propene is predominantly formed from the butenoxide intermediate (3_Zrh).

The reaction behavior on hydroxylated Al₂O₃ differs greatly from the other investigated surfaces. At low conversions, the retro-Prins products are predominantly produced from the butenoxide intermediate (3_Alh) at 70% selectivity, which is in qualitative agreement with the experiments. Beyond 15% conversion, thermodynamics overtakes kinetics and the catalyst becomes more selective for butadiene. Moreover, on hydroxylated Al₂O₃, the γ-proton elimination is the rate-determining step.

The main difference in product selectivities between hydroxylated Al₂O₃ and the ZrO₂ (dry and hydrous) lies in the barriers for retro-Prins and elimination reactions of the butenoxide intermediate. On hydroxylated Al₂O₃ and ZrO₂, the butenoxide intermediate undergoes E2 and E1cB reactions, respectively. The presence of a neighboring Lewis-acidic metal site facilitates the E1cB mechanism, which is more facile than an E2, which is preferred when such Lewis acid centers are not available. We believe that the interaction between the terminal carbon (C5) and the surface Lewis-acidic metal site in the E1cB transition state makes this step more facile than E2, where the C5-metal interaction is lacking. The local topology around the reacting site surely greatly influences the reaction mechanism and thereby the selectivity.