Plasmon damping depends on the chemical nature of the nanoparticle interface

Damping of gold nanorod plasmons by surface-adsorbed molecules is best explained by scattering off adsorbate-induced dipoles.


INTRODUCTION
Localized surface plasmons, the light-induced collective oscillation of conduction band electrons in metal nanoparticles, have had a tremendous impact on the field of chemistry: chemical-specific sensing became possible via surface-enhanced Raman scattering since its discovery in the 1970s (1)(2)(3)(4)(5), while surface plasmon-mediated hot carrier generation recently produced a subfield of photochemistry (6,7). Hot electrons and holes can selectively catalyze chemical reactions through charge or energy transfer to surface adsorbates by carefully matching plasmon resonance energies with molecular energy levels (8)(9)(10)(11)(12)(13). Nevertheless, plasmons are described exclusively by electromagnetic theory that treats the surrounding chemical environment as an effective medium (14,15), neglecting any dependence on chemical composition. Although the influence of adsorbates on nanoparticle plasmons has been noted as far back as the 1980s (15)(16)(17), the even stronger size and shape dependence of plasmons coupled with ensemble measurements have thus far prevented a consistent, quantitative analysis and theoretical understanding.
A localized surface plasmon decays through radiation or by scattering with bulk defects and at the nanoparticle surface (18,19). The time scale of this dephasing is on the order of at most a few tens of femtoseconds (18), making direct time-resolved measurements difficult (20,21). An equivalent and more common approach to quantifying plasmon decay is to evaluate the resonance linewidth and intensity (18,(22)(23)(24), as long as single-particle spectroscopy is used to yield nanoparticle intrinsic values free from ensemble averaging of heterogeneous size and shape distributions. When molecules chemically bind to the surfaces of metal nanoparticles, their plasmon decay can be accelerated as manifested by additional damping (i.e., increase in linewidth and decrease in intensity), referred to as chemical interface damping (15,23,24). However, plasmon damping by molecular adsorbates is still poorly understood and is often simply invoked when all other explanations fail. The obstacles hindering a quantitative understanding of chemical effects on plasmon damping are rooted in both experiment and theory. The latter requires a quantum mechanical approach for the electronic structure of molecules adsorbed on a metal surface, while the former suffers from the complexity that the different damping channels are highly interrelated (18,19,25,26). Specifically, bulk damping depends on the plasmon resonance energy, which, in turn, is affected by the refractive index of the surrounding medium (25). It is therefore difficult to isolate the difference in only chemical interface damping caused by different adsorbates.

RESULTS AND DISCUSSION
To overcome these challenges, we studied two chemical isomers with the same molecular size: 1,7-dicarbadodecaborane(12)-1-thiol and 1,7-dicarbadodecaborane(12)-9-thiol (M1 and M9 in Fig. 1B, respectively). These carboranethiols have the same size, geometry, and binding group, and scanning tunneling microscopy revealed that they adsorb with the same density on gold surfaces (27,28). M1 and M9 are, however, chemically different. Both carboranethiols consist of an icosahedral boron cage, in which two boron atoms are replaced by positively charged carbon atoms (27)(28)(29). The sulfur atom in M1 is attached to a carbon atom giving it more electronegative character and higher acidity (pK a = 5.30) compared to M9 (pK a = 9.45), in which the sulfur atom is attached to a boron atom instead (29). By comparing the change in homogeneous plasmon linewidth of single gold nanorods due to adsorption of these two carboranethiols, we are able to quantitatively determine chemical interface damping and interpret the mechanism based on calculations obtained with embedded correlated wavefunction (ECW) theory. In particular, we find that charge transfer, invoked as the main explanation of chemical interface damping in many instances before (7,12,30), does not play a role here. Instead, the plasmon is damped by scattering off dipoles induced by the carboranethiols.
Plasmon damping was determined from the linewidth  of singleparticle dark-field scattering spectra (Fig. 1A). Because we measured the same individual gold nanorods (22 ± 2 nm by 66 ± 4 nm; section S1) after removal of the initial surfactants and during the entire carboranethiol adsorption process (see Methods), we only need to consider changes in linewidth ∆ and can equate them directly to plasmon damping by M1 and M9 while ignoring all other energy loss pathways. A small difference in resonance energy shift ( fig. S4) can be neglected for plasmon energies below 1.76 eV (31). We observed the plasmon linewidth of 115 and 130 single gold nanorods for more than 120 min during M1 and M9 carboranethiol adsorption, respectively (Fig. 1C). The plasmon linewidth broadened over time t, (t), following kinetics typical for a Langmuir adsorption isotherm, i.e., (t) =  ∞ [1 − exp(−kt)] (23). On the basis of Langmuirian kinetics, we obtained the plasmon broadening of fully carboranethiol-covered gold nanorods at infinite time  ∞ and the adsorption time constant k. An alternative way to visualize the difference in plasmon damping for the M1 and M9 carboranethiols is to plot the histograms of all single-particle linewidths measured before and after 120 min (section S3).
M1 carboranethiols damp plasmons 56% more strongly than M9 carboranethiols. We obtain values of  ∞,M1 = 25.4 ± 0.4 meV and  ∞,M9 = 16.3 ± 0.4 meV for gold nanorods covered with M1 or M9 carboranethiols (Fig. 1D). The relative scattering intensity was also reduced to 65 and 75% after t = 114 to 120 min and t = 116 to 123 min for the M1 and M9 carboranethiols, respectively ( Fig. 1E) (23). The larger reduction in scattering intensity for M1-coated gold nanorods is in quantitative agreement with the stronger damping inferred from the increase in linewidth (23). Consistent changes in both linewidth and intensity as well as comparable resonance energy shifts ( fig. S4) support the conclusion that the plasmon more strongly interacts with the more electronegative M1 carboranethiols.
The adsorption kinetics of both carboranethiols are highly reproducible and the same within our experimental error, implying a comparable coverage of adsorbed molecules. Complementary analytical methods that determine the density of adsorbates on individual gold nanoparticles and are applicable to our experimental measurement scheme do not exist (32). We therefore used the adsorption kinetics of the plasmon broadening as an intrinsic control to exclude that differences in  ∞ were not caused by different numbers of adsorbed molecules. The adsorption time constants k for the M1 and M9 carboranethiols (1/k M1 = 46 ± 2 min and 1/k M9 = 50 ± 3 min) are the same within their errors (Fig. 1C). The time constants k were reproducible in six independent adsorption experiments for both carboranethiols ( Fig. 2A), while, at the same time, plasmon broadening was always stronger for M1 than M9 (Fig. 2B). This conclusion that M1 and M9 similarly adsorb to the gold nanorods is consistent with previous scanning tunneling microscopy measurements (27,28), especially when considering studies that have found gold nanorod surfaces to be mostly flat (33) and molecular coverages to be comparable between gold nanorods and bulk surfaces (34).
Plasmons such as electrons scatter at defects (internal and at the particle surface), phonons, and electrons, losing their initial momentum and energy (Fig. 3A) (15,18). The frequency of these scattering events is described by the metal-specific electron mean free path and the effective particle size in the case of nanostructures. Nordlander and co-workers (35) proposed that electron-phonon relaxation in metal nanoparticles is enhanced when surface adsorbates induce electric dipoles at the nanoparticle surface. These dipoles act as additional scattering centers, and relaxation times were found to scale with the magnitude of the induced dipole moments. We therefore consider here the possibility that induced dipoles are created when the carboranethiols adsorb to the gold nanorods and similarly lead to accelerated plasmon decay.
Using ECW theory, we indeed find that dipoles are induced by the carboranethiols at the nanoparticle interface and that their magnitude and orientation differ for M1 and M9. Details of the calculations and the values of all electric dipoles are given in the sections S4 and S5,  Fig. 3 (B to D) summarizes the main result. Both carboranethiols have permanent electric dipole moments due to the substituted carbon atoms in the boron cage (27,28). When the carboranethiols adsorb on a metal surface, the freely moving conduction band electrons respond to the molecular dipole by creating image dipoles. The magnitude and orientation of these adsorbate-induced electric dipoles in the metal (Fig. 3D) are obtained by subtracting the electric dipoles of the carboranethiols in the gas phase (Fig. 3C) from the electric dipoles calculated for carboranethiols on an embedded gold cluster (Fig. 3B). It is justified to omit the ethanol solvent because it will interact only weakly with both the Au surface (ethanol only physisorbs on Au) and these inorganic boron carbides M1 and M9, resulting in a random orientation of solvent molecules with no net polarization.
Interactions between adsorbate-induced surface dipoles and the dipole of the experimentally probed longitudinal plasmon mode are maximized when they line up parallel to each other. To quantitatively assess the difference in plasmon damping for M1 and M9, we there-fore consider the magnitude of the dipole component parallel to the long nanorod axis. We find that the induced electric dipoles along the gold surface plane  xy are 4.03 and 2.74 D for M1 and M9 carboranethiols, respectively. The ratio of their magnitude  xy,M1 / xy,M9 = 1.47 is in quantitative agreement with the experimental observed plasmon broadening ratio  ∞,M1 / ∞,M9 = 1.56. We have considered here only the carboranethiols adsorbed at the nanorod sides because Zijlstra et al. (24) previously showed greatly reduced damping of end-only functionalized gold nanorods compared to functionalization at the sides and ends (see also section S1). It is also important to mention that these induced dipole moments, of course, do not directly predict the rate of electron scattering.
A long-standing theory by Persson (17), which links chemical interface damping with charge transfer into empty electronic states created via surface adsorption, agrees qualitatively with our results. However, this theory fails to quantify the difference in damping by M1 and M9 carboranethiols when we calculate, using density functional theory (DFT), the adsorbate density of states (see section S6). For the carboranethiols studied here, we therefore find no evidence for charge transfer to the adsorbate molecules following plasmon excitation. We cannot exclude though that a charge-transfer mechanism might dominate chemical interface damping in other systems.
We have shown by comparing the surface adsorption of two chemical isomers that, for metal nanoparticles, the nature of the chemical interface affects the loss of energy stored in a plasmon. Quantitative agreement was achieved with ECW theory calculations that correlate the degree of plasmon damping with the magnitude of the dipole moment induced in the metal parallel to the plasmon oscillation. The induced surface dipoles act as additional scattering centers for plasmon dephasing. Our results pave the way for a more detailed understanding of the chemical interface of plasmonic nano particles, necessary for the design and optimization of plasmon-driven chemistry. fully M1 (red) and M9 (blue) carboranethiol-coated gold nanorods from three independent experiments (#1, #2, and #3) obtained from Langmuir adsorption kinetics. The plasmon linewidth broadening ∆ ∞ differs between M1 and M9 carboranethiols but is consistent among the three experiments carried out for each carboranethiol. Corresponding adsorption isotherms with Langmuirian kinetics are given in section S3.  (15,18). Surface adsorbates create additional scattering centers in the form of induced dipoles (35). The scattering efficiency depends on the average distance an electron needs to travel to reach the induced dipoles, i.e., the nanorod surface (15,18,23,51). (B to D) ECW theory yields the total dipole moment  total of the M1 and M9 carboranethiols adsorbed on gold (B; red arrows) and in the gas phase (C; yellow arrows). The difference between these dipole moments gives the induced dipole moment in gold (D; violet arrows) due to carboranethiol adsorption (section S5).

Synthesis of gold nanorods
Gold nanorods were synthesized using the method by Ye et al. (36). The amounts of chemicals were used as given in the supporting information by Ye et al. (36) for the 22 nm by 66 nm gold nanorods also yielding 22 nm by 66 nm gold nanorods in our synthesis. Dimensions of gold nanorods were estimated by analyzing transmission electron microscope (FEI Tecnai G2 Spirit Twin) images using an image recognition tool of the MATLAB software. Details of the average dimensions of the synthesized gold nanorods are given in section S1.

Single-particle spectroscopy
Scattering spectra of single gold nanorods were obtained using a custom-built dark-field microscope equipped with a spectrometer. Details of the custom-built microscope are given in section S2. A microfluidic cell was used in the experiments to change the liquid around the gold nanorods. Details regarding the fabrication of the microfluidic cells are given below. A low concentration solution of 22 nm by 66 nm gold nanorod was flushed through the microfluidic cell. Then, ethanol was flushed through the cell and gold nanorods attached to the glass substrate. Thirty-one to 46 gold nanorods were selected, and scattering spectra were obtained from each selected gold nanorod. These spectra were used as reference. Then, a 3 mM ethanolic solution of M1 or M9 carboranethiol was flushed through the microfluidic cell, and the spectra of the selected gold nanorods were repeatedly obtained for about 120 min. This experiment was repeated three times for both carboranethiols. The results of each separate experiment are given in section S3.

Microfluidic cell
Two coverslips #1 (Menzel-Gläser) were connected with parafilm as a spacer. A CO 2 laser cutter (Trotec Speedy 100) was used to cut channels in the parafilm and holes into the coverslips. The holes in the coverslip were placed at the end of each channel in the parafilm. The microfluidic cell was sealed by placing it on a hot plate at 120°C for a few seconds and applying pressure. Silicon tubings (NeoLab) were attached to the holes in the coverslip, and liquid was pulled through the cell using syringes.
Theoretical methods ECW theory was used for the calculations to obtain the dipole moments via density functional embedding theory (DFET) (37). The correlated wavefunction method used was the complete active space self-consistent field method (38). Starting structures for ECW calculations and calculations to obtain the density of states were obtained from periodic slab DFT using the Perdew-Burke-Ernzerhof exchangecorrelation functional (39) with D3 dispersion corrections (40,41). Hartree-Fock theory (42) was used to obtain the gas-phase dipole moments. A comprehensive summary of the protocol is included in section S4. Vienna Ab Initio Simulation Package (43)(44)(45)(46) was used for the periodic DFT and DFET optimization of the embedding potential using an in-house embedding potential optimization code (47). Molpro (48) was used for the ECW and gas-phase calculations. Additional software used includes VESTA (49) version 3.2.1 for visualizing densities and the calculated embedding potential, VMD (50) for plotting and rendering the dipoles, and MATLAB for data analysis.