Fast Molecular Compression by a Hyperthermal Collision Gives Bond-Selective Mechanochemistry

Elucidating mechanisms of a selective chemical reaction is fundamental towards control over its outcome. For molecules colliding on a surface, accelerating the molecule towards the surface has not been considered as a means to attain a selective reaction. Here we show bond-selective reaction of a molecule induced by its translational kinetic energy towards a surface. We use electrospray ion beam deposition to collide the molecule at low, hyperthermal translational energy (2 - 50 eV) with a Cu(100) surface and image the outcome at single-molecule level using Scanning Tunneling Microscopy (STM). A large mechanical impulse generated from the collision compresses the molecule and bends specific bonds within, making them react selectively. The compression-induced dynamics leads to reaction products that are inaccessible by thermal pathways since the compression timescale (sub-picoseconds) is much shorter than the thermalization timescale (nanoseconds or longer). The bond-selective, compression-induced chemistry, exemplified here by an organic dye molecule (Reichardt's Dye, C41H30NO+) colliding on a Cu-surface allows the exploration of mechanochemistry, especially for large molecules possessing complex chemical functionality that may have limited thermal stability.


MAIN TEXT
Chemistry is concerned with the manipulation of bonds between atoms with the goal to use chemical reactions to form desired substances. Achieving this goal requires an understanding of how selectivity emerges. Chemical reactions that occur following a molecule-surface collisions are important, technologically and for fundamental studies, in fields as diverse as heterogeneous catalysis [1][2][3][4][5][6][7][8][9], epitaxial material fabrication [10][11][12], biomolecular analysis [13][14][15][16][17], and astrochemistry [10]. Bond-selective reactions in a molecule-surface collision have been demonstrated by exciting specific vibrational modes of a molecule right before its surface impact [5,7,8]. This approach succeeds due to the sudden energy accumulation in a specific molecular degree-of-freedom triggering a reaction that promptly occurs before the deposited energy spreads to other degrees-of-freedom not involved in the reaction, i.e. before thermalization [18,19].
Imparting energy into the molecular center-of-mass motion (i.e. translation) towards the surface offers an alternative means to induce surface reactions, which is attractive because translational energy of a molecular ion is readily achieved by acceleration in an electric field.
However, simply accelerating molecules towards the surface has been deemed unsuitable for obtaining bond-selective reactions, because the collision excites soft modes of the molecule [20,21] which are poorly coupled to stiff stretching modes that promote bond-breaking reactions [9,19]. As a result, reactions would happen late after the thermalization occurs which diverges reaction pathways towards non-selective outcomes [10,[22][23][24][25][26].
It remains unclear whether excited soft modes in the absence of thermalization could give a selective reaction because previous studies have only detected the reaction products in the gasphase [10,[22][23][24][25][26] whereby excess energy from the collision cannot be removed from the molecular product and subsequently cause further reactions. We avoid this issue by using the surface to remove excess energy from the collision products. This is readily achieved by carrying out the experiment at lower, near-threshold energies, which necessitates the detection of the adsorbed collision outcome on surface.
Here we show that the large excitation of soft modes in a molecule surface collision, i.e. extensive compression of a molecule, occurring at timescales faster than thermalization, does lead to a selective, non-thermal reaction path. We have used scanning tunneling microscopy (STM) to detect the surface-bound products from collisions between a large molecular ions and a metal surface, carried out at low hyperthermal energies (2 -50 eV). Singly-protonated Reichardt's Dye (RD, C 41 H 30 NO + ) was collided at normal incident angle with a Cu(100) surface held at room temperature using electrospray ion beam deposition (ES-IBD) [27]. The collision outcome, examined by STM at 11 K (see Supplementary Fig 1), revealed reactive pathways that selectively cleaved a single C-N bond in RD. Energy-dependent experiments and ab-initio Molecular Dynamics (MD) calculations revealed orientation-dependent dynamics that selectively bend specific C-N bonds in the molecule, ignoring the minimum energy path.
Our work gives insight into the emergence of bond-selective mechanochemical reactions in molecule-surface collisions. Collision-induced mechanochemistry [28,29] promises to be generally applicable to large molecules such as peptides or even proteins [13][14][15][16][17], providing a new tool to perform non-thermal on-surface synthesis of novel molecular materials. The INTACT pathway gives an adsorbed RD. The CRACK pathway breaks a CN(AC) in the parent RD to give one α-fragment, while the SPLIT pathway breaks the CN(AL) to give one βand one Δ-fragment. Computed geometries (GEOM) show a broken CN(AC) (red dashed line) in the α-fragment, and a broken CN(AL) (blue dashed line) between β-and Δ-fragment. EXPERIMENT Figure 1 shows three outcomes obtained from the collision between an RD ion and a Cu(100) surface: one non-reactive, and two reactive outcomes, as revealed by STM imaging and simulation. These are the outcomes of specific collision dynamics that are accessed by aiming the ion beam at normal incidence angle to the surface, as shown in Fig 1A. The non-reactive pathway, INTACT, was found to yield an adsorbed intact RD with its two phenyl rings oriented vertical from the surface as shown in Fig 1B. We expect the proton that was attached to the Oatom in the RD ion to undergo spontaneous dehydrogenation on surface at room temperature given the low computed barrier of 0.08 eV.
The reactive pathways were found to be bond-selective to the C-N bonds, breaking either one out of the two types of C-N bonds in RD (see Fig 1A): the C-N bond pointing at an angle 'across' the N-O axis of the molecule, termed C-N(AC); or the C-N bond pointing 'along' the N-O axis, termed C-N(AL). In the CRACK pathway, a single C-N(AC) was broken to give a dissociated RD fragment (α-fragment) as shown in Fig 1B. In the SPLIT pathway, the C-N(AL) bond was broken to give two products as shown in Fig 1B: a diphenylphenoxy (β-fragment), and a triphenylpyridine (Δ-fragment). The observation of CRACK and SPLIT pathways thus establishes the existence of bond-selective pathways due to molecule-surface collision at hyperthermal translational energy. To gain insight into the dynamics of these bond-selective pathways, we measured their respective reaction probabilities against the kinetic energy of the molecular beam. We varied the collision energy by decelerating the ions approaching the surface, which shifted the energy distribution of the ion beam (see Fig 2A) without changing its width (see Methods). Since the ion beam is aimed normal to the surface, the collision energy is the kinetic energy, which corresponds to the molecular translation along the surface normal. The result of this energydependent measurement was fitted to a model, inspired by the Sudden Vector Projection (SVP) description of molecular collisions at the gas-surface interface [19]. We compute the reaction probability (P RXN,X ) for each pathway (X) by projecting the molecular translation vector to the reaction coordinate vector to estimate how much of the translation energy (E COL ) is utilized to propel the system towards the transition state (see Supplementary Fig 2 and Methods for details).
The model yields the relation: The fitting gave a translational energy threshold (E T in Eq. (1)) of 6.0 eV for CRACK, and 9.5 eV for SPLIT, as shown in Fig 2B. The existence of this threshold, which marks the minimal energy needed for the reaction, thereby evidences the translational energy as the cause of the reaction, ruling out surface-to-molecule charge transfer [10,24,25] as a sole cause of the reaction.
Further dynamical insight is obtained by the observation of the reaction probabilities approaching a limiting value, P SAT,X , estimated to be 0.43 for CRACK, and 0.35 for SPLIT at high energies using Eq. 1. Such a saturation of reaction probability below unity has been observed in gas-surface scattering experiment [25,30,31], and in surface-induced dissociation of proteins [16]. From the diatomic-surface scattering studies [30,31], the saturation was understood to be due to a steric effect in which there was a limited range of orientations in the approaching molecule (also known as 'cone of acceptance'see ref [32]) that upon collision with the surface would lead to a reactive outcome. Such a strong dependence on initial orientation is characteristic of a direct reaction in which a single-collision event causes a direct energy transfer from the molecule translation into the reaction coordinate [1,30,31,33].
Transferring insights from the diatomic-surface scattering studies to the present work, the saturation in reaction probability observed for CRACK and SPLIT is therefore indicative of an orientation-dependent direct reactions caused by a single-collision event, ruling out a multicollision event [33,34] via a precursor state [35].  Schematics of different initial RD geometries simulated for 15 eV collision energy, illustrated here as the surface approaching the molecule (the arrows mark the surface normal). The red and blue arrows indicate initial geometries that give CRACK and SPLIT outcomes respectively; and the grey arrows, for INTACT. B. Time-dependent quantities of RD averaged between all trajectories with different initial geometries: the molecule-surface distance (d mol-surf ), the stopping force exerted by the surface on RD (F mol ), the change in RD cross-section along the surface plane direction (ΔA, dotted line), and the compression of RD along the surface normal (ΔZ, solid line). The data is shown for 15 eV incident energy whereby the stopping force went up to ~2.4 nN for ~125 fs, compressing the entire molecule by ~45% and expanding its area by ~25% laterally. The shaded area gives the standard deviation between all trajectories.
The collision dynamics for all three pathways in general showed the molecule to be compressed onto surface due to the mechanical impulse from the molecule-surface impact. As shown in Fig 3B, the stopping force exerted by the surface on the incident molecule was computed to be operative in the nano-Newton regime at sub-picosecond timescale (see also Supplementary Fig 5). This impulse is understood to move the system to a region on the potential energy surface along the compression coordinate. For the reactive CRACK and SPLIT outcome, the system was impulsively propelled to a transition state en route to a product potential well.  Fig 4). This C-N bond bending was noted to modify the local geometry of the C-atom from a trigonal planar (sp 2 C-atom) to a trigonal pyramidal (sp 3 Catom), in effect converting a p-orbital into a sp 3 -hybridized C-dangling bond. Upon contact with the surface, this bent C-N bond breaks, and concurrently a new C-Cu bond forms, indicating a reactive event whose transition state is stabilized by the formation of a new bond [36]. In both pathways, the trajectory shows the C-N bond to dissociate in a single attempt, evidencing a direct energy transfer from the translational energy to the reaction coordinate. The trajectories shown are for the minimum translational energy required to give a reactive outcome: 12 eV for the CRACK pathway and 14 eV for SPLIT. The bottom panels show time-dependent quantities of RD: the angle gives the average angle formed between the C-atom and two of its neighboring atoms, which checks whether the C-atom in the C-N bond is in a sp 2 -or sp 3 -like geometry; and the distance gives the separation between the C-and N-atom in the reacting C-N bond as a black line, and the separation between the C-atom of the reacting C-N bond and the nearest Cu-atom as a green line. The yellow area marks the region where the compression from the collision bends the C-N bonds; the dashed line marks the beginning of the C-N rupture.
The preferential breaking of the C-N bond for the bond-selective CRACK and SPLIT pathways was understood to be due to its pronounced reactivity, which we attribute to two factors: (i) the destabilization due to C-N bending, and (ii) the C-N antibonding orbital being the lowest unoccupied molecular orbital (LUMO) of RD (see Supplementary Fig 6). The importance of these two factors were suggested by the correlation between C-N dissociation, and the charge flow from the surface to the LUMO of the compressed RD (i.e. RD with bent C-N bonds) (see Supplementary Fig 7). Similar to mechanism proposed for C-H dissociation of benzene on Cusurface [37], we expect the orbital mixing caused by the C-N bond-bending increases the propensity of the RD LUMO to hybridize, and form a bond with the surface. The proposed mechanism in which bond-bending alters the electronic structure of a reacting molecule is noted to be topical in mechanochemistry [28,29].
Finally, to address the possibility of the hyperthermal collision giving a chemical reaction via a precursor state of RD, we examined the minimum energy pathway (MEP) of an adsorbed RD. We consider this alternative pathway because an incident RD that had failed to react upon its first collision would have its energy increasingly equipartitioned among all its degrees of freedom while being trapped on surface, allowing the system to search for the MEP to react. The computed barrier along the MEP from an adsorbed RD to give an α-fragment was 1.23 eV, and to give a β-and Δ-fragments, 0.28 eV (see Supplementary Fig 8). The MEP barrier thus predict a reactive outcome dominated by the β-and Δ-fragments, in agreement with the thermal reaction experiment of adsorbed RD at ~350 K, which only gave β-and Δ-fragments as products. The hyperthermal collision clearly deviates from the MEP prediction since the α-fragment was the major product, thereby ruling out reactions via precursor state.
In conclusion, we have reported here the first observation of a bond-selective reaction from a hyperthermal collision of a polyatomic ion with a metal surface. The collision gave a mechanical impulse that compresses the incident molecule, causing a specific chemical bond to be activated. From a fundamental standpoint, we expect the collision-induced molecular compression described here to be a general phenomenon across surfaces with similar stiffness; and we expect such mechanism to play a central role in surface-induced dissociation experiments in tandem mass spectroscopy [13][14][15][16][17]. The experimental method described here provides a new general tool to study and apply compressive mechanochemistry for any molecules that can be electrosprayed [27]. Here the compression is operative on the entire molecule, instead of locally [38], for subpicoseconds, instead of permanently [29], thereby opening a new avenue to explore impulsive mechanochemistry. Additionally, the method described here offers new scalable method to non-thermally activate molecules to generate species that are inaccessible from conventional thermochemistry [12,27].

COMPETING INTERESTS
The authors declare no competing financial interests. for every collision energy. Some ambiguous species were excluded from the analysis; this refers to species that were in close proximity to one another, or species adsorbed at the step edges. The exclusion of such species did not bias the probabilities reported in Fig 2, since similar populations (~10%) of such species were found for every collision energy. We expect these species to possess the same relative populations of species as the adsorbates observed on the surface terrace.

Experiment
We estimated that there is no significant scattering of molecules to the gas phase upon collision since most of the incident molecules remained on the surface, even at 50 eV collision energy. This is evidenced by the good agreement between the total deposited charge and the observed coverage, as shown in Extended Data Fig 1. Typically we deposited a total charge of ~50 pAh (1 pAh = 1 picoampere hour) on a deposition area of 12.5 mm 2 for which we expect a density of ~225 molecules in a 50×50 nm 2 area. The observed high sticking probability for large polyatomic ions at 50 eV collision energy is consistent with that reported in the literature [2].

Data Analysis: Threshold Energies
To fit the experimental data that we inferred to be due to an impulsive collision dynamics, we constructed a mathematical model inspired by the Sudden Vector Projection method that has been applied to model impulsive dynamics in the gas phase and on surfaces [3]. For every reactive pathway, i.e. CRACK and SPLIT, we postulated: (1) Unique, non-overlapping reaction coordinate (RC) oriented at a specific axis from the molecule, as illustrated in Extended Data Fig 2. (2) Each RC-axis possesses a unique threshold energy (E T ) for which the reaction would occur if energy supplied along this coordinate is larger than E T .
In the context of a surface approaching the molecule as shown in Extended Data Fig 2, the kinetic energy supplied along a specific RC (E RC ) was determined by projecting the surface-approach-velocity vector onto the direction vector of the RC, which leads to: where E COL is the kinetic energy of the molecule approaching the surface at normal incident angle, and φ is the angle between the surface-approach-velocity vector and the RC-vector. We limit the domain of φ to [-/2, /2], thereby only considering approach angles that point towards the hemisphere whose axis is the RC-vector (illustrated in Extended Data Fig 2). This expression thus allows premise (2) to be written mathematically as: For gaseous molecules impacting on the surface at random orientation as in the experiment, we assume an equiprobable distribution of φ, expressed mathematically as: The cumulative probability at a given E COL can therefore be written as: Since p RXN is unity only for a range of φ that gives E RC ≥ E T , we demarcate the maximum range of φ by a constant φ max . As a result, the expression simplifies to: where: max ( COL ) = cos −1 √ T COL which gives the cumulative probability for one reaction coordinate at a given E COL as: Each collision can give multiple outcomes, and each outcome can have a number of reaction coordinates. Thus we impose a normalization condition where the sum of probability for all pathways, both reactive and non-reactive, at a specific E COL to be equal to 1, expressed as: We enforce this condition by scaling the event probability by a factor P SAT,X that is unique for each pathway, and independent from E COL , thereby yielding the final expression for pathway X (X could be INTACT, CRACK or SPLIT): where P SAT,X is the saturation reaction probability observed at high energy.
To account for the spread in collision energy as in the experiment, we perform a convolution integral in which we replaced the exact E COL by a Gaussian distribution centered at ̅ COL with a standard deviation of ΔE, yielding the expression of pathway X:  [5,6] calculations were performed with the plane wave-pseudopotential package Quantum ESPRESSO [7], using Ultrasoft pseudopotentials [8] with a wave function (charge) kinetic energy cutoff of 408 eV (4080 eV) and a GGA-PBE [9] exchange-correlation functional. The rVV10 [10] self-consistent van der Waals interaction was included. The Brillouin-zone was sampled with the k = Γ point, with a Methfessel-Paxton [11] smearing of 0.27 eV. The Cu (100) surface was modeled with a periodically repeated slab of three layers, with a vacuum gap between the adsorbed molecule and the bottom layer of the slab replica of ~10 Å. Only forces on molecule atoms and surface atoms belonging to the first two layers were allowed to relax, up to 0.026 eV/Å. STM simulations were performed by calculating the integrated local density of states (ILDOS) within the Tersoff-Hamann method [12]. STM simulations of the intact molecule, αfragment, Δ-fragment, and β-fragment decorated with a Cu-adatom were obtained at a constant current using the LEV00 package v3.4 (see ref [13]).
NEB calculations are shown in Supplementary Fig 8. In Note also that while the α-fragment is 0.4 eV less stable than the intact molecule, the β-and Δfragments are 0.3 eV more stable. In both NEBs, the state in the reaction coordinate R c =1 corresponds to the intact molecule in Figure 1B   (ref [17]). The Born-Oppenheimer molecular dynamics (MD) and barrier height calculations were performed at the Γ-point, with a cutoff energy at 400 eV. The Cu(100) surface was modeled as a (9 × 8) slab with four layers of atoms, where the lowest layer was frozen; and a vacuum gap of 25 Å. The projected-Density-of-States (pDOS) calculations employed a Gaussian smearing with σ = 0.25 eV. The molecular models and the charge density were visualized using the VESTA software [18].
The MD calculations were performed to simulate the collision dynamics between RD and the Cu(100) surface on the ground potential energy surface. The initial state was a hydrogenated RD with its center-of-mass placed 12 Å above the first layer of the Cu-slab. Each atom in the molecule was given an identical velocity along the surface normal towards the Cu-slab, which sets the center-of-mass translational energy to be the experimental collision energy (E COL ).
Additionally all atoms in the molecule and in the Cu-slab were initialized with random velocities sampled from a Boltzmann distribution at 298 K. The MD simulations were performed with a time-step of 0.5 fs as a microcanonical system that conserved the number of atoms (N), the volume of the supercell (V), and the total energy of the system (E).
The barrier height in the ground potential energy surface for the dehydrogenation of a singly hydrogenated RD on surface was obtained by climbing-image NEB calculations [5,6].
The calculations were performed with at least 5 images between the initial and the final state.
The band was relaxed until the forces orthogonal to the reaction coordinate were lower than 0.02 eV/Å.
Supporting Information Fig S4. Fitting of collision model into the experiment and ab-initio results.
The fit to experimental data used the energy-profile-broadened version of the fitting equation (see Eq (2) in Methods), while the fitting to ab-initio molecular dynamics results used the non-broadened version of the fitting equation (see Eq (1) in Methods). The fit was performed by enforcing a normalization condition whereby the sum of probabilities from three outcome (CRACK, SPLIT, INTACT) must equal to unity at all collision energies.

Supporting Information Fig S6. Computed LUMO for a singly-protonated RD in gas-phase. (A)
The LUMO was computed to be a π* antibonding orbital around the C-N bonds (isosurface: 0.03 eÅ -3 ). This is evidenced by the presence of nodal plane (marked by dashed lines) between C-and N-atom in both C-N(AC) and C-N(AL).