Report

Real-space and real-time observation of a plasmon-induced chemical reaction of a single molecule

See allHide authors and affiliations

Science  04 May 2018:
Vol. 360, Issue 6388, pp. 521-526
DOI: 10.1126/science.aao0872

Direct plasmon chemistry

Light can excite plasmons at a metal surface, which can then decay and create hot electrons that induce chemical reactions of adsorbed molecules. Kazuma et al. used a scanning tunneling microscope (STM) to induce and map out the surface dissociation of a dimethyl disulfide molecule on silver and copper surfaces. A silver STM tip created localized plasmons at different distances from the molecule. The plasmons drove the reaction directly by exciting the valence electrons of the molecule into unoccupied states and cleaving the sulfur-sulfur bond.

Science, this issue p. 521

Abstract

Plasmon-induced chemical reactions of molecules adsorbed on metal nanostructures are attracting increased attention for photocatalytic reactions. However, the mechanism remains controversial because of the difficulty of direct observation of the chemical reactions in the plasmonic field, which is strongly localized near the metal surface. We used a scanning tunneling microscope (STM) to achieve real-space and real-time observation of a plasmon-induced chemical reaction at the single-molecule level. A single dimethyl disulfide molecule on silver and copper surfaces was dissociated by the optically excited plasmon at the STM junction. The STM study combined with theoretical calculations shows that this plasmon-induced chemical reaction occurred by a direct intramolecular excitation mechanism.

Localized surface plasmon (LSP) resonances of metal nanostructures can focus light near the metal surface to sizes below the diffraction limit (1), and the generated localized electric field can be used for near-field optical spectroscopies (2, 3). In addition, LSPs facilitate highly efficient conversion of solar energy in photovoltaics (4, 5) and photocatalysts (5, 6). In particular, plasmon-induced chemical reactions of molecules adsorbed on metal nanostructures are attracting increased attention as photocatalysts (7, 8) that can form bonds (913) or dissociate them (1316). An indirect hot-electron transfer mechanism (Fig. 1A) (7, 8) has been invoked for these reactions. Electron-hole pairs are generated in the metal nanostructures by nonradiative decay of the LSP (7, 17, 18), and the hot electrons transfer to form a transient negative ion (TNI) state of the adsorbed molecule (7). Recently, plasmon-induced dissociations of O2 (14) and H2 (15) molecules were observed and explained by a mechanism in which the dissociation reactions proceed through vibrational excitation after the transfer of hot electrons generated TNI states.

Fig. 1 Excitation mechanisms for plasmon-induced chemical reactions.

(A) Indirect hot-electron transfer mechanism. Hot electrons (e) generated via nonradiative decay of an LSP transferred to form the TNI states of the molecule. (B) Direct intramolecular excitation mechanism. The LSP induces direct excitation from the occupied state to the unoccupied state of the adsorbate. (C) Charge transfer mechanism. The electrons are resonantly transferred from the metal to the molecule.

We propose a direct intramolecular excitation mechanism (Fig. 1B) on the basis of direct observation of a plasmon-induced chemical reaction of a single molecule. The LSP resonantly excites electrons from the occupied state to the unoccupied state in the electronic structure of an adsorbate. Single molecules in the strongly localized plasmonic field near the metal surface can be observed in real space and real time. We successfully made such observations through experimental studies with the LSP excited within the nanogap between a metal substrate and a Ag tip of a scanning tunneling microscope (STM).

Figure 2, A and B, illustrates our experimental scheme for investigating the plasmon-induced chemical reaction with the STM. Dimethyl disulfide [(CH3S)2] was selected as a target molecule for the plasmon-induced chemical reaction. To excite the LSP optically, the Ag tip with a curvature radius of ~60 nm (fig. S1) was positioned over the bare metal surface, and the sample bias voltage (Vs) and tunneling current (It) were set to 20 mV and 0.2 nA, respectively. Tunneling electrons at a bias of 20 mV cannot excite vibrational modes related to any kind of reaction, such as rotation, desorption, or dissociation of the molecule (19, 20).

Fig. 2 Real-space investigations of the plasmon-induced chemical reaction.

(A) Schematic illustration of the experiment for the real-space investigation of the plasmon-induced chemical reaction in the nanogap between a Ag tip and a metal substrate. The tip was positioned over the metal surface during light irradiation with the feedback loop turned on to maintain the gap distance. The Vs and It were kept at 20 mV and 0.2 nA. hν, Planck’s constant (h) multiplied by frequency (ν). (B) Division of the STM image into four areas depicted with 10-nm wide concentric rings for analysis. (C) Simulated spatial distribution of the electric field at the 1-nm gap under p-polarized light at 532 nm. E0, incident electric field. Topographic STM images of (CH3S)2 molecules on Ag(111) (D) before and (E) after irradiation with p-polarized light at 532 nm (~7.6 × 1017 photons cm−2 s−1, 2 s) (Vs = 20 mV, It = 0.2 nA, 43 nm by 43 nm). The tip was positioned at the center of area one during light irradiation. (F) Time dependence of the dissociation ratio (N/N0) under irradiation with p-polarized light at 532 nm (~5.9 × 1016 photons cm−2 s−1) in the four areas shown in (B). Each data point represents the average of results from six trials. The dotted lines denote single exponential functions fitted to the data points [ln(N/N0) = −kt]. Error bars indicate SD. (G) The rate constant k obtained at areas one through four and the calculated lateral profile of electric field intensity at 0.1 nm above the substrate surface (z = 0.1 nm) under 533-nm light. x = 0 nm corresponds to the center of the tip.

The LSP generates a strong electric field in the nanogap (Fig. 2C and fig. S2). Figure 2, D and E, shows the spatial distribution of isolated (CH3S)2 molecules on Ag(111) before and after the excitation of the LSP with p-polarized light at 532 nm. Although individual (CH3S)2 molecules appear as elliptical protrusions in the STM images, some molecules near the tip position were transformed into two identical ball-shaped protrusions after the excitation of the LSP (fig. S3). The dissociated chemical species have the same appearance as CH3S molecules obtained by injecting tunneling electrons into a (CH3S)2 molecule (1921). This implies that the S–S bond in (CH3S)2 was dissociated by the LSP. Notably, other reactions, such as rotation or desorption, were not observed.

The dissociation ratio (N/N0), which we defined as the number of (CH3S)2 molecules (N) after LSP excitation divided by the number of preadsorbed molecules (N0), was measured for quantitative analyses. Figure 2F shows plots of ln(N/N0) as a function of the irradiation time (t) in the four areas depicted with concentric rings (Fig. 2B). The linearity of the plots shows that the dissociation, (CH3S)2 → 2CH3S, is a first-order reaction. The slopes of the lines in Fig. 2F determine the dissociation rate constant (k). The rate constant was largest in area one and decreased with lateral distance from the tip (Fig. 2G). We calculated the electromagnetic field intensity in the nanogap (Egap) (Fig. 2C) using the Ag tip with a curvature radius of 60 nm and a cone angle of 15° estimated from the scanning electron microscopy images shown in fig. S1 (see the supplementary text and fig. S2). The lateral distribution of Egap is also shown in Fig. 2G. From comparison of k with Egap, we conclude that the plasmon-induced dissociation has a strong correlation with the electric field intensity of the optically excited LSP.

To explore a plausible mechanism for the plasmon-induced dissociation, we first examined the wavelength (λ) dependence of the dissociation yield on Ag(111) (fig. S3). The rate constant divided by the number of incident photons per second is equivalent to the yield of plasmon-induced dissociation (YLSP). The wavelength at the maximum intensity of YLSPMax) and the threshold wavelength of YLSPTh) obtained in area one (Fig. 3A) were ~532 nm (~2.33 eV) and ~780 nm (~1.59 eV), respectively. The plasmon-induced dissociation of (CH3S)2 was also examined on a Cu(111) substrate (Fig. 3B and fig. S4), which has a different electronic structure and plasmonic properties. The YLSP λ spectrum for Cu(111) had λMax = ~670 nm (~1.85 eV) and λTh = ~980 nm (~1.27 eV).

Fig. 3 Wavelength dependence of the plasmon-induced chemical reaction and of the plasmonic electric field.

Wavelength dependence of YLSP of (CH3S)2 molecules on (A) Ag(111) (blue diamonds) and (B) Cu(111) (red circles). Insets in (A) and (B) show YLSP for 700 to 980 nm. Each data point represents the average of results from six trials. The photodissociation yields without the excitation of the LSP [Yphoton, previously reported in (21)] (black circles) are also shown. The yield is determined from k divided by the number of incident photons per second (~6.0 × 1016 to 6.5 × 1016 photons cm−2 s−1 for both plasmon-induced dissociation and photodissociation). Error bars indicate SD. (C) Calculated electric field intensity for a 1-nm gap between a Ag tip and the metal substrates under p-polarized light. The simulated point is z = 0.1 nm above the substrate surfaces and x = 0 nm.

In contrast, the photodissociation yield (Yphoton) measured when the sample was exposed to light with the tip retracted by more than 2 μm from the surface exhibited peak and threshold wavelengths at ~450 nm (~2.76 eV) and ~635 nm (~1.95 eV) on Ag(111) and ~450 and ~670 nm on Cu(111), respectively. Our previous work (21) revealed that photodissociation of the S−S bond in (CH3S)2 molecules adsorbed on Ag(111) and Cu(111) surfaces occurs through direct electronic excitation from the highest occupied molecular orbital (HOMO)– to the lowest unoccupied molecular orbital (LUMO)–derived orbitals of (CH3S)2, that is, from the nonbonding lone pair–type orbitals on the S atoms (nS) to the antibonding orbital localized at the S−S bond (σ*SS) (supplementary text).

The hybridization between (CH3S)2 and the metal substrate reduced the optical energy gap into the range of visible light. Furthermore, LUMO-derived molecular states with less overlap with the metal substrate were formed, which resulted in longer excited-state lifetimes than would be the case for strong overlap with the substrate. Thus, photodissociation occurs through the direct excitation between the frontier MOs, and the shape of the Yphoton λ spectra reflected the densities of states (DOSs) of both the HOMO and LUMO (fig. S5). In contrast, the YLSP λ spectra had a shape similar to that of the simulated Egap λ spectra (Fig. 3C) in the wavelength regions where photodissociation occurs. This similarity indicates that the LSP was an excitation source and that the YLSP strongly depended on the energy profiles of the LSP. A similar wavelength dependence on Ag and Cu was also observed in the enhancement of surface-enhanced Raman scattering (SERS) intensities (22). In addition, the maximum intensity of YLSP on Ag(111) and Cu(111) was ~400 and ~300 times as high as that of Yphoton, respectively, which was caused by the strong enhancement of the electric field by the LSP (Fig. 3C).

In the indirect hot-electron transfer mechanism (Fig. 1A), the YLSP is determined both by the energy distribution of hot electrons and by the DOSs of the LUMOs. Energy distributions of hot electrons and holes generated by the decay of LSPs are sensitive to the electronic band structure of metals (17, 18). If the energy of the LSPs was lower than the threshold energy for direct interband transitions of the metals, both electrons and holes would be equitably distributed from zero to the energy of the LSP through phonon-assisted transitions (18). In contrast, at higher energies, direct transitions dominate, and the relative probability density of hot carriers dramatically increases.

If the plasmon-induced dissociation of (CH3S)2 occurred through hot-electron transfer, the YLSP λ spectra would have exhibited a step change at the threshold energy of the interband transition at ~3.0 and ~2.0 eV for Ag and Cu, respectively (23), and increased at the higher energies, because the LUMO states are distributed broadly above the Fermi level (EF) (fig. S5). In addition, the hot electrons generated by the direct transitions of Cu have a broad energy distribution above EF (18) and can be transferred into the LUMO states, resulting in the excitation of vibrational modes lying along the reaction coordinate. However, the expected spectral change at the threshold energy of the interband transition of the metal was not observed. Moreover, the overall shapes of the YLSP λ spectra reflected not only the shape of the Egap λ spectra (Fig. 3C) but also the shape of the Yphoton λ spectra, that is, the energy distribution of the DOSs for both the HOMO (nS) and the LUMO (σ*SS) (Fig. 3, A and B). Thus, we could exclude the indirect hot-electron transfer mechanism for this plasmon-induced dissociation. Furthermore, hot-hole transfer in Cu, where hot holes are much more energetic than hot electrons (17, 18), could also be excluded by the shape of the Yphoton λ spectrum. We conclude that the LSP efficiently induces and enhances the dissociation reaction through the same reaction pathway as photodissociation (nS → σ*SS), on the basis of the direct intramolecular excitation mechanism (Fig. 1B).

The YLSP λ spectra also had tails extending to longer wavelengths where photodissociation never occurred. This finding suggests that the LSP also enabled direct intramolecular excitation from the MO in-gap states near EF (supplementary text and figs. S5 to S7) to σ*SS (MOin-gap → σ*SS), as well as the HOMO-LUMO transition (nS → σ*SS). The computationally estimated energy gaps between EF and the edge of the LUMO state are ~1.0 and ~0.80 eV on Ag(111) and Cu(111), respectively (fig. S5). This model is consistent with the threshold energy of the YLSP on Ag(111) being greater than that on Cu(111). Dissociation was also apparently induced by direct intramolecular excitation of MOin-gap → σ*SS. The DOS of the in-gap states is much smaller than that of the frontier electronic states (figs. S5 to S7), and thus the direct excitation of MOin-gap → σ*SS is expected to be a much less efficient process than that of HOMO-LUMO transition (nS → σ*SS). However, the LSP generated a strong electric field localized at the interface between the adsorbate and the metal and enabled dissociation even through inefficient excitation pathways (MOin-gap → σ*SS).

A charge transfer from metals to molecules (Fig. 1C) was proposed to explain chemical enhancement effects in SERS (2426) on the basis of LSPs. The Raman intensities of pyridine molecules adsorbed on coinage metals such as Ag, Cu, and Au were enhanced by electron transfer from the highest occupied state (the valence shell s orbital) of the metal near the EF to the unoccupied MOs (24, 25). In addition, the charge transfer mechanism was invoked recently to describe a plasmon-induced chemical transformation (27) in which the electrons near the EF of plasmonic metal nanoparticles were transferred to adsorbed molecules. In the plasmon-induced dissociation of (CH3S)2, the charge transfer from near the EF to σ*SS (EF → σ*SS) was also taken into account to explain the tails of the YLSP λ spectra in the low-energy region because the energy gaps between EF and σ*SS were about 1.0 to 2.0 eV and 0.8 to 1.5 eV on Ag(111) and Cu(111), respectively (fig. S5).

Real-time observation with an STM allowed us to measure the rates of plasmon-induced chemical reactions for single molecules, thereby providing insights into the elementary reaction pathways that cannot be accessed by conventional spectroscopies and the analyses of YLSP λ spectra. It is highly sensitive to the change in the gap distance (d) (28), and thus real-time information can be collected by tracing It under light irradiation (Fig. 4A). Figure 4B shows the current trace when the STM tip was positioned over a target molecule adsorbed on Ag(111) and exposed to p-polarized light at 532 nm. A sudden drop of It reflected the change in d from d1 to d2 (Fig. 4A) caused by molecular dissociation, and the dissociation rate was determined by the inverse of the time required for the plasmon-induced dissociation (tR).

Fig. 4 Real-time STM results for the plasmon-induced chemical reaction and IET-induced reactions of a single molecule.

(A) Schematic illustration of the real-time observation of the plasmon-induced chemical reaction. (B) Current trace for detecting the dissociation event for the target molecule (STM images) on Ag(111) induced by the LSP excited with p-polarized light at 532 nm (~2.7 × 1015 photons cm−2 s−1). The Ag tip was positioned above the molecule marked by the asterisk in the STM image. The fixed gap resistance (Vs = 20 mV and It = 2.0 nA) was applied and the feedback loop was turned off to maintain the tip height at t = 2.0 s. Light irradiation started at t = 4.0 s. The light was turned off and the feedback loop was turned on at t = 18 s. The tunneling conditions at t = 0 to 2.0 s and 18 to 20 s are Vs = 20 mV and It = 0.2 nA. The time required for the plasmon-induced dissociation (tR) is directly read from the current trace. (C) Gap distance d dependence of the dissociation rate (1/tR) obtained under p-polarized light at 532 nm (~1.3 × 1015 photons cm−2 s−1). Each data point represents the average of results for 12 molecules. Error bars indicate SD. (D) Gap distance dependence of the calculated electric field intensity at 532 nm. The simulated point is z = 0.1 nm above the substrate surface and x = 0 nm. (E to H) IET-induced reactions in the dark. Current trace was measured on the molecules with the feedback loop turned off. The tunneling conditions were (E) Vs = 0.30 V and It = 8.0 nA, (F) 0.38 V and 8.0 nA, and (H) 2.3 V and 1.0 nA. The current changes indicated by red and green arrows correspond to rotation and dissociation, respectively. (G) Action spectrum for the rotation and dissociation of the (CH3S)2 molecules induced by injecting tunneling electrons. The initial current was set to 8.0 nA. Each data point represents the average of results from 16 trials. Error bars indicate SD. At 360 mV, the dissociation and the rotation occurred 1 time and 15 times in 16 trials, respectively. The scale bars in the STM images in (B), (E), (F), and (H) are 0.5 nm. Schematic illustrations of the potential energy surface for the plasmon-induced chemical reactions on the basis of (I) the indirect hot-electron transfer mechanism and (J) the direct intramolecular excitation mechanism are shown. Embedded Image, Planck’s constant h divided by 2π Embedded Image multiplied by the angular frequency Embedded Image.

Notably, no reaction could be induced with tunneling electrons at a Vs of 20 mV, and thermal expansion of the tip was negligible because It was stable on the metal surface under light irradiation (fig. S8). The gap distance d (Fig. 4C) is controlled by It, where Embedded Image (Vs = 20 mV; A, coefficient; and Φ, barrier height). We investigated the dependence of d on the dissociation rate (1/tR) at the single-molecule level (Fig. 4C and fig. S9). We note that 1/tR exhibited d dependence similar to that of the calculated electric field intensity (Fig. 4D and fig. S10). This relation indicates that the reaction is caused by the LSP and suggests that the reactivity is determined by the coupling between the electric field of the LSP and a transition dipole moment of the molecule.

In the indirect hot-electron transfer mechanism, the reactions are initiated from the TNI states formed by the hot-electron transfer to molecules via an inelastic electron tunneling (IET) process (8) (Fig. 1A). We can obtain further knowledge of reaction pathways initiated from the TNI states formed by electron transfer from the metal to the molecule via the IET process with the STM (19, 20). In addition, the IET process examined with the STM enables us to describe the elementary processes of vibrational or electronic excitation resulting in molecular motion or reaction, which reflects the local DOS of the adsorbed molecules (28). Both the hot electrons of the LSP and the tunneling electrons had a broad energy distribution from the EF to the energy of the LSP (17, 18) and the applied Vs, respectively (fig. S11). Thus, the reaction pathways in the same energy regions should be the same regardless of the excitation source: hot electrons or tunneling electrons. Rotation (Fig. 4E) and dissociation (Fig. 4F) of (CH3S)2 on Ag(111) were induced through vibrational excitation with inelastically tunneled electrons at energies higher than ~0.28 and ~0.36 eV, respectively (Fig. 4G). Dissociation occurred through vibrational excitation of the C–H stretch mode (19) and a combination of the C–H stretch and the S–S stretch modes (20). Moreover, dissociation induced with tunneling electrons was accompanied by rotation, which resulted in small changes in It followed by its sudden drop (Fig. 4F). Rotation before dissociation was also observed in the dissociation of O2 on Pt(111) through vibrational excitation via the IET process with an STM (29).

We conclude that the energy of the TNI states formed by electron transfer from the metal to the molecule was dissipated to the vibrationally excited states according to the nondissociative potential energy surface (Fig. 4I and fig. S12), which resulted in both rotation and dissociation of (CH3S)2, and rotation is a precursor for dissociation. The small changes of It caused by rotation before its sudden drop were also observed for the dissociation reaction induced with tunneling electrons at 2.3 V (Fig. 4H) and 2.0 and 1.0 V (fig. S13), an energy region almost equal to that of the hot electrons of the LSP excited with light at 532, 620, and 1240 nm, respectively. Thus, the IET process through vibrational excitation at the lower energy region is always included in the reaction pathway when the electrons are distributed from the EF to the upper energy level, which is higher than the vibrational energies of the ground state.

If the plasmon-induced dissociation proceeded through vibrational excitation, current changes caused by rotation should appear before the dissociation. However, changes in It before the sudden drop were not observed (Fig. 4B and fig. S9). This difference excluded the hot-electron–mediated process, which always involved vibrational excitation in the TNI state (Fig. 4I), as an elementary pathway for the plasmon-induced dissociation of (CH3S)2. By considering the mechanistic insight obtained from the YLSP λ spectra (Fig. 3), we concluded that the plasmon-induced dissociation of (CH3S)2 occurred through the direct dissociation pathway from neutral excited states generated by direct intramolecular excitation (Fig. 4J). In addition, if the potential energy surface of the excited state formed by charge transfer from the metal states to σ*SS was dissociative for the S–S bond cleavage (fig. S12C), the charge transfer mechanism could also contribute to the plasmon-induced dissociation in the low-energy region (supplementary text). However, it is difficult to quantify the contribution of the charge transfer mechanism, because the excited state formed by charge transfer (fig. S12C) cannot be simply described in the STM experiment because of the broad energy distribution of the tunneling electrons.

The LUMOs of (CH3S)2 were weakly hybridized with the metal substrates (21). The weak hybridization suppressed the relaxation of the excited state and thus allowed access to the dissociative potential energy surface from the neutral excited state (Fig. 4J), which was theoretically predicted for the photodissociation of (CH3S)2 molecules in the gas phase (30). In contrast, the MOs of O2 and H2, for which plasmon-induced dissociation was explained by the indirect hot-electron transfer mechanism, were strongly hybridized with metal substrates (14, 15). This suggests that the degree of hybridization between MOs and metals plays a crucial role in the plasmon-induced chemical reactions.

Our real-space and real-time STM study combined with theoretical calculations revealed that the plasmon-induced dissociation of the S–S bond in a single (CH3S)2 molecule on Ag and Cu surfaces occurred principally by the direct intramolecular excitation to the LUMO state of the antibonding S–S (σ*SS) orbital through a decay of the optically excited LSP in the nanogap between the Ag tip and the metal surface. The present results underline that the plasmon-induced chemical reactions of the molecule with the electronic states less hybridized with metals are explained by the direct intramolecular excitation mechanism but not by the indirect hot-electron transfer mechanism. These findings provide deep insights into the interaction between LSPs and molecules at metal surfaces for designing efficient plasmon-induced photocatalysis in a highly controlled fashion.

Supplementary Materials

www.sciencemag.org/content/360/6388/521/suppl/DC1

Materials and Methods

Supplementary Text

Figs. S1 to S13

Table S1

References (3151)

References and Notes

Acknowledgments: We appreciate Y. Hasegawa for supporting the preparation of Ag tips. We thank J. Yoshinobu for helpful discussions. Funding: The present work was supported in part by a Grant-in-Aid for Scientific Research (A) (15H02025), a Grant-in-Aid for Young Scientists (B) (16K17862), and the RIKEN postdoctoral researchers (SPDR) program. J.J. acknowledges the financial support of the National Research Foundation under the Next Generation Carbon Upcycling Project (grant 2017M1A2A2043144) of the Ministry of Science and Information and Communication Technology, Republic of Korea. H.U. was supported by Ministry of Education, Culture, Sports, Science and Technology Grants-in-Aid for Scientific Research (KAKENHI) grant C-2539000, which allows him to continue to work after retirement from the University of Toyama. M.T. acknowledges support from a grant from the NSF (CHE-1464816). We are grateful for the use of the HOKUSAI-GreatWave supercomputer system of RIKEN. Author contributions: E.K. designed and carried out the experiments and the finite-difference time-domain calculations. J.J. carried out the density functional theory calculations. E.K., J.J., H.U., M.T., and Y.K. contributed to the interpretation of the results and wrote the manuscript. Y.K. planned and supervised the project. Competing interests: The authors declare no competing financial interests. Data and materials availability: All data are presented in the main text and supplementary materials.
View Abstract

Stay Connected to Science

Subjects

Navigate This Article