On-surface cyclodehydrogenation reaction pathway determined by selective molecular deuterations

Understanding the reaction mechanisms of dehydrogenative Caryl–Caryl coupling is the key to directed formation of π-extended polycyclic aromatic hydrocarbons. Here we utilize isotopic labeling to identify the exact pathway of cyclodehydrogenation reaction in the on-surface synthesis of model atomically precise graphene nanoribbons (GNRs). Using selectively deuterated molecular precursors, we grow seven-atom-wide armchair GNRs on a Au(111) surface that display a specific hydrogen/deuterium (H/D) pattern with characteristic Raman modes. A distinct hydrogen shift across the fjord of Caryl–Caryl coupling is revealed by monitoring the ratios of gas-phase by-products of H2, HD, and D2 with in situ mass spectrometry. The identified reaction pathway consists of a conrotatory electrocyclization and a distinct [1,9]-sigmatropic D shift followed by H/D eliminations, which is further substantiated by nudged elastic band simulations. Our results not only clarify the cyclodehydrogenation process in GNR synthesis but also present a rational strategy for designing on-surface reactions towards nanographene structures with precise hydrogen/deuterium isotope labeling patterns.

S5 phonon frequencies and phonon eigenvectors (i.e., atomic vibrations). Raman intensities of phonon modes were then calculated using the in-house developed Raman modeling package. 6,7 The computation of Raman intensity essentially requires the derivatives of the dielectric tensor ) ( L E   with respect to atomic displacements, which can be achieved by the finite difference method as well. For both positive and negative atomic displacements (δ = 0.03 Å), the frequency-dependent dielectric tensors were computed by VASP and then their derivatives can be obtained. Note that the ) ( L E   dielectric tensor should be calculated at the incident laser photon energy . Since DFT ) ( L E   L E often underestimates the optical gaps (for 7-aGNRs, the experimental optical gap is about 2.1 eV (ref. 8  From the simulated Raman spectrum with a smaller broadening (2 cm -1 ), one can find that the BLM is also split, with a smaller splitting of about 7.7 cm -1 , compared to the larger splitting of about 19.7 cm -1 of the BLM3, but the SLM does not split (Fig. S8). While there is no apparent splitting for the BLM in experiment, a slight broadening is indeed observed with high-resolution Raman spectroscopy when comparing the pristine and deuterated GNRs (see Fig. S9). This could be assigned to two reasons. First, the splitting is overestimated in the simulation. For example, the experimentally observed splitting of BLM3 is about 13.0 cm -1 , which is smaller than the splitting of 19.7 cm -1 from the simulation. Therefore, the experimental splitting of the BLM should be smaller than the calculated value of 7.7 cm -1 . Second, the low percentage of isotopic impurity may slightly S6 modify the H/D patterns, which would further suppress the BLM splitting.

Mass spectrometry
A commercial RGA100 residual gas analyzer (Stanford Research Systems) was used. By using the O100MAX Maximum Insertion nipple, a short sample-ionizer length of about 2 cm was achieved and used for all mass spectrometry measurements in experiment. Each time the RGA100 is fully degassed before recording the signal. The polymers were further annealed at 520 K for 1 h to fully remove the residual bromine (Br) atoms generated during polymerization 9 (Fig. S10). Coverage was controlled to be lower than one monolayer (around 0.5~0.9 monolayer) to avoid influences from the multiple layer molecular precursors 10 (Fig. S12). Then the mass spectra of HD (3 amu), D 2 (4 amu), H 2 (2 amu), and HBr (81 amu) were recorded at annealing of 670 K for at least 25 min. As reference, only H 2 are detected without HD and D 2 for pristine precursors (Fig. S10). H 2 , HD and D 2 signals were also measured for the clean Au substrate as background (Fig. S11).

CI-NEB simulations
The first-principles density functional theory (DFT) calculations for energy profiles were performed with the Quantum Espresso code, 11 using ultrasoft pseudopotentials 12 and Perdew-Burke-Ernzerhof (PBE) exchange-correlation functional. 13 The kinetic energy cutoff for the plane wave basis of Kohn-Sham wavefunctions was set at 24 Ry, and that for the charge density at 200 Ry. The supercells consist of four anthrylene units with periodic boundary condition, which were relaxed until forces on atoms reached a threshold of 0.026 eV/Å. The periodic direction of the polymer is aligned along the [110] direction of Au to allow for minimum lattice mismatch and strain. A nonlocal van der Waals (vdW) correction, 14 i.e., the self-consistent vdW-DF method, as used in our previous report, 15 was chosen to calculate energies of the polymer and the GNR on the metal substrate. The energy barriers of the reaction were calculated using the climbing image nudged elastic band (CI-NEB) method. 16 The forces on images were relaxed until they reached a threshold of 0.1 eV/Å. Different reaction pathways involving [1,2], [1,3]-sigmatropic, and [1,9]-sigmatropic hydrogen shift were examined (Fig. S15).

The Four different cyclodehydrogenation pathways discussed in this work
The direct hydrogen elimination pathway proposed by Björk et al. 17 is shown in Fig. S1a. From the initial state i, the neighboring anthrylene units first rotate about the polymer axis to approach each other, allowing two benzyne groups (C 6 H 2 D 2 ) on the same side of the polymer to form a single The [1,9] pathway proposed in this work based on the [1,9]-sigmatropic hydrogen shift is shown in Fig. S1c. First of all, Int1 is formed in the same way as in the [1,2] pathway. Following that, the D atom on the top side of the newly formed bond migrates by a [1,9]-sigmatropic hydrogen shift across the fjord to an edge C atom of a neighboring anthrylene unit, as highlighted by the red arrow and line segments, giving Int2. Next, the D atom at the other bonding C and the H atom at the edge C, both facing down to the Au substrate, are sequentially eliminated as ad-atoms on the Au surface, respectively giving Int3 and state 1, thereby restoring the aromaticity of the scaffold. Subsequently, a neighboring C aryl -C aryl bond forms in the same way as in a one-side domino-like mechanism, [18][19][20] S8 giving state 2 with two D atoms located at the previously down-tilted anthrylene edge. Finally, the other side of the polymer goes through the same reaction path, which leads to state f with the H/D pattern as identified by Raman spectroscopy and DFT calculations.
Two different [1,3] pathways, based on the suprafacial and antarafacial [1,3]-sigmatropic H shift, are shown as Figs. S1d 1 and S1d 2 , respectively. First of all, Int1 and Int2 are formed in the same way as in the [1,2] pathway. However, in the following step the D atom on the top side of the Then the H or D atoms on the edge facing down to the Au substrate are eliminated, giving 1 with different isotopes on the edge, thereby restoring the aromaticity of the scaffold. Subsequently, a neighboring C aryl -C aryl bond forms in the same way as in a one-side domino-like mechanism, [18][19][20] giving state 2 with two D or two H atoms located at the previously down-tilted anthrylene edge.
Finally, the other side of the polymer goes through the same reaction path, which leads to state f. The suprafacial [1,3] pathway produces the same H/D pattern as the [1,9] pathway, whereas the antarafacial [1,3] pathway gives a pristine 7-aGNR.

D elimination and migration order in different pathways
In the direct elimination pathway (Fig. S1a), D atoms will be directly eliminated after the C sp 3 -C sp 3 bond formation without undergoing migration. In both the [1,2] pathway ( Fig. S1b) and [1,3] pathway (Fig. S1d), the D facing down to the Au surface will first eliminate after the C sp 3 -C sp 3 bond formation, and then the other D facing up to the vacuum side will migrate between two immediate neighboring C atoms (thus the term of [1,2] shift) or across the fjord to the edge of a neighboring anthrylene unit, respectively. Differently, in the [1,9] pathway, the D facing up to the vacuum will migrate first also across the fjord to the edge of a neighboring anthrylene unit, while the D facing S9 down to the Au surface remains attached on the intermediate structure. The terms of [1,3] and [1,9] pathways are named according to the order terms of respective [1,3]-and [1,9]-sigmatropic hydrogen shifts.

Estimated gas-phase products in different pathways
In the direct elimination pathway, 17 the two D atoms at the two bonding C atoms are eliminated to form D 2 only (see Fig. S1a). In the [1,2] pathway, 18 two immediately neighboring H atoms on the ribbon edge are eliminated to give prevailing H 2 , leaving the well-separated D atoms to predominantly form D 2 (see Fig. S1b). In the [1,3] pathway, HD will be predominantly formed with a suprafacial [1,3]-sigmatropic hydrogen shift, while only D 2 will be generated with an antarafacial [1,3]-sigmatropic hydrogen shift (see Fig. S1d). Differently, as shown in Fig. S1c and Fig. 4a in the main text, in the [1,9] pathway, the elimination of D and H atoms happens sequentially and has a lateral separation of only one C-C bond. While full scrambling of the eliminated H and D atoms will give a H 2 /HD/D 2 ratio of exact 1:2:1 (ref. 21 ), the proximity between the eliminated H and D atoms suggests the more preferable formation of HD, thereby giving an expected HD ratio larger than 2.
The exact ratio between H 2 , HD and D 2 cannot be predicted due to two origins. The main reason is that the molecular precursors are not 100% isotopically pure. The other reason is that the kinetic isotope effect could slightly affect the ratio. However, one can still expect the dominant gas-phase products in different pathways, such as, dominant D 2 from direct elimination pathway, 17  from the antarafacial [1,3] pathway, and dominant HD from the [1,9] pathway, as illustrated and summarized in Fig. S13. S10 Table   Fig. S1 Four different cyclodehydrogenation pathways discussed in this work. (a)

Supplementary Figures and
Cyclodehydrogenation based on the pathway proposed by Björk et al. 17 In this pathway, the gas product will be only D 2 . (b) Cyclodehydrogenation based on the pathway proposed by Blankenburg S11 et al., 18 namely the [1,2] pathway. According to the elimination sequence for the D and H atoms, one can expect that the dominant gas-phase products will be H 2 and D 2 . (c) Cyclodehydrogenation proposed in this work based on the [1,9]-sigmatropic hydrogen shift, namely the [1,9] pathway.
According to the elimination sequence for the D and H atoms, one can expect that the dominant gasphase products will be HD. . Therefore, in the text we mainly focus on the [1,2], [1,3] and [1,9] pathways, where the latter two involving [1,3]-and [1,9]-sigmatropic hydrogen shifts are referred to as suprafacial pathways following the terminologies of pericyclic reactions. The data were obtained with a smaller broadening (2 cm -1 ), compared to those in Fig. 2b (16 cm -1 ). Only pattern 1 shows splitting for BLM and BLM3, while SLM does not for all patterns. For pattern 1, the BLM's splitting is about 7.7 cm -1 , much smaller than that for BLM3 (19.7 cm -1 ). In the experiment, we indeed observe a broadened BLM in deuterated GNRs compared to that in the pristine GNR (Fig. S9). (c) Experimental Raman spectra of the pristine and deuterated GNRs, same as those in Fig. 2a but with a narrow frequency range, for comparison with (b).   This is in contrast to the scenario as revealed in ref. 29 , where H 2 /D 2 molecules of various partial pressures (10 −5 ~ 10 −1 torr) are present, and therefore the hydrogen molecules can undergo an endothermic chemisorption at elevated temperatures, following the Le Chatelier's principle.  [1,9] pathway, same as Fig. 4a in the main text, while only considering one C-C bond formation. The three pathways share the same initial (state i), Int1, and final (state f) states. The [1,2] and [1,3] pathways have the same Int2', which is different from the Int2 in the [1,9] pathway, due to different H elimination orders. The [1,9] and [1,3] pathways have the same Int3, which is different from the Int3' in the [1,2] pathway, due to the different H shifts. One can see that the [1,3] pathway involves a hydrogen shift across the fjord of bond formation, similar to that in the [1,9] pathway.