Phenyl radical + propene: a prototypical reaction surface for aromatic-catalyzed 1,2-hydrogen-migration and subsequent resonance-stabilized radical formation †

The C 9 H 11 potential energy surface (PES) was experimentally and theoretically explored because it is a relatively simple, prototypical alkylaromatic radical system. Although the C 9 H 11 PES has already been extensively studied both experimentally (under single-collision and thermal conditions) and theoretically, new insights were made in this work by taking a new experimental approach: flash photolysis combined with time-resolved molecular beam mass spectrometry (MBMS) and visible laser absorbance. The C 9 H 11 PES was experimentally accessed by photolytic generation of the phenyl radical and subsequent reaction with excess propene (C 6 H 5 + C 3 H 6 ). The overall kinetics of C 6 H 5 + C 3 H 6 was measured using laser absorbance with high time-resolution from 300 to 700 K and was found to be in agreement with earlier measurements over a lower temperature range. Five major product channels of C 6 H 5 + C 3 H 6 were observed with MBMS at 600 and 700 K, four of which were expected: hydrogen (H)-abstraction (measured by the stable benzene, C 6 H 6 , product), methyl radical (CH 3 )-loss (styrene detected), H-loss (phenylpropene isomers detected) and radical adduct stabilization. The fifth, unexpected product observed was the benzyl radical, which was rationalized by the inclusion of a previously unreported pathway on the C 9 H 11 PES: aromatic-catalysed 1,2-H-migration and subsequent resonance stabilized radical (RSR, benzyl radical in this case) formation. The current theoretical understanding of the C 9 H 11 PES was supported (including the aromatic-catalyzed pathway) by quantitative comparisons between modelled and experimental MBMS results. At 700 K, the branching to styrene + CH 3 was 2–4 times greater than that of any other product channel, while benzyl radical + C 2 H 4 from the aromatic-catalyzed pathway accounted for B 10% of the branching. Single-collision conditions were also simulated on the updated PES to explain why previous crossed molecular beam experiments did not see evidence of the aromatic-catalyzed pathway. This experimentally validated knowledge of the C 9 H 11 PES was added to the database of the open-source Reaction Mechanism Generator (RMG), which was then used to generalize the findings on the C 9 H 11 PES to a slightly more complicated alkylaromatic system.


Introduction
The addition of the phenyl radical, C 6 H 5 , to the unsaturated CQC bond in propene, C 3 H 6 , produces two different propylbenzene radicals depending on the addition site, which are representative of a broader class of alkylaromatic radicals.As defined here, alkylaromatic radicals consist of an aromatic group, which can either be a single benzene ring or a polycyclic aromatic hydrocarbon (PAH) of arbitrary size, and at least one alkyl side chain of arbitrary length and branching on which the radical site resides.Alkylaromatic radicals can form either by ''bottom-up'' growth of smaller molecules, such as C 6 H 5 + C 3 H 6 , 1 or by ''top-down'' decomposition of larger molecules, such as in the combustion or pyrolysis of gasoline, diesel and This journal is © the Owner Societies 2018 jet fuels which contain significant amounts of closed-shell alkylaromatic compounds like propylbenzene. 2 Once formed, the subsequent decomposition, growth or oxidation of alkylaromatic radicals can dictate the extent of PAH and soot formation, with potentially deleterious effects on climate 3 and human health. 4Soot, or ''coke'', formation is also usually undesirable in industrial processes such as ethane steam cracking, 5 where fouling of the reactor can necessitate costly shutdowns for cleaning.Alkylaromatic linkages and radicals are also important in the chemistry of heavy oils, coal, kerogen, lignin and many polymers.Therefore, understanding all of the ways in which alkylaromatic radicals can react is critical to a number of diverse applications including combustion, industrial cracking, organic geochemistry, and utilization of waste materials and biomass.
The two propylbenzene radicals directly formed by C 6 H 5 + C 3 H 6 have served as relatively simple surrogates for more complicated alkylaromatic radicals in many top-down experimental and theoretical studies of propylbenzene oxidation and pyrolysis, recently summarized by Yuan et al., 2 and in nine bottom-up studies utilizing C 6 H 5 + C 3 H 6 (including this work) summarized in Table 1.From this large body of work on the C 9 H 11 potential energy surface (PES) a consensus has emerged (with some caveats discussed below) regarding the major expected reaction pathways of the C 6 H 5 + C 3 H 6 system, shown in Scheme 1. C 6 H 5 mostly reacts with C 3 H 6 by addition to the terminal and central CQC sites, and by hydrogen, H, abstraction from the allylic carbon in C 3 H 6 .The two propylbenzene radicals formed by addition can easily interconvert via a 1,2phenyl-migration, and are usually assumed to be equilibrated at combustion-relevant temperatures.The central-addition product will mostly b-scission a methyl radical, CH 3 , to form styrene, whereas the terminal-addition product will b-scission an H on one of the two neighbouring carbons to form 3-or 1phenylpropene.Alternatively, if the radical site on the terminal addition product can shift to the end of the alkyl chain via a 1,2-H-migration, the resulting alkylaromatic radical can either bscission an ethene, C 2 H 4 , to form a benzyl radical or it can undergo a secondary ring-closure to form a five-membered ring.Subsequent b-scission of an H from the bicyclic radical forms indane, which can form indene following loss of two more H atoms. Indene is a precursor to PAH and soot, and it is mostly because of this potential route to indene that C 6 H 5 + C 3 H 6 has been of such interest. 1However, the 1,2-H-migration that enables indane formation has a barrier at least 6 kcal mol À1 higher than the various b-scissions mentioned above, rendering that pathway uncompetitive, even under combustion conditions. 11The lack of indane formation from C 6 H 5 + C 3 H 6 and the dominance of styrene and phenylpropene isomers as products is supported both by calculations of the product and B30 kcal mol À1 (ref.8) based on both the measured exoergicity of the H-loss channel and from experiments with deuterated C 3 H 6 .The thermal experiment, conducted in a fast pyrolysis reactor maintained at 1200-1500 K in 300 Torr of C 3 H 6 , ruled out indane formation based on the measured photoionization efficiency (PIE) curves of the product(s) at a mass-to-charge ratio, m/z, of 118 amu, corresponding to the mass of the various H-loss products with the molecular formula C 9 H 10 .
Although there is qualitative agreement in the literature regarding what the major products of C 6 H 5 + C 3 H 6 should (styrene and phenylpropene isomers) and should not (indane) be, quantitatively there are three discrepancies.First, the extent of H-abstraction is predicted to be significant, increasing from B10% of the product branching at 300 K to B80% at 3000 K, 11 but this has never been quantified experimentally.Of the six previous experimental studies listed in Table 1, only the photolysis experiments of Hefter et al. in liquid C 3 H 6 reported measurable products of H-abstraction: benzene, C 6 H 6 , and allyl radical, the most thermodynamically stable of the C 3 H 5 isomers. 6 that work, C 6 H 5 was found to prefer radical addition to the terminal CQC carbon in propene over the other two major entrance reactions (Scheme 1), consistent with the low temperatures used (182-213 K).Of the two remaining experiments conducted in a thermal reactor, the fast pyrolysis experiment of Zhang et al. was done under aggressive conditions that would encourage many side, secondary, tertiary and higher reactions, 10 making it impossible to extract quantitative information about the underlying chemistry without a detailed kinetic mechanism.The Cavity Ring Down Spectrometry (CRDS) experiments of Park et al. were conducted under more controlled conditions, and the probe laser wavelength used, 504.8 nm, was tuned to a known absorbance feature of C 6 H 5 , 14 where few other radicals absorb, enabling selective detection of C 6 H 5 . 7While convenient for accurate quantification of overall consumption rates, selective detection of a photolytically generated radical reactant (C 6 H 5 in this case) leaves the experimentalist blind to subsequent product formation, which was the case in Park et al. 7 Finally, all three of the CMB experiments with QMS detection were unable to detect either C 3 H 5 or C 6 H 6 , likely due to lower sensitivity for light species and overlap with the large 13 C peak of scattered C 6 H 5 , 12 respectively.
The second quantitative discrepancy is the extent of CH 3 -loss relative to H-loss.Kaiser et al. computed relative product yields under single-collision conditions by combining G3(MP2,CC)//B3LYP calculations of the C 9 H 11 PES with microcanonical Rice-Ramsperger-Kassel-Marcus (RRKM) theory for energy-dependent rate coefficients, k(E), and then applying the steady-state approximation to the set of differential equations describing the reaction network. 9This calculation does not include the contribution of H-abstraction because as noted above it was not possible to observe the products of H-abstraction in the CMB experiments.The single-collision product calculations of Kaiser et al. were quantitatively consistent with CMB experiments reported in that same work at E col E 10 kcal mol À1 , where yields of CH 3 -and H-loss products were measured as B70% and B30%, respectively.At higher E col it was predicted that the yield of CH 3 -loss would drop steadily to B20% at B45 kcal mol À1 to be replaced by various H-loss products (mostly 3-phenylpropene).This prediction was also consistent with earlier CMB experiments at E col 4 30 kcal mol À1 , where only H-loss products were observed, and the CH 3 -loss yield was approximated as o10% due to the detection limit. 8However, Albert et al. measured a much higher relative product yield of CH 3 -to H-loss at intermediate E col 's of B20 and 25 kcal mol À1 than expected based on the calculations of Kaiser et al.: B10 : 1 and B3 : 1, respectively. 12ne proposed explanation for the experimental discrepancy between Albert et al. and Kaiser et al. is that the latter used 80 eV electron impact ionization, necessitating the deconvolution of fragments, whereas the former used 9.9 eV ''soft'' vacuum ultraviolet (VUV) photoionization (PI).Furthermore, the G3(MP2,CC)//B3LYP C 9 H 11 PES first reported by Kaiser et al. was later combined with RRKM theory to solve the master equation (ME) describing temperature-and pressure-dependent kinetics under thermal conditions. 11These thermal calculations predict similar yields of CH 3 and H-loss products from 1200 to 1500 K and at 300 Torr, whereas Zhang et al. measured B5Â more styrene from CH 3 -loss than the sum of all H-loss products under the same conditions.As mentioned earlier, however, this discrepancy could be due to secondary chemistry occurring during the pyrolysis.
Finally, there is disagreement regarding the identity of the dominant H-loss isomer.While there is wide agreement that styrene is the dominant CH 3 -loss product, there are conflicting reports in the literature as to whether 3-phenylpropene or 1-phenylpropene dominates H-loss.Both the single-collision and thermal calculations of Kaiser et al. 9 and Kislov et al., 11 respectively, predict that 3-phenylpropene will have the highest yield of any H-loss product by far over all conditions considered.This is in nominal agreement with the PIE curves measured at m/z = 118 amu by Zhang et al., which could be fit by assuming almost 100% 3-phenylpropene, but given the number of potential 118 amu isomers (indane, both cis-and trans-1-phenylpropene, 2-phenylpropene and 3-phenylpropene) this fit is best considered as just one possible solution. 10In contrast, the earlier CMB experiments of Zhang et al. identified 1-phenylpropene as the dominant H-loss isomer at E col E 45 kcal mol À1 based on experiments with deuterated isotopologues of C 3 H 6 . 8Albert et al. also tentatively assigned the H-loss product observed in B20-25 kcal mol À1 CMB experiments to 1-phenylpropene, based on analogy to the products observed from C 6 H 5 + 2-butene, 2-C 4 H 8 , in the same work. 12his work aims to alleviate the first two of the three discrepancies summarized above (extent of H-abstraction and the extent of CH 3 -to H-loss) by applying a different detection technique to the problem: time-resolved molecular beam MS (MBMS) with VUV PI.The reaction conditions used here are most similar to those in the work of Park et al. 7 (photolytically generated C 6 H 5 in a thermal environment), but the use of timeresolved MBMS for detection allows for quantification of all products as they are formed, including C 6 H 6 from H-abstraction.Additionally, unlike the fast pyrolysis experiments of Zhang, which also used MBMS under thermal conditions, the primary products of C 6 H 5 + C 3 H 6 can be easily distinguished from later generation products based on their faster appearance, enabling more conclusive quantification of CH 3 -to H-loss as a function of temperature and pressure.The experimental apparatus used in this work is also equipped to perform multiple-pass laser absorbance measurements, allowing for precise quantification of the overall C 6 H 5 consumption rate, which is compared to the CRDS measurements of Park et al. 7 Finally, benzyl radical was observed as an unexpected primary product of C 6 H 5 + C 3 H 6 .This finding was rationalized by the inclusion of a new ''aromatic-catalyzed'' 1,2-hydrogen-migration pathway on the C 9 H 11 PES.All experimental measurements were quantitatively compared to the predictions of the Reaction Mechanism Generator (RMG), which was then used to predict the products of a slightly more complicated alkylaromatic system, 1-naphthyl + 2-butene, in order to demonstrate the general applicability of the insights made into the C 6 H 5 + C 3 H 6 system.

Experimental
A previous version of the experimental apparatus has been described before, 15,16 and the latest version will be described in detail in a forthcoming publication. 17A brief description is provided here, with an emphasis on two major changes made to the apparatus since the last publication.
The apparatus consists of a custom quartz flow-reactor housed in a high vacuum chamber equipped with a time-offlight MS (TOF-MS).Switching from a stainless steel reactor with B6 cm inner diameter to a more compact quartz reactor described below was the first major change to the apparatus.The new reactor is 1 m long with a 16 mm inner diameter (2.5 mm wall thickness) in the central 0.4 m and a 36 mm inner diameter (2 mm wall thickness) everywhere else.Gases are mixed upstream and pumped through the reactor by a Roots blower.Pressure is controlled in the reactor by throttling a butterfly valve at the exit.The contents of the flow reactor are coaxially flashed by a collimated 266 nm laser beam (fourth harmonic frequency of an Nd:YAG laser) at a set repetition rate (usually 1 Hz).The total gas flow through the reactor is set such that one flash per refresh (FPR) conditions are maintained.The photolysis beam is expanded by a set of telescoping lenses and clipped by an adjustable iris to the same diameter as the narrow section of the reactor (16 mm).This approach to setting the photolysis beam diameter (expand and then clip) is intended to reduce radial inhomogeneities at the edges of the beam.A ''funnel-shaped'' pinhole, similar to that used by Wyatt et al., 18 is drilled in the center of the reactor with a 275 mm diameter at the narrowest point.A small portion of the reactor contents are sampled through the pinhole, and the center of the resulting gas expansion is skimmed by a Beam Dynamics skimmer (Model 16.3, 1.0 mm orifice diameter), forming a molecular beam.After traversing B50 mm, the molecular beam intersects with a focused 118 nm (10.5 eV) PI laser beam (ninth harmonic frequency of an Nd:YAG laser).Cations formed in the ionization region are accelerated, focused, and guided to the detector of the Kore TOF-MS (ETP electron multiplier, model AF824) by a set of ion optics.
The second major change to the apparatus was to separate the residual 355 nm radiation from the desired 118 nm PI radiation by an off-axis MgF 2 lens.Once the two beams have been dispersed, only the 118 nm beam is allowed to enter the ionization region by passing through a pinhole, while the 355 nm beam is blocked.Preventing the excess 355 nm radiation from reaching the ionization region was necessary to avoid excessive fragmentation and multiphoton ionization.The approach to separating the two wavelengths described here (essentially using the MgF 2 lens as a prism) was adapted from Tonokura et al. 19 As mentioned earlier, the apparatus is also equipped to do multiple-pass UV-visible laser absorbance, and single-pass IR absorbance of the iodine, I, atom.Details of this portion of the apparatus have been given numerous times before, 20,21 and will not be repeated here.Although the quartz reactor is much narrower than the previous reactor, it was designed to accommodate the multiple-pass Herriott cell, hence the custom ''bow-tie'' geometry of the reactor. 17reviously, with the larger diameter, stainless steel reactor, the radicals formed by photolysis were concentrated in a relatively small central section of the reactor (B1.3 cm diameter), and would diffuse outwards radially on a timescale commensurate with the kinetics of interest (B10 ms at B10 Torr).With the new reactor, by adjusting the photolysis beam diameter to match the inner diameter of the central 0.4 m, radial diffusion out of the MBMS sampling region can be eliminated as a loss process.The longest kinetic time-scales that can be measured with MBMS is now set by the time it takes for the gas to flow through the narrow, central section (B50 ms for a 1 Hz experiment).
Switching the reactor material from stainless steel to quartz was also intended to reduce wall reactions that were most clearly manifested by fast hydrogenation of C 6 H 5 to C 6 H 6 .However, even after treating the quartz reactor with boric acid following the procedure of Krasnoperov et al., 22 this manifestation of wall reactions was never entirely eliminated (Fig. S9, S18, S29 and S33, ESI †).Nonetheless, control experiments described in the Results and discussion section demonstrate that wall reactions do not noticeably affect the main observations and conclusions made in this work.
The reactor is heated by nichrome ribbon wire in two temperature-controlled zones.There is a short, 5 cm ''pre-heat zone'' just upstream of where the Herriott cell and the photolysis laser overlap.The gas entering this zone is still at room temperature and is quickly heated to the desired temperature.The following ''reaction zone'' encompasses almost the entire remaining length of the reactor and maintains near isothermal conditions at the desired temperature.The thermocouples used to control the power supplied to each heater are located inside the reactor at the exit of each zone (out of the path of the photolysis laser).Axial temperature profiles were measured under different experimental conditions and are reported in the Results and discussion section.For the absorbance experiments, the temperature reported is the average T(z) over the central 0.5 m, where the Herriott cell and photolysis laser overlap.For the MBMS experiments, the temperature reported is the average in the 0.2 m narrow section on the inlet side of the sampling pinhole.Uncertainties reported are two standard deviations over the same axial range.
Both iodobenzene (C 6 H 5 I; Sigma-Aldrich, 98%) and nitrosobenzene (C 6 H 5 NO; Sigma-Aldrich, Z97%) were used as photolytic precursors of C 6 H 5 because both are known to almost entirely photodissociate to C 6 H 5 + I/NO at 266 nm. 23,24Iodobenzene was purged of oxygen by successive freeze-pump-thaw cycles, while nitrosobenzene, a powder at room temperature, was simply degassed.Both precursors were placed in air-tight bubblers and introduced into the reactor by flowing a controlled amount of helium through/over them.Because of the strong reactivity of C 6 H 5 towards O 2 , 14 even B100 ppm of residual air in the reactor could have a measurable impact on the experiments.The clearest indicators of O 2 present somewhere in the reactor or associated gas lines are a strong visible absorbance signal from the phenylperoxy radical, C 6 H 5 OO, 25 and a product at m/z = 94 amu (likely phenol, C 6 H 5 OH, also seen by Zhang et al. 10 ) in the MS following photolytic C 6 H 5 generation.When C 6 H 5 I was used as the precursor, it was possible to eliminate all signs of O 2 from the experiment.With C 6 H 5 NO, however, there was always some residual O 2 in the precursor, manifested by both C 6 H 5 OO absorbance and C 6 H 5 OH MS signal.C 6 H 5 I can also be used up to 900 K before thermal decomposition on the residence time scale of this experiment (1 s) becomes important, 26 whereas for C 6 H 5 NO the limit is only 700 K. 24 The I atom from C 6 H 5 I photolysis can also be quantified by IR absorbance, providing an estimate of the initial C 6 H 5 concentration.For all of these reasons, C 6 H 5 I was the precursor of choice for almost all of the experiments reported here.
Helium, He, was used as the bath gas in all experiments reported here and was purchased from Airgas (UHP grade, 99.999%).C 3 H 6 was also purchased from Airgas (CP grade, 99%) and used without further purification.

Theoretical
Much of the C 9 H 11 PES has already been explored both by Park et al. using G2M(RCC6)//B3LYP 7 and by Kislov et al. using G3(MP2,CC)//B3LYP. 11Wang et al. reported CBS-QB3 calculations on a portion of the PES, focused on the 1,2-phenyl-migration connecting the terminal and central addition products. 13This work took the C 9 H 11 PES calculated by Kislov et al., and added on one new C 9 H 11 isomer, three new transition states (TSs) and one new bimolecular product channel (benzyl radical, C 7 H 7 , + ethene, C 2 H 4 ) that together comprise the newly discovered aromaticcatalyzed 1,2-H-migration and subsequent resonance stabilized radical (RSR) formation pathway.For consistency, the same level of theory was used.Briefly, geometries and frequencies of the new minima and saddle points were optimized using B3LYP/ 6-311G(d,p).Starting from the optimized geometries, relaxed potential energy scans were performed (also using B3LYP/ 6-311G(d,p)) as a function of the dihedral angle in 101 increments around all C-C and nascent C-C bonds that are likely to undergo internal rotation for all stationary points on the C 9 H 11 PES.In some cases the scan revealed that the starting geometry was not the conformational minimum and the geometry was re-optimized accordingly.The 0 K electronic energies of the optimized geometries were refined by using a simplified version of the composite G3(MP2,CC) method: The first term on the right-hand side is the ground state electronic energy calculated with a high level of theory (spinrestricted coupled cluster) and a small basis set (6-311G(d,p)).
The second term is an estimation of the error introduced by using a small basis set instead of a larger one (G3large, in this case).The last term is the zero point energy (ZPE) correction, calculated based on the B3LYP frequencies scaled by a recommended factor of 0.967. 27The G3(MP2,CC)//B3LYP method has been applied to many aromatic-containing PESs because it offers a good compromise between accuracy and computational cost. 1,28Gaussian03 29 was used for all B3LYP calculations, and Molpro 30 was used for all coupled cluster calculations.The geometries, frequencies, rotational constants, dihedral angle scans and electronic energies for all stationary points on the C 9 H 11 PES are reported in the ESI.† Both high-pressure (P) and P-dependent thermal rate coefficients, k(T) and k(T,P), were calculated on the G3(MP2,CC)//B3LYP C 9 H 11 PES using Cantherm, an open-source software included with RMG. 31 k(T)'s for every elementary chemical reaction were calculated by canonical transition state theory (TST) using the rigid-rotor harmonic oscillator (RRHO) approximation with low-energy (o10 kcal mol À1 barrier) internal rotations modelled as onedimensional hindered rotors (1D-HR).Tunnelling was modelled using the 1D-asymmetric Eckart correction. 32Phenomenological k(T,P)'s were fit by applying RRKM/ME under the modified strong collision (MSC) approximation (RRHO and 1D-HR approximations also used in calculating densities of states). 33or the two barrierless exit channels found by Kislov et al., the high-P variational TST k(T)'s calculated in that work were input to Cantherm and converted into k(E)'s for solution of the ME by inverse Laplace transform. 33Lennard-Jones and collisional energy-transfer (CET) parameters were taken from Mebel et al. for an argon bath gas, 1 and scaled down for a helium bath gas using the calculations of Jasper et al. 34 The same custom scripts used in Kaiser et al. 9 for simulating single-collision experiments on the C 9 H 11 PES were also used here on the expanded PES.k(E)'s were calculated using RRKM at a fixed E corresponding to E col .Using the calculated k(E) values, a system of ordinary differential equations (ODEs) was constructed that describes the connectivity of the PES.The initial composition was set to either 100% of the terminal or 100% of the central addition adduct.The ODEs were solved under the steady-state approximation in order to determine predicted product branching as a function of E col for comparison to CMB experiments.The product branching for the two different initial adducts were very similar, and the total product branching was estimated by summing the two sets of calculations weighted by the expected thermal branching between terminal and central addition.The temperature used to evaluate the thermal branching was calculated as 2/3 Â E col /R, or the temperature at which the average kinetic energy in a Boltzmann distribution is E col .
Finally, all elementary reactions on the C 9 H 11 PES were added as training reactions to the RMG database. 31 the RMG database is currently divided into B50 reaction families common to gas-phase combustion chemistry, such as intramolecular H-migration (abbreviated as intra_H_migration).Each family uses reaction templates to describe the critical details for a given reaction.In the example given, the reaction template specifies a 1-4-H-migration from a carbon in a benzene ring (identified by the atom label ''Cb'') to a secondary carbon radical.The template can also provide information about atoms not directly involved in the reaction.For example, one of the atoms between the two reacting ends is another Cb, and one of the atoms bonded to the radical carbon is a Cs (a carbon with only single bonds, i.e., an sp 3 carbon).Such details can have large impacts on the real kinetics, so it is critical for RMG to be aware of these details to make good estimates for reactions with unknown kinetics.Training reactions are the most specific representations of a reaction in the RMG database.When RMG proposes a reaction with unknown kinetics during automatic mechanism generation, it will match the new reaction to similar training reactions based on their similar reaction templates, utilizing the kinetic parameters of the training reactions to estimate parameters for the new reaction.For example, this work is primarily concerned with C 6 H 5 + C 3 H 6 , but any arbitrary alkylaromatic radical system, such as 1-naphthyl + 2-butene, 1-C 10 H 7 + 2-C 4 H 8 , will undergo many of the same reactions with similar kinetics and heats of reaction.By training the RMG database on reliable kinetics for the C 9 H 11 system, the knowledge gleaned from that specific system will automatically be applied to every analogous alkylaromatic system, including 1-C 10 H 7 + 2-C 4 H 8 .This analogy is explored in more detail in Section 4.6.The aromatic-catalyzed 1,2-H-migration proceeds in two steps.First i1 undergoes a 1,4-H-migration, transferring one of the H's on an ortho-carbon to the secondary carbon radical on the alkyl side-chain via a 5-membered-ring TS (Fig. 1).The barrier for this first, rate-limiting step in the aromatic-catalyzed 1,2-H-migration is energetically similar to the barrier for b-scission of CH 3 from i2 (B30 kcal mol À1 ), which is known to be a major loss channel for equilibrated i1/i2.The new intermediate formed (i12) has a radical site on an orthocarbon and a fully saturated propyl side-chain.i12 can then undergo a 1,5-H-migration, transferring one of the H's on the primary carbon at the end of the alkyl side-chain back to the ortho-carbon via a low barrier (B8 kcal mol À1 ) 6-memberedring TS (Fig. 1) forming i4.To summarize, an overall reaction (i1 -i4) that has a high barrier as a one-step process (1,2-Hshift) is more energetically feasible as a two-step process (1,4-Hshift + 1,5-H-shift) enabled by the presence of a catalyst that is unaltered at the end of the process (the phenyl side group).Hence, the two-step i1-i12 -i4 process was given the name ''aromatic-catalyzed'' 1,2-H-shift in acknowledgement of the central role of the aromatic side group (phenyl in this case) as both a source and a sink of the H-atom, much like a conventional Bronsted acid catalyst that acts as both source and sink of protons.This pathway is accessible to many radicals containing aromatic-groups as long as there is an H on at least one orthocarbon site.The second step is fast if there is at least one H on a carbon in the chain in a b, g or d position relative to the aromatic group.

Results and discussion
Given the favourable energetics of the previously unreported aromatic-catalyzed 1,2-H-shift, it was deemed worthwhile to revisit calculations of the C 6 H 5 + C 3 H 6 product branching under both thermal (T, P, see Fig. 2) and single-collision (E, see Fig. 4) conditions, and to re-assess previous experimental results in this new context.
The thermal product branching was calculated by simulating the network of Cantherm-generated phenomenological k(T,P)'s on the C 9 H 11 PES in an isothermal, isobaric, homogeneous batch reactor in CHEMKIN 35   5 Â 10 16 cm À3 , respectively, for all T, P conditions, with the helium bath gas providing the balance of the gas density.These specific reaction conditions were chosen to closely simulate the conditions of the experiment reported in this work in order to facilitate comparison.The k(T,P)'s taken from Cantherm include both thermally activated, well-to-well rates (i1 -i3, for example) and chemically activated, well-skipping rates (i1 -p10 + C 2 H 4 , for example).When simulated together, the combined thermally and chemically activated k(T,P)'s provide a complete picture of the initial chemical evolution of the thermal system, neglecting side and secondary reactions not on the same PES (considered later).
Qualitatively, the thermal product branching shown in Fig. 2 is very similar to that calculated by Kislov et al., 11 albeit their definition of product branching is slightly different.At low-T stabilized i1 dominates the product distribution, but from 600 to 1000 K styrene, p1, becomes the prominent product.With a further increase in T, branching to p1 gradually decreases to be replaced mostly with C 6 H 6 from the three H-abstraction channels.Unlike the predictions of Kislov et al., a peak branching of B10% benzyl radical, p10, is predicted at B600 K and like styrene diminishes with T. Given the chemical complexity of previous thermal experiments probing C 6 H 5 + C 3 H 6 , 6,10 the lack of observable benzyl radical in those works could easily be attributed to side reactions.However, in the product mass spectrum reported by Zhang et al. there is a peak at m/z = 91 amu (mass of the benzyl radical) though it is not discussed. 10lso unlike Kislov et al., the predicted pressure dependence from 10 to 1000 Torr over the whole T-range is negligible.This prediction was not sensitive to the choice of CET parameters or the method used to solve the ME. 33The discrepancy with Kislov et al. is likely due to the difference in the definition of product branching, specifically the inclusion of both thermally and chemically activated pathways in the simulation.The predicted lack of P-dependence is shown to be consistent with experiments in Section 4.3.
Finally, although the branching to indane, p5, is about an order of magnitude higher than in Kislov et al. due to the aromatic-catalyzed 1,2-H-shift promoting isomerization to i4, the overall amount of p5 formed is still small (B1% peak), consistent with the experiments of Zhang et al. 10 Overall, the predicted thermal product branching reported here does not significantly alter the current understanding of C 6 H 5 + C 3 H 6 (styrene, phenylpropene isomers and benzene are all still major products) other than the inclusion of a non-negligible, pressure-independent B10% yield of benzyl radical and C 2 H 4 .
An approach identical to Kaiser et al. 9 for quantifying product branching under single-collision conditions was taken here and has already been described in the Theoretical section.Just as in the thermal case, the single-collision product branching shown in Fig. 4 is very similar to what has previously been reported, this time by Kaiser et al., with some important exceptions.In terms of similarities, p1 dominates at low E col and is gradually replaced by p2 with increasing E col .This was the same qualitative observation made by all three CMB experiments regarding the effect of E col on the CH 3 -loss to H-loss ratio. 8,9,12One main difference is that there is up to B15% branching to the benzyl radical, p10, at the lowest E col , followed by a steady decline.Considering that all of the CMB experiments were conducted at E col Z 10 kcal mol À1 , any p10 formed in those experiments was likely below the detection limit.The other important difference is an increased yield of p5 (up to 3%), but that would have also been too small to detect in the CMB experiments, making it consistent with the observations made therein. 8,9o conclude this section, although the newly discovered aromatic-catalyzed 1,2-H-migration is predicted to be both energetically and kinetically competitive with the other reactions on the C 9 H 11 PES available to C 6 H 5 + C 3 H 6 , none of the six previously published experiments on this system have seen clear evidence for it.In the case of the thermal experiments this is either because of complex chemistry with side and secondary reactions, 6,10 or because the detection technique was not intended for product measurements, as in the CRDS experiments of Park et al. 7 In the case of the CMB experiments, detection sensitivity is the issue.Most of the remainder of this work will present experimental measurements coupled with detailed modelling that together quantitatively support the predictions above.

Overall k(T) measured by 505.3 nm absorbance
In addition to MBMS and laser absorbance experiments to quantify product branching, discussed in the following sections, 505.3 nm absorbance of C 6 H 5 was also used to measure the total consumption rate coefficient, k total (T), of C 6 H 5 .These measurements of k total (T) are presented first because later results depend on them.
C 6 H 5 is known to have a low-lying electronic transition at 504.8 nm at room temperature, 14 but in this work 505.3 nm was used to probe C 6 H 5 under all T, P conditions due to slightly higher absorbance observed at that wavelength.The other radicals formed in the C 6 H 5 + C 3 H 6 system (CH 3 , 36 C 3 H 5 37 and C 7 H 7 38 ) are not expected to absorb significantly at such a relatively long wavelength.Therefore, C 6 H 5 can be probed selectively at 505.3 nm.However, as mentioned earlier, if O 2 is present in the reactor at a similar concentration as C 6 H 5 , C 6 H 5 OO will also absorb strongly and broadly at B500 nm, 25 providing a convenient leak indicator.k total (T) consists of contributions from five different C 6 H 5 + C 3 H 6 entrance channels (two radical additions and three H-abstractions): From the calculations of the barrier heights for these five elementary reactions (Fig. 3) it is expected that terminal radical addition to form i1 (1.4 kcal mol À1 barrier) will dominate C 6 H 5 consumption, followed by central addition to i2 (2.6 kcal mol À1 ) and H-abstraction from the allylic carbon (3.6 kcal mol À1 ).The other two H-abstractions have higher barriers (6.4 and 8.4 kcal mol À1 ) and are predicted to be negligible below 1000 K.
Background absorbance traces are recorded with He only (or He and C 3 H 6 only) following photolysis to account for spurious absorbance, which is likely due to heating of the Herriott cell optics from scattered photolysis light and subsequent steering of the probe laser beam slightly off of the signal photodiode.The background absorbance was subtracted from absorbance traces recorded with C 6 H 5 present.
Each background-subtracted and normalized 505.3 nm absorbance trace was fit to an exponential decay assuming pseudo-first-order kinetics: a is a vertical shift factor to account for slight offsets (usually AE10%) in the baseline due to noise, imperfect background subtraction and artifacts from the AC-coupled detection electronics.k 0 is the pseudo-first-order decay rate, which consists of a contribution from C 6 H 5 + C 3 H 6 and from k wall .k wall is a fit parameter that accounts for all of the other C 6 H 5 loss processes (wall reaction, self-reaction, reaction with impurities like O 2 , reaction with the precursor, transport out of the sampling volume, etc.).Fig. 5 shows representative 505.3 nm absorbance traces fit to exponential decays.Table 2 summarizes the conditions of the fourteen 505.3 nm absorbance experiments.Each experiment was conducted at a fixed T and P, and consisted of at least five absorbance traces at different concentrations of propene, [C 3 H 6 ].The nominal T was varied from 300 to 700 K, and P was mostly kept at 10 Torr, except for one control experiment at 25 Torr to demonstrate the independence of k total (T) from P. C 6 H 5 I was used as the photolytic C 6 H 5 precursor in all but two of the experiments for the three reasons mentioned in the Experimental section.Initial radical concentrations were quantified by single-pass IR absorbance of the I atom, following the same procedure described previously. 41,42For a given experiment, [C 3 H 6 ] was varied over a wide range to verify a linear relationship with k 0 .Some of the fit k wall values are rather large (41000 s À1 ), likely due to a small air leak during those experiments that was later fixed (importantly, all of the MBMS results presented in the next section were obtained without the leak).However, in all experiments [C 3 H 6 ] was increased such that the decay of C 6 H 5 due to C 6 H 5 + C 3 H 6 was 5-10Â the contribution of k wall .The vertical shift factor, a, was usually between À0.1 and 0.1, and in the worst case it was À0.25.Besides varying the precursor identity and P, control experiments were also conducted using different photolysis laser fluences, precursor concentrations, radical concentrations, and flashes per refresh (FPR), in order to verify the independence of a given k total (T) from all of these parameters.The reported uncertainty in k total (T) is at least 10% due to possible systematic uncertainty in [C 3 H 6 ], as well as fitting uncertainty. 21For all experiments, the overlap pathlength between the Herriott cell and the photolysis laser was B20 m.The absolute 505.3 nm absorbance signal observed for a given [I] 0 (Q[C 6 H 5 ] 0 ) was consistent with a C 6 H 5 cross-section of B2 Â 10 À19 cm 2 under all conditions, similar to the measurement of Tonokura et al. 14 (3.6 AE 1.6 Â 10 À19 cm 2 at 504.8 nm and 298 K).
Fig. 6 compares the fourteen measurements of k total (T) in this work against the previous measurements of Park et al. using CRDS. 7The two sets of measurements agree well with each other, including the various control experiments.By using a different chemical precursor that does not significantly thermally decompose at o900 K (C 6 H 5 I), 26 this work is able to extend the upper T at which k total (T) is measured by B200 K compared to the previous work (C 6 H 5 NO, o700 K 24 ).
Also shown is the predicted k total (T), obtained by summing the TST calculated rate coefficients for the five different entrance channels.The barrier for terminal addition had to be lowered by 0.75 kcal mol À1 in order to obtain the agreement between experiment and theory shown in Fig. 6.As expected, terminal addition to i1 is predicted to be the dominant entrance channel, especially at lower T, with central addition and allylic H-abstraction B1-2 orders of magnitude slower over the experimental range.The calculated entrance rates in Fig. 6 (including the adjusted rate for terminal addition) were used in all subsequent C 6 H 5 + C 3 H 6 models discussed in the following sections.

Products measured by MBMS
Table 3 summarizes the conditions of the 15 MBMS experiments.Only nominal temperatures of 600 and 700 K were explored because for T o 600 K secondary reactions involving the I atom will dominate the reaction flux (discussed later).Also, at lower T a higher [C 3 H 6 ] is needed to ensure fast and near-unity reaction of C 6 H 5 with C 3 H 6 , but higher [C 3 H 6 ] tends to lower the MBMS sensitivity both due to attenuation of VUV photons and due to a decrease in the density of the molecular  beam for a heavier carrier gas. 43P was 10 Torr for all of the experiments except for two conducted at 25 Torr and another two conducted at 50 Torr to test the effect of pressure on the product distribution.
For a given T,P-condition, the time-dependent product branching was typically measured over four different [C 3 H 6 ], including an experiment without C 3 H 6 .Control experiments were also conducted at different photolysis laser fluence, precursor concentration and radical concentration to check the effect of these variables.Both 505.3 nm absorbance of C 6 H 5 and IR absorbance of the I atom were measured simultaneously with all of the MBMS experiments.
Fig. 8 shows representative, time-resolved mass spectra that have had the pre-photolysis background subtracted out so that only transient signals are observed.The spectra are divided into two m/z regions for clarity: a lower range from 70 to 125 amu where all of the primary C 6 H 5 + C 3 H 6 product signals appear, and a higher range from 125 to 250 amu where the side and secondary product signals appear, many of them corresponding to iodide-containing species.No transient signals were observed below 77 amu (C 6 H 5 ) for any experiment.Qualitatively, all of the primary C 6 H 5 + C 3 H 6 products (see Fig. 2 for prediction of product branching) are observed at their expected parent m/z: 78 amu for C 6 H 6 , 91 amu for p10 (benzyl radical), 104 amu for p1 (styrene), 118 amu for p2-p4 (phenylpropene isomers) and 119 amu for i1 and i2.It should be reiterated that initially the 91 amu product was unexpected based on the previous C 6 H 5 + C 3 H 6 literature, but can now be rationalized by the aromaticcatalyzed 1,2-H-migration that facilitates p10 formation discussed in Section 4.1.The side and secondary products can also be assigned intuitively: 127 and 128 amu correspond to the I atom and HI, 134 amu is the recombination product of i1 + CH 3 (named i1-CH 3 ), 142 amu is methyl iodide (CH 3 I), 154 amu is biphenyl (C 6 H 5 -C 6 H 5 ), 168 amu is allyl iodide (C 3 H 5 I) and 246 amu is the recombination product of i1 + I (named i1-I).There is also a small transient signal at m/z = 160 amu that is difficult to discern in Fig. 8 corresponding to i1 + allyl radical (named i1-C 3 H 5 ).
All of the transient MBMS signals were integrated and plotted against reaction time, as shown for three representative experiments in Fig. 7 (experiments #2, 7 and 15).The thick solid lines in those plots correspond to model predictions, which will be described in the following section.Qualitatively, there are several features of Fig. 7 that are consistent with the product branching predictions of Fig. 2. First, the product branching has a strong dependence on T, which can be seen by comparing experiments #2 and 7, conducted at nominal T's of 600 and 700 K, respectively, and 10 Torr.Second, p1 (104 amu) is the dominant product at 700 K, assuming similar photoionization cross sections (PICS) for all products.The 104 amu signal is 3-4Â any of the other product signals, which is in nearly quantitative agreement with Fig. 2. Third, C 6 H 6 is clearly observed as a primary product.The importance of C 6 H 6 with increasing T was predicted both here and in Kislov et al., 11 but has never before been directly confirmed experimentally and  was one of the main motivations for this work.Finally, the product branching has negligible P-dependence as seen by comparing experiments #7 and #15, both of which were nominally conducted at 700 K but at different P's (10 and 50 Torr, respectively).Although the time-dependence of the product growth is slower at higher-P, that is likely due to slower radial diffusion impeding the transport of transient species to the MBMS sampling volume rather than a chemistry effect (Section S1.5, ESI †).Disregarding the time-dependence, both experiments #7 and #15 reach nearly identical steady-state product distributions, in agreement with the lack of P-dependence exhibited in the predictions of Fig. 2. The 25 Torr experiment (#13) exhibits a behaviour intermediate to the 10 and 50 Torr experiments (Fig. S30-S32, ESI †).Fig. S32 and S36 in the ESI † show that the instantaneous primary product ratios (relative to styrene/104 amu) at 25 and 50 Torr are essentially time-independent and in agreement with the model, further supporting the claim that the distinct time-dependence of the absolute signal at higher-P is due to transport rather than chemistry effects.Although there is great qualitative agreement between the measurements of Fig. 7 and the predictions of Fig. 2, there are some clear discrepancies.Most notably, at 600 K (experiment #2) the largest product signals are at m/z = 118 and 119 amu, nominally corresponding to p2-p4 and i1.The dominance of 119 amu at later times is especially unexpected given its assumed identity as a radical.Clearly there is more going on than suggested by Fig. 2, and a more detailed model is needed, specifically, a model that includes not just the major reactions on the C 9 H 11 PES, but also side and secondary reactions on other PESs, including those with the I atom.Ideally, transport effects should also be accounted for in the model, as well as the weighting of each species' signal by its PICS, the contribution of 13 C isotopologues and dissociative ionization (fragmentation).Such a model was developed (model predictions shown as thick solid lines in Fig. 7) and is described in the following section.Another purpose behind constructing a model for this flash photolysis system with MBMS detection is to test if the current theoretical understanding of C 6 H 5 + C 3 H 6 is sufficient to quantitatively explain the experimental measurements.Of specific interest is whether the proposed aromatic-catalyzed 1,2-H-migration can really explain the 91 amu signal.After briefly describing the development of the model, its predictions will be compared in detail to each transient MBMS signal in order to test this understanding (next section).
Regarding the unexpected dominance of 118 and 119 amu at 600 K, careful modelling (and a critical experiment) revealed that both m/z's are actually mostly attributable to fragments of i1-I.At higher T's, i1 undergoes unimolecular isomerization and decomposition reactions much faster compared to its reaction with the I atom, so the role of the I atom is largely diminished at 700 K.At lower T's, the converse is true; hence MBMS experiments were not conducted below 600 K.
Time-profiles for all 15 MBMS experiments (with model comparisons and commentary) are plotted in Fig. S9-S36  (ESI †).It should be mentioned that for the experiments without C 3 H 6 , evidence of catalysis by the walls was observed.Specifically, fast hydrogenation of C 6 H 5 to C 6 H 6 was observed that was inexplicable by gas-phase chemistry (i.e., C 6 H 5 self-reaction or reaction with the precursor).Slower, but still inexplicable, hydrogenation of the I atom to HI was also observed.For this reason, MBMS experiments were conducted at various [C 3 H 6 ] to confirm that sufficient C 3 H 6 had been added such that C 6 H 5 mostly reacted with C 3 H 6 in the gas-phase, rather than abstracting an H-atom from the wall.

Modelling of flash photolysis with MBMS
The overall approach to modelling the MBMS experiments is summarized in Fig. 9 and many of the individual steps are discussed in detail in the indicated ESI † sections.Briefly, after being ''trained'' on the relevant alkylaromatic chemistry, RMG was used to automatically construct a chemical mechanism for the important hydrocarbon (HC) chemistry occurring in the flash photolysis experiment.A separate sub-mechanism for I atom chemistry was manually created, largely taking inspiration  C isotopologues of major species were created assuming a natural abundance of 1.1% Â # of carbon atoms.Only isotopologues with zero or one 13 C are considered (or observed).Simulated species signals were then fragmented, again using either known, measured or estimated fragmentation patterns.Many of the iodide containing species, R-I, fragment extensively to R + due to the relatively weak C-I bond, 47 which was important  those two entrance channels.However, the measurements and the simulations already overlap considering the uncertainty in the model (at least AE15% due to the PICS).Therefore it was deemed unnecessary (and essentially meaningless) to fit the barrier heights.Also, at 600 K, both the model and the experiment show a slower growth time scale for 104 amu (B10 ms), which is formed in two steps (C 6 H 5 + C 3 H 6 -i1 -p1 + CH 3 ), compared to 78 amu (B1 ms), which is formed in only one step (C 6 H 5 + C 3 H 6 -C 6 H 6 + C 3 H 5 ).This demonstrates that although the time resolution of the current MBMS experiments is only B1 ms, important differences in chemical time-scales can still be resolved.
91 amu, shown in Fig. 11, is modelled as a complicated mixture of three different species signals all with distinct t-and T-dependence: p10 (benzyl radical) and fragments of both i1 and p10-I.For 600 K, the agreement at early times is due to fitting the extent of i1 fragmentation to 91 amu.The measurements at later time were not fit, however, and are in excellent agreement with the model, which mostly consists of the p10-I contribution.
Even more remarkable is that the model still performs well at 700 K, where nothing was fit, and the contribution of i1 is greatly diminished because of its fast unimolecular isomerization/ decomposition at that T. Once again, at later times the model is dominated by p10-I, in accordance with the experiment.The measurements in Fig. 11 and the ability of the model to quantitatively explain them provide strong support for the never-before-seen aromatic-catalyzed 1,2-H-migration proposed in Section 4.1.Without the addition of this pathway to Kislov et al.'s original PES, 11 virtually no p10 or p10-I would be predicted, and the long-time 91 amu signal observed experimentally would be unaccounted for.
Both 118 and 119 amu, shown in Fig. 12 and 13, respectively, have large contributions from fragmentation of i1-I.119 amu in particular is dominated by i1-I in the simulations at both 600 and 700 K, with a negligible contribution from the parent cation of i1.118 amu exhibits a more diverse mix of simulated species: at 600 K i1-I is the largest contributor to the 118 amu signal, but the phenylpropene isomers p2 and p3 also make a  Fig. 14 compares the I atom and HI at 127 and 128 amu, respectively.The rise time of the 127 amu signal was fit for k sampling , and its absolute value was roughly fit for [I] 0 .Therefore, the agreement between the modelled and measured I atom at early times is unremarkable.At longer times, however, the simulated decay of the I atom was not fit and generally follows the trend of the measured 127 amu signal, especially considering that the I atom PICS is B35% uncertain (Table S3, ESI †).Given the importance of i1 recombination with the I atom at 600 K in explaining many of the observations made at that T, it is reassuring that the absolute I atom signal itself is also described well by the model over the entire measured time range.
As discussed in Section S1.2 (ESI †), HI is mostly formed through wall catalysis, which was not included in the model.Therefore, HI is underpredicted, especially at 600 K.At 700 K, the wall catalysis effect is reduced, probably because of less  adsorption on the wall at higher T, 48 but HI is still underpredicted.Nonetheless, matching HI is not critical to the main goals and conclusions of this work, and is not discussed further.Fig. 15 compares the C 6 H 5 signal measured both by MBMS at 77 amu and by simultaneously recorded 505.3 nm absorbance.As a short-lived species, C 6 H 5 is difficult to resolve with MBMS, an issue that is compounded by the overlap of 77 amu with a fragment of C 6 H 5 I, which is present in relatively high concentration.However, the decay of C 6 H 5 was clearly resolved by 505.3 nm absorbance, which is a much faster and non-intrusive detection technique for measuring overall kinetics.The simulated C 6 H 5 decay (without sampling effects, of course) is in good agreement with the 505.3 nm absorbance, which is expected given that the terminal addition rate of C 6 H 5 + C 3 H 6 was previously fit to absorbance measurements in Section 4.2.Fig. 15 and Fig. 9-13 illustrate the complementary nature of laser absorbance coupled with MBMS for kinetic studies: laser absorbance can resolve the time dependence of only one species (or at most a few species if there is spectral overlap that can be deconvoluted) with high t-resolution, whereas MBMS can resolve the time dependence of many species with lower t-resolution.Of course, the t-resolution of the MBMS measurements reported here can and should be improved in the future, but MBMS is unlikely to supplant laser absorbance as the preferred method for overall kinetics quantification for multiple reasons to be discussed in an upcoming publication. 17iphenyl, C 6 H 5 -C 6 H 5 , is also measured with MBMS at 154 amu (Fig. 15), and the model is in near-quantitative agreement, especially considering that the PICS of C 6 H 5 -C 6 H 5 was estimated by the method of Bobeldijk et al. 49 The C 6 H 5 self-recombination rate (C 6 H 5 + C 6 H 5 -C 6 H 5 -C 6 H 5 ) used in the model is also an unadjusted estimate used directly from RMG.
Measured and modelled m/z signals corresponding to the products of i1 bimolecular recombination with the H atom (120 amu), CH 3 (134 amu) and C 3 H 5 (160 amu) are shown in Fig. 16 and 17.At 600 K, the signal at m/z = 120 amu is actually mostly attributable to the 13 C isotopologues of various 119 amu species/fragments (specifically, the fragment of i1-I), with propylbenzene (from i1 adding an H either by recombination, H-abstraction or disproportionation) contributing B1/3 of the total modelled signal.At 700 K, the simulated signal is about an order of magnitude lower, below the detection limit of the MBMS, consistent with the lack of any measurable 120 amu signal.134 and 160 amu are entirely attributable to the expected species: i1-CH 3 and i1-C 3 H 5 .At 600 K, the signal is discernible but near the detection limit, and by 700 K the simulated and measured signals have dropped below the detection limit.As mentioned before in a different context, this is because i1 is too short-lived at higher T's to undergo bimolecular reactions (such as recombination with I, CH 3 or C 3 H 5 ).Although not shown, in the model another secondary product of i1 bimolecular reaction, i1-dimer, has a similar predicted signal and T-dependence as the species in Fig. 16.However, no transient signal at the corresponding parent m/z (238 amu) was observed even at 600 K (Fig. 8).Given that the PICS of i1-dimer was estimated as 2Â the PICS of propylbenzene (2 Â 30 = 60 MB), 49 its signal is probably slightly overpredicted, which being so close to the detection limit can make the difference between seeing it and not seeing it experimentally.
Finally, Fig. 18 shows the parent m/z's of all of the important iodide-containing species: 142 amu for CH 3 I, 168 amu for C 3 H 5 I and 246 amu for i1-I.The parent m/z signal for p10-I at 218 amu is below the detection limit due to its severe (B10 : 1) 50 fragmentation to C 7 H 7 + .The modelled C 3 H 5 I and i1-I both quantitatively match their respective m/z signal measurements at both T's (keeping in mind that the PICS and fragmentation pattern of i1-I were fit to the 600 K experiments).CH 3 I, however, is overpredicted, especially at 700 K.As discussed in Section S1.2 (ESI †), this is probably due to fall-off effects, not included in the I atom sub-mechanism, lowering the effective rate of CH 3 + I -CH 3 I at higher T's and lower P's than where the original rate measurement was conducted (400 K, 82 Torr CH 3 I). 51lthough the CH 3 + I -CH 3 I rate could be fit to match the 142 amu MBMS signal, given the uncertainties in the many other CH 3 reactions and given that CH 3 was not observed with

Products measured by absorbance
Attempts were also made to quantify product branching by laser absorbance given that at room temperature three of the important radicals in the C 6 H 5 + C 3 H 6 system, C 6 H 5 , C 3 H 5 and p10, are known to exhibit distinct visible absorbance features at 504.8, 14 408.4 37 and 447.7 nm, 38 respectively (Fig. 19).However, the strength of these features is expected to drop with increasing T due to line-broadening. 16In particular, the peak absorbance cross section for p10 at 447.7 nm is known to sharply decrease by B1 order of magnitude when increasing T from 300 to 600 K, 55 such that at T Z 600 K all three radicals in Fig. 19 will have similar absorbance cross sections of B2 Â 10 À19 cm 2 .Table S4 (ESI †) summarizes the conditions of the experiments to probe for products (C 3 H 5 and p10) with laser absorbance.The experiment #'s are a continuation of those for the MBMS experiments.Experiments #2-4, 16-18 and 22-24 were conducted under nearly identical conditions (nominally 600 K, 10 Torr and a range of [C 3 H 6 ], the radical concentration was slightly higher for the product probing experiments) but the probe laser wavelength was varied from 505.3, 404.8 and 447.7 nm, respectively.A similar procedure was followed for experiments #7-9, 19-21 and 25-27, which were conducted at 700 K and 10 Torr.
The insets of Fig. 19 compare the 700 K backgroundsubtracted and normalized absorbance measurements at the three probe wavelengths.All of the other absorbance traces are shown in Table S5 (ESI †).As already discussed (Section 4.2) 505.3 nm absorbance of C 6 H 5 typically returns to the baseline within AE10% due to noise and imperfect background subtraction (Table 2).C 6 H 5 continues to contribute substantially to the absorbance at the lower wavelengths as well, similar to previous measurements of its UV-visible spectrum. 56At 447.7 nm, the absorbance also essentially returns to the baseline, but at 408.4 nm there is a clear, reproducible baseline-shift of B30%, which exceeds the typical baseline shift fluctuations of AE10%.From Scheme 3 alone it can already be seen that simply adding one more carbon to the alkyl chain of i1 opens up new chemistry (left branch).However, there are two more novel applications of the ''aromatic-catalyzed'' concept to the 1-C 10 H 7 + 2-C 4 H 8 system that were not found by RMG.In the first, rather than the ortho-carbon (labelled 2 in Scheme 4) serving as the ''active site'' in the aromatic catalyst (source and sink of H), carbon 9 could instead function as the active site.In fact, carbon 9 is probably a better active site than carbon 2 because in the rate-limiting step of the aromatic-catalyzed 1,2-H-migration carbon 2 loses its H in a 5-membered-ring TS (Fig. 1 and 3) whereas carbon 9 would form a more favourable 6-membered-ring TS.In the second application, carbon 2 could still be the active site, but instead of abstracting an H back from carbon 11 in the second step, as in Scheme 3, it could abstract an H from carbon 14.In this case, the net effect would be a 1,3-H-migration.The fact that RMG was unable to extrapolate the aromatic-catalyzed 1,2-H-migration, on which it was trained, to a different active site or to a 1,3-Hmigration (or to a combination of both: a 1-3-H-migration using carbon 9 as the active site) points to areas of future work.Fig. 20 shows RMG's quantitative predictions of the product branching for 1-C 10 H 7 + 2-C 4 H 8 at 600 K and 1 atm.Just as in the case of C 6 H 5 + C 3 H 6 , the major products are from H-abstraction (red line) and radical addition followed by CH 3 This journal is © the Owner Societies 2018 loss (black) or H loss (magenta and green).All of the ''aromaticcatalyzed products'' each have a yield of less than 1%, and in total they only account for B4% of the product branching.However, if RMG had found the other three aromatic-catalyzed reactions described in the previous paragraph the branching would be higher.Nonetheless, the general applicability of aromatic-catalyzed reactions has been demonstrated and applied to the example 1-C 10 H 7 + 2-C 4 H 8 system using RMG.Given the staggering number of different alkylaromatic structures that would be encountered in a real application (e.g., gasoline, kerogen), a tool like RMG is needed to extrapolate findings made for model systems (e.g., aromatic-catalyzed 1,2-H-migration in C 6 H 5 + C 3 H 6 ) to all possible analogous systems, even if the extrapolation is currently unsophisticated.

Conclusions
By applying a different experimental approach (flash photolysis with time-resolved MBMS) to a chemical system that has already been studied extensively both experimentally and computationally (C 6 H 5 + C 3 H 6 ), several new insights were made.
First, and most importantly, a previously unreported aromatic-catalyzed 1,2-H-migration was proposed to explain unexpected benzyl radical formation observed experimentally from C 6 H 5 + C 3 H 6 .Quantum calculations of the energetics and RRKM/ME calculations of the kinetics for the new pathway were both favourable, predicting up to B10% and B15% product branching to the benzyl radical under thermal and single collision conditions, respectively.In order to gain further confidence in these predictions, a detailed model for the flash photolysis/MBMS experiment was developed that includes both hydrocarbon and iodine chemistry, as well a simple empirical model for transport.The model could quantitatively explain the complicated time-and temperature-dependence of the unexpected product (as well as the other four primary product m/z's) observed experimentally, providing strong support for the aromaticcatalyzed pathway as the main route to the benzyl radical (and resonance stabilized radicals more generally).Second, the extent of H-abstraction from C 6 H 5 + C 3 H 6 was quantified experimentally for the first time, using MBMS detection of the stable product (C 6 H 6 ), and was found to be in good agreement with the model described above (and the theoretical calculations upon which the chemistry portion of the model relies).Attempts to quantify H-abstraction by probing for the radical product (C 3 H 5 ) with laser absorbance were less quantitatively successful, however, either due to poorly understood secondary chemistry involving the I atom or due to spectral overlap of other radicals at the relatively short wavelength used (408.4nm).
Finally, the competition between CH 3 -loss and H-loss following radical addition of C 6 H 5 + C 3 H 6 was quantified experimentally and matched by the model.Under the conditions of these experiments (600 and 700 K) CH 3 -loss was dominant.
The only outstanding discrepancy in the C 6 H 5 + C 3 H 6 literature that this work did not address experimentally was the isomeric identity of the H-loss product.However, trapping of the effluent of the flash photolysis reactor followed by GC/MS detection of the stable C 6 H 5 + C 3 H 6 products might be a practical approach to addressing this issue in the future.
Given that the current theoretical understanding of C 6 H 5 + C 3 H 6 was sufficient to quantitatively explain the many experimental results of this work, this knowledge was ''taught'' to RMG using the language of training reactions.The ability of RMG to apply this knowledge to a slightly more complicated alkylaromatic system, 1-C 10 H 7 + 2-C 4 H 8 , was demonstrated, although areas for improvement were clearly identified.Specifically, RMG only applied the aromatic-catalyzed concept narrowly to the kind of 1,2-Hmigration seen in the C 6 H 5 + C 3 H 6 system, instead of applying it more broadly to 1,3-H-migrations with different ''active-sites'' of the aromatic catalyst.Despite these shortcomings, RMG and similar automated tools are currently the best options for extrapolating detailed chemistry insights, such as the ones made in this work, to real applications.Hopefully the overall framework presented here for translating fundamental experimental and theoretical insights into broader applications using RMG will serve as a guide for future fruitful work.For example, experiments with naphthyl radicals or butene isomers would be a logical extension of this work.

Scheme 1
Scheme 1 Major expected reaction pathways of phenyl radical + propene.
Scheme 2 gives an example of one of the training reactions added in this work and its position in the larger RMG database hierarchy.High-P kinetic data are provided with each training reaction and in this case the TST calculated k(T)'s were used.The kinetic portion of This journal is © the Owner Societies 2018 Phys.Chem.Chem.Phys., 2018, 20, 13191--13214 | 13197

4. 1
Fig. 3 shows a reduced version of Kislov et al.'s G3(MP2,CC)// B3LYP C 9 H 11 PES 11 appended with the aromatic-catalyzed 1,2-H-migration and subsequent RSR formation pathway calculated in this work.For clarity, only the kinetically important features of the PES are shown, and the same intermediate and product names are used as in the previous work (e.g., i1 for intermediate 1 and p1 for product 1, which is styrene).The major expected pathways of C 6 H 5 + C 3 H 6 forming styrene (p1), benzene (C 6 H 6 ) and phenylpropene isomers (p2-p4) have already been discussed in the Introduction and will not be repeated here.The relatively high-energy (B40 kcal mol À1 ) direct 1,2-Hshift from i1 to i4 is highlighted in red, and the lower-energy (B30 kcal mol À1 ) aromatic-catalyzed 1,2-H-shift is highlighted in green.Once i4 is formed, it can produce indane (p5) in two steps, as discussed in the Introduction, or it can simply b-scission a C-C bond to form the benzyl radical, C 7 H 7 (p10), and ethene, C 2 H 4 .It is because of the latter product channel, benzyl radical + C 2 H 4 , that the aromatic-catalyzed 1,2-H-migration was inferred from experiments described in the following sections.The aromatic-catalyzed 1,2-H-migration proceeds in two steps.First i1 undergoes a 1,4-H-migration, transferring one of the H's on an ortho-carbon to the secondary carbon radical for 50 milliseconds.The initial C 6 H 5 and C 3 H 6 concentrations were kept fixed at 3 Â 10 12 and

Fig. 2
Fig. 2 Predicted product branching of phenyl radical + propene on a linear (top, products with 410% branching) and logarithmic (bottom, products with o10% branching) scale as a function of T and P for otherwise fixed conditions: 50 millisecond reaction time, [C 3 H 6 ] = 5 Â 10 16 cm À3 and [C 6 H 5 ] 0 = 3 Â 10 12 cm À3 ([He] provides the balance of the gas density).Dashed lines are predictions at 10 Torr and solid lines are 1000 Torr.

Fig. 3
Fig. 3 Simplified phenyl radical + propene PES calculated using G3(MP2,CC)//B3LYP/6-311G(d,p).Radical addition pathways are shown on the left and hydrogen abstraction on the right.The direct and aromatic-catalyzed 1,2-hydrogen-migrations are highlighted in red and green, respectively.Energies and geometries for all stationary points were calculated by Kislov et al., 11 except for those in green and the b-scission of i4 to p10 + C 2 H 4 , which were calculated in this work.

Fig. 4
Fig.4Predicted product branching of phenyl radical + propene on a linear (top, products with 410% branching) and logarithmic (bottom, products with o10% branching) scale as a function of collision energy under single-collision conditions.

Fig. 5
Fig. 5 Representative 505.3 nm absorbance decays measured (markers) at 691 K, 10 Torr.Lines are exponential fits.Only every 10th point is shown for clarity.
44 and Tranter et al.,45 and combined with the HC mechanism.The main contribution of the I atom to the observed chemistry is as a recombination co-reactant: R + I = R-I, where R = H, CH 3 , C 3 H 5 , p10, i1 or another I.The initial radical concentration was fit to MBMS time-profiles of the photolytically produced I atom, and the combined HC + I atom mechanism was simulated in an isothermal, isobaric batch reactor in order to obtain concentration profiles for each species i, C i .Each C i is then weighted by the PICS for i, s PI,i , at 10.5 eV.PICS are mostly obtained from the literature, except for species where the literature values are unavailable, in which case either estimates or measurements of the PICS were made.The PICS weighted C i profiles were converted to instantaneous MBMS signals by applying the mass discrimination factor, R(m/z).R(m/z) was fit to the MBMS signals of internal standards, S int-std , also present in the flash photolysis reactor during experiments at known, low concentrations.Transport effects were accounted for by a simple model adapted from Baeza-Romero et al.46 that lumps all of the time-smearing associated with MB sampling into a single first-order rate coefficient, k sampling , which is fit to the measured rise time of the I atom MBMS profile.Once S i,sampled profiles had been calculated,13

Fig. 7
Fig. 7 Time profiles of primary phenyl radical + propene products measured with MBMS under indicated conditions.Markers are experimental measurements and lines are model results.

Fig. 10
Fig. 10 Measured (markers) and modelled (thick lines) 78 and 104 amu MBMS signals under indicated conditions.These are the only two primary phenyl radical + propene product m/z's that are entirely attributable to one species: benzene and styrene, respectively.

Fig. 11
Fig. 11 Measured (markers) and modelled (thick lines) 91 amu MBMS signal under indicated conditions.The thin lines correspond to different species contributing to the overall modelled 91 amu signal: the C 7 H 7

Fig. 12
Fig. 12 Measured (markers) and modelled (thick lines) 118 amu MBMS signal under indicated conditions.The thin lines correspond to different species contributing to the overall modelled 118 amu signal: the C 9 H 10 + fragment of i1-I (black) and the parent cations of 1-, 2-and 3-phenylpropene (blue, red and green, respectively).

Fig. 13
Fig. 13 Measured (markers) and modelled (thick lines) 119 amu MBMS signal under indicated conditions.The thin lines correspond to different species contributing to the overall modelled 119 amu signal: the parent cations of i1 and i2 (blue and green, respectively), the C 9 H 11 + and 13 C-containing C 9 H 10 + fragments of i1-I (red and orange, respectively)and the parent cations of13 C-containing 1-, 2-and 3-phenylpropene (cyan, gold and black, respectively).

Fig. 14
Fig. 14 Measured (markers) and modelled (thick lines) 127 (black) and 128 (blue) amu MBMS signals under indicated conditions, which are exclusively attributable to I and HI.

Fig. 16
Fig. 16 Measured (markers) and modelled (thick lines) 134 (blue) and 160 (green) amu MBMS signals under indicated conditions.Both signals are exclusively attributed to the products of i1 recombination with other radicals: methyl and allyl radical, respectively.Both signals were too low to be detected at 707 K, therefore only the modelled signal is shown under that condition.

Fig. 17
Fig. 17Measured (markers) and modelled (thick lines) 120 amu MBMS signal under indicated conditions.The thin lines correspond to different species contributing to the overall modelled 120 amu signal: the parent cation of propylbenzene (blue), the 13 C-containing C 9 H 11 + fragment of i1-I Fig. 17Measured (markers) and modelled (thick lines) 120 amu MBMS signal under indicated conditions.The thin lines correspond to different species contributing to the overall modelled 120 amu signal: the parent cation of propylbenzene (blue), the 13 C-containing C 9 H 11 + fragment of i1-I(cyan) and the parent cations of13 C-containing i1 and i2 (green and red, respectively).The overall 120 amu signal at 707 K was too low to be detected, therefore only the modelled signal is shown under that condition.

Fig. 19
Fig. 19 Room temperature visible absorbance spectra measured by Tonokura et al. for allyl, 37 benzyl 38 and phenyl 14 radicals.The insets show representative absorbance traces (markers are measured and lines are modelled) measured in this work at the different wavelengths indicated and otherwise identical conditions (707 K, 10 Torr, [C 3 H 6 ] = 7.5 Â 10 15 cm À3 ).

Table 2
Summary of 505.3 nm absorbance experiments.Uncertainties represent two standard deviations Precursor concentrations calculated assuming that He exiting the bubbler is saturated with C 6 H 5 X at its room temperature vapor pressure: 0.92 Torr for C 6 H 5 I 39 and 0.6 Torr for C 6 H 5 NO.40 dPhotolysis laser fluence.e Units are molecule, s, cm.f Vertical shift factor in fits to normalized absorbance traces.g Repetition rate of the photolysis laser doubled to 2 Hz to check the effect of flashes per refresh (FPR).
a Identity of X in the precursor, C 6 H 5 X. b 10% uncertainty in all values due to systematic uncertainty in mass flow controller calibrations.c

Table 3
Conditions of MBMS experiments.Uncertainties represent two standard deviations from recent work by Comandini et al.