Deducing the Adsorption Geometry of Rhodamine 6G from the ......scattering cross section is then computed for each normal Figure 1. Procedure for computing the Raman spectra. (1a)

Deducing the Adsorption Geometry of Rhodamine 6G from theSurface-Induced Mode Renormalization in Surface-Enhanced RamanSpectroscopyColin Van Dyck,*,†,∥ Bo Fu,‡,∥ Richard P. Van Duyne,§ George C. Schatz,§ and Mark A. Ratner*,§

†National Institute for Nanotechnology (NINT), 11421 Saskatchewan Drive NW, Edmonton, AB, T6G 2M9, Canada‡Department of Physics and Astronomy, Northwestern University, 2145 Sheridan Road, Evanston, Illinois 60208, United States§Department of Chemistry, Northwestern University, 2145 Sheridan Road, Evanston, Illinois 60208, United States

*S Supporting Information

ABSTRACT: Surface-enhanced Raman spectroscopy probes adsorbates on a plasmonicsubstrate and offers high sensitivity with molecular identification capabilities. In thisstudy, we present a refined methodology for considering the supporting substrate in thecomputation of the Raman spectra. The supporting substrate is taken into account byemploying a periodic slab model when doing the geometry optimization and normalmode analysis, and then the Raman spectrum is calculated for the isolated molecule butwith the normal modes from the surface structure. We find that the interaction with thesurface induces internal distortion in the molecule, and spectral shifts in the computedRaman spectrum. By comparing a low temperature surface-enhanced Ramanspectroscopy measurement of Rhodamine 6G (R6G) with the computed Ramanspectra of a series of adsorption geometries, we propose that the binding state captured in the experiment tends to possess theleast internal distortion. This binding state involves upward orientation of ethylamine on R6G, and our calculations indicate thatthis is the lowest energy adsorption structure. Following this route, it is possible to infer both a molecular orientation and anadsorption geometry of the molecule from its Raman spectrum. Importantly, we note that, if the substrate correction isestablished to play a role, we also discuss that this corrected approach still has several shortcomings that significantly limit itsoverall accuracy in comparison with experimental spectra.

■ INTRODUCTION

Surface-enhanced Raman spectroscopy (SERS)1,2 is a well-established and powerful spectroscopic tool for acquiringvibrational properties of molecules adsorbed on surfaces ofplasmonic nanostructures such as nanoparticles (NP), metallicfilms on nanosphere (MFON), and so on. It is well-known thatthe Raman scattering signal in SERS is significantly amplifiedby many orders of magnitude (compared to the traditionalRaman experiment of molecules in solution) due to the hugeelectromagnetic enhancement offered by the plasmonicsubstrate.3 When combined with the resonance enhancementof some dye molecules (due to molecular electronicexcitation), SERS can detect and identify the Raman spectrumof a single molecule. Rhodamine-6G (R6G) adsorbed on asilver nanoparticle (AgNP) is such a system, regarded as thefirst success of single molecule SERS (SMSERS) detection.4,5

The absorption spectrum of R6G shows a strong peak at 530nm and thus enables a strong enough Raman signal for singlemolecule detection with the commonly used 532 nm laser.This makes it a perfect testbed system for developing newcharacterization techniques relying on Raman spectroscopy.Due to its active role in the study of SMSERS, the Ramanspectra of R6G have been extensively studied bothexperimentally6−16 and theoretically.17−20

Recently, R6G has been used in combination with SERS todemonstrate single (few) molecule electrochemistry experi-ments.15 To interpret these results, we showed15,21 that R6Gcan adsorb in many different geometries on a silver surface.This leads to a large distribution of reduction potentials, asobserved experimentally.15 For each single molecule electro-chemical event, the reduction of R6G is monitored bywatching the SMSERS signal loss. Although a relationship isestablished between adsorption geometries and the reductionpotential distribution, the association of an individualreduction event with a specific binding geometry is stilllacking. Such a relationship at the single molecule level is ofgreat interest both for understanding the electron transferprocess at the nanoscale and for the design of highly efficientelectrocatalytic electrodes.In principle, the SERS signal of an adsorbed molecule is

directly correlated with adsorption geometry. Accordingly, onecan imagine identifying this geometry from a careful analysis ofthe Raman spectrum. To do so by a comparison of theory andexperiment, computation of the Raman spectrum requiresconsideration of the effect of the metal substrate. However, the

Received: September 22, 2017Revised: December 8, 2017Published: December 11, 2017

Article

pubs.acs.org/JPCCCite This: J. Phys. Chem. C 2018, 122, 465−473

© 2017 American Chemical Society 465 DOI: 10.1021/acs.jpcc.7b09441J. Phys. Chem. C 2018, 122, 465−473

currently reported Raman spectra of R6G in the literature wereonly calculated for its gas phase geometry,16−18,20 and in fact itis extremely challenging to calculate Raman spectra thatinclude more than a few atoms in the metal substrate.In this study, we consider the effect of substrate in the

computation of Raman spectra with a slab model (Figure 1).Our computational procedure considers interaction of R6Gwith the substrate during geometry optimization and thenormal modes evaluation. This allows for investigating theeffect of adsorption geometry variations on the Ramansignature of R6G, as adsorbed on a Ag-(111) surface. Theconsideration of the interaction with such a large substrate liesat the current edge of the ab initio normal mode capabilities foradsorbed molecules, due to a high computational cost.Although the interaction with the substrate is still neglectedin the computation of the molecular polarizabilities, theinduced changes in the normal mode frequencies are captured.We observe that the interaction with the substrate maysignificantly distort the molecule and consequently inducespectral shifts in the corresponding Raman spectrum. Bycomparing our several results with low temperature SERS (LT-SERS) experiments for R6G done in ultrahigh vacuum,14 weshow that the least distorted configuration, which also has thestrongest adsorption, is the most likely one to be probedduring the SERS experiment. This fact offers a pathway to infergeometrical details of the adsorbate, allowing us to exploit theinformation contained in the Raman signal to determine theadsorption geometry of the molecule. However, thequantitative prediction of the Raman spectrum is still limitedby a few shortcomings that exist at both theoretical andexperimental levels.

■ METHODOLOGY AND COMPUTATIONAL DETAILS

In the commonly reported procedure for calculating theRaman spectrum, the optimized geometry and normal modesof the molecule are obtained in a first computational step. Thepolarizability derivative and the Raman scattering cross sectionof each normal mode are then computed with a linear responsetime-dependent density functional theory (LR-TDDFT)method22,23 in a second step.22,24 This procedure has beenapplied for commonly used dye molecules in SERS, e.g., R6Gand Nile Blue, and good agreement with experimental spectrahas been demonstrated.25,26 In this study, we refine the firstcomputational step by including the interaction between themetal surface and the adsorbed molecule.

To describe the substrate, either a cluster model or aperiodic slab model could be considered. Intuitively, a clusteris a reasonable atomistic model for a AgNP. However, we optfor a slab model that we used in a previous study.21 This ismotivated by three main reasons. First, the AgNPs used in theSERS experiment range from 20 to 100 nm in diameter.15

Nonetheless, only clusters made of a small number of metallicatoms are tractable computationally. For example, severalstudies of the pyridine-Ag20 system using ab initio methodshave been reported in the literature.27−31 Although thesestudies provided some fruitful insights into the molecule−substrate interaction, a Ag cluster exhibits a discrete electronicspectrum. On the contrary, the nanoparticles employedexperimentally are much larger than those consideredpreviously, and possess a bulk-like electronic structure.32,33

Therefore, a cluster approach will not be able to accuratelyreproduce the energetic state of the adsorbate. Second, thesmall size of R6G (∼1 nm) makes the periodic slab a goodapproximation of the nanoparticle surface. Third, it has beendemonstrated experimentally that transition metal nano-particles exhibit well-defined crystal facets.34−36 Noteworthy,the (111)-like facet is frequently observed for silver nano-particles.37−39

The slab model is introduced using periodic boundaryconditions, with a unit cell that includes the R6G molecule andtwo (111) silver layers (144 atoms) and of 25.8 × 19.9 Å2

dimensions. These layers are frozen during the geometryoptimization and the normal mode evaluation. The consid-eration of such a large sized silver surface might constitute acurrent limit in terms of computational resources. We rely on aplane-wave basis set as implemented in the Vienna Ab initioSimulation Package (VASP).40,41 The electronic structure iscomputed at the DFT/PBE level of theory.42 The missingdispersion effect is corrected by the Tkatchenko-Scheffler (TS)scheme.43 This dispersion correction scheme has proven itsvalidity in our previous study21 which overcame the failure ofPBE in describing the physisorption of R6G on the Ag surface.More details on this computational procedure are given in ourprevious study and in the Supporting Information. All thenormal mode frequencies are rescaled by a factor of 1.01 toaccount for the missing anharmonicity. This rescaling factorwas determined at our level of theory by following anestablished and widely used procedure44−46 on the basis of amolecule database set compiled by Alecu et al.46 (see theSupporting Information for details). The resonant Ramanscattering cross section is then computed for each normal

Figure 1. Procedure for computing the Raman spectra. (1a) Geometry optimization of the adsorbed R6G using a silver slab surface and periodicDFT with PBE XC functional plus the van der Waals dispersion correction proposed by Tkatchenko and Scheffler. (1b) Normal mode evaluationusing the same modeling and theory as in step (1a). (2a) Computation of the polarizability derivative for each normal mode of R6G withoutincluding the surface, using the B3LYP XC functional. (2b) Calculation of Raman scattering cross sections from the polarizability derivativesobtained in step (2a). The complete formula for the Raman scattering cross section is discussed in the Supporting Information.

The Journal of Physical Chemistry C Article

DOI: 10.1021/acs.jpcc.7b09441J. Phys. Chem. C 2018, 122, 465−473

466

mode at the TDDFT/B3LYP level of theory, using a 6-311G** basis set, as implemented in the AOResponse moduleof the NWChem DFT package. An excitation with a laserwavelength of 532 nm is used for the consistency with theexperiment.23,47,48 The scattering cross section is computed forthe isolated R6G with the adsorption geometry obtained in thefirst computational step, after the removal of the silver surface.The silver slab is excluded here due to the extremecomputational cost. However, this approximation is not

expected to lead to large deviations in the overall Raman

signals because the component of chemical enhancement due

to charge transfer, which is usually observed upon covalent

bonding, can be neglected for physisorbed molecules like

R6G.27 Still, due to the lack of surface enhancement in the

calculation of the polarizability derivatives, we stress that our

proposed method does not intend to quantitatively reproduce

the relative intensities of the characteristic Raman peaks.

Figure 2. (a) Raman spectrum of R6G taken from the chosen LT-SERS experiment (Adapted with permission from ref 14. Copyright 2014American Chemical Society). Nine characteristic modes are labeled alphabetically. Computed Raman spectra of (b) isolated R6G, (c) adsorbedR6G on Ag slab in ethylamine down geometry, (d) adsorbed R6G on Ag slab in twisted flip geometry, and (e) adsorbed R6G on Ag slab inethylamine up geometry. Characteristic modes are marked with their corresponding frequencies in figures from (a) to (e).

The Journal of Physical Chemistry C Article

DOI: 10.1021/acs.jpcc.7b09441J. Phys. Chem. C 2018, 122, 465−473

467

■ RESULTS AND DISCUSSION

In this study, we compare our simulations with the Ramanspectrum of R6G obtained in a low temperature, ultra-high-vacuum SERS (LT-SERS) experiment.14 By choosing LT-SERS, we aim at simplifying the investigation of theadsorbate−substrate interaction by excluding solvent andthermal effects which are present in ambient, room temper-ature SERS. These effects make it much more difficult to findthe source of inconsistency between simulations and theexperiments. In our atomistic model of R6G-AgNP, theinterslab space is assumed to be vacuum. Also, the optimizedgeometry obtained by the conjugate gradient algorithm isexpected to correspond to the ground state at absolute zerotemperature. Thus, our atomistic model agrees with theexperimental conditions associated with LT-SERS. Note-worthy, the chosen low temperature Raman spectrum alsoagrees reasonably well with room temperature ambient SERSexperiments for R6G.15,16

The experimental spectrum of the chosen LT-SERSmeasurement (green line) is presented in Figure 2a. Ninecharacteristic peaks of R6G (at 1652, 1608, 1579, 1512, 1364,1318, 1187, 775, and 614 cm−1) are marked by correspondingwavenumbers and labeled from A to I for convenience. Thesevibrational modes are chosen because they can be recognizedin SERS experiments under different conditions.4,5,14,16

We start the discussion with an analysis of the Ramanspectrum of the isolated R6G in its gas phase geometry. Thiscalculation excludes the silver surface in the first computationalstep. We note that this calculation assumes a starting geometrywith the ethylamine groups pointing in the direction that isopposite to the tail of the phenyl group, as highlighted by acircle in Figure 5. This geometry has been commonly used inprevious theoretical characterizations of R6G.17,18,20 Asillustrated in Figure 2b, the computed Raman spectrum ofisolated R6G presents a spectrum that matches well with theexperiment. All of the nine characteristic peaks can berecognized from the computed Raman spectrum. The modefrequencies differ from the corresponding experimental valuesby less than 6 cm−1 for 6 of the 9 modes. Only the reportedfrequencies for mode A and mode G are off by more than 10cm−1. We note that mode G may also be a combination of twomodes that lie at 1173 and 1201 cm−1, as discussed below. Thecomputed frequency of mode C is off by 9 cm−1, if we attributeit to the mode at 1569 cm−1. We will further discuss thesethree modes in the next paragraphs.The success of the isolated molecule model is consistent

with several computed Raman spectra of R6G using similarmethodologies.17,18,20 This suggests that the adsorbateobserved in the experiment should not differ significantlyfrom its gas phase geometry. This indication provides us ahelpful guideline in selecting the most appropriate adsorptiongeometry for computing the Raman spectrum of R6G, aspresented below.Before proceeding to the calculation that includes the metal

surface, we stress that, due to the neglect of the substrate in thecomputation of polarizability derivatives, our comparisonbetween simulated Raman spectra and LT-SERS is mainlyfocused on the spectral positions of the chosen characteristicpeaks. Despite being used as a well-established technique todetect and identify molecular adsorbates, we acknowledge thatSERS experiments can suffer from reproducibility issues,including the position of characteristic peaks. This is especially

true considering that different substrates or solvents areregularly used in different experimental setups. In this context,it is important to highlight here that we focus on a specificsubstrate,14 in ultra-high-vacuum conditions, which greatlyincreases the reproducibility of SERS experiments.Moreover, the RT-SERS (room temperature) spectrum is

compared to the LT-SERS spectrum in our experimentalreference.14 This comparison shows a limited decrease of thepeak broadening about 12% on average. As some broadeningremains at low temperature, it is not of thermal nature.Importantly, this broadening is further reduced significantly,about 50% on average, when using the LT-TERS character-ization technique. As this technique intrinsically reduces thenumber of sampled molecules (≤104 in TERS vs ∼109 inSERS) by a very large factor, it is reasonable to assume that thebroadening of the peaks in the LT-SERS experiment is givenby an accumulation of signals emitted by molecules in slightlydifferent adsorption geometries.Since the isotropic contributions of the polarizability

derivatives are dominant in SERS,24 the Raman scattering isnot sensitive to the relative molecular orientation and we canreasonably assume that the ensemble of molecules probedduring the experiment14 (∼109) generates an equivalentscattering intensity. For these reasons, we assume here thatthe SERS experimental Raman signal is dominated by the mostenergetically favorable configuration and the binding stateenergy plays an important role to determine this configuration.Our choice of the Ag-(111) slab is thus further justified as it isthe most stable exposed facet of the AgNP surface.34−36

The crooked shape of R6G leads to several differentadsorption geometries as was suggested in our previous work.21

Regarding the molecular orientation, our previous studyrevealed that the “parallel” geometry of R6G (shown in Figure5c) is the most stable binding state on the Ag-(111) surfaceamong a large number of tested configurations.21 Therefore,this binding state is chosen as the first attempt to simulate theRaman spectrum for LT-SERS. The fact that the “parallel”geometry is the least distorted geometry among all the possiblebinding geometries we tested on the Ag-(111) surface21 alsomakes it a reasonable choice for assessing the Raman signatureof R6G captured in the experiment. In the later discussion, this“parallel geometry” will be referred to as the “ethylaminedown” configuration for convenience, because its two ethyl-amine groups point downward onto the silver surface.We note that the geometries in our previous study were

assessed “on the fly”; i.e., we optimized the geometry of themolecule adsorbed on the surface, starting from a set ofintuition-driven geometries. Then, we compared the obtainedset of adsorption energies, after convergence of theoptimization algorithm. A better approach would be a “thermalannealing” procedure, allowing for a direct determination of aglobal minimum structure, given by the possibility to overcomepotential energy barriers. Unfortunately, the size of our systemrenders molecular dynamics approach computationally intract-able. We then envision the determination of the adsorptiongeometry as a trial and error procedure, guided by theadsorption energy and the comparison between simulated andmeasured Raman spectra. This is the procedure that we followin this paper.From a calculation that includes the metal surface, we

compute the Raman spectrum of the “ethylamine down”configuration, as represented in Figure 2c (with comparison tothe isolated molecule calculation). This shows that interaction

The Journal of Physical Chemistry C Article

DOI: 10.1021/acs.jpcc.7b09441J. Phys. Chem. C 2018, 122, 465−473

468

with the surface does not lead to obvious changes in the overallRaman spectrum concerning the spectral locations and therelative intensities of the modes (when neglecting the substratein the computation of the polarizability derivatives). Thisgenerally indicates that the “ethylamine down” configuration isa reasonable assumption for the R6G configuration probed inSERS.By comparing Figure 2c to the isolated molecule spectrum

Figure 2b, only modes A and I show large mode frequencychanges (>10 cm−1), apart from mode C that is discussedbelow. The “ethylamine down” configuration gives these twomodes at 1655 and 619 cm−1, respectively. The accuracy ofmode A is significantly improved while mode I is still off theexperiment by 4 cm−1 (vs 5 cm−1 for the isolated moleculecalculation). According to the mode visualization obtained forthe isolated molecule model (Figure S2), mode A correspondsto a xanthene plane deformation and mode I is associated within-plane stretching in the xanthene plane and the phenyl ring.Our assignment is consistent with previous studies.14,17,18

Therefore, the spectral shift of mode A is an indicator of thesubstrate−adsorbate interaction, which is dominated by thexanthene/silver van der Waals interactions, as also noted in ourprevious work.21 The spectral shift of mode I may be affectedby the distortion of the molecule as this latter hascontributions of both the xanthene and the phenyl planes,while the angle between these two parts is clearly distorted, asevidenced by a comparison between spectra (b) and (c) inFigure 2.In the computed Raman spectrum of the isolated R6G, the

frequencies of modes A, C, and G do not follow theexperiment. The comparison between the Raman spectra ofisolated R6G and the “ethylamine down” configuration revealsthat the unsatisfied predictions in these modes are caused bydifferent reasons.First, the prediction of mode A is off the experimental value

by 12 cm−1 due to the missing adsorbate−substrateinteraction. Indeed, it is only off by 3 cm−1 in the “ethylaminedown” configuration. Such an improvement in mode A thencorroborates the necessity and validity of the surface inclusion.By considering the isolated molecule model as a weak bindingstate, it is reasonable to postulate that both 1655 and 1664cm−1 may be observed in the experiment if the bindingstrength could be weakened due to an external disturbance,e.g., thermal fluctuation. This postulate is consistent with theexperimental data in the seminal paper by Nie and Emory.4

Figure 5 in this reference, and reproduced here in Figure 3,presented nine Raman spectra taken from the same nano-particle captured at different times. According to theassumption made in that paper, there was only one moleculeon each nanoparticle. These nine Raman spectra exhibited anabrupt change in Mode A from 1653−1655 cm −1 to 1661−1664 cm−1 together with a reduction in the mode intensity.This is consistent with a partial desorption, according to oursimulations.Second, the deviation in mode C could be attributed to a

failure of our methodology in evaluating its correct Ramanintensity. The normal mode evaluation in fact shows a mode at1580 cm−1 for the isolated molecule. This mode is more likelythe origin of mode C, but is unidentifiable in the spectrum dueto its limited intensity. Our mode visualization (see theSupporting Information) shows that the mode at 1570 cm−1 isassociated with a xanthene plane deformation, while the modeat 1580 cm−1 is a motion of the lower phenyl ring. This specific

motion has been assigned to mode C in previous studies.14,17

Here, we choose to adopt the same assignment: mode C is themode at 1580 cm−1. This means that we consider that themode intensities in an approximate region in between 1555and 1585 cm−1 are not qualitatively accurate. Indeed, theintense theoretical peak at 1570 cm−1 is not observedexperimentally. The same situation occurs in the “ethylaminedown” configuration. A tiny peak is given by the normal modeevaluation at 1576 cm−1 and an intense peak lies at 1561 cm−1.We suspect that the qualitatively correct intensities in thisregion may only be captured by a model including thechemical enhancement effect,27 i.e., the interaction with thesurface in the polarizability derivative calculation.Third, we suspect that mode G originates from the addition

of the two modes at 1173 and 1200 cm−1. Indeed, we observethat, among the nine modes we discuss here, these two areamong the least affected ones by the metal/moleculeinteraction. However, mode G is also the mode that is themost broadened at the experimental level (26 cm−1 incomparison to 12 cm−1 for mode I for example), which wecannot explain by fluctuation in the metal/molecule inter-action. This highly suggests that mode G is a superposition of

Figure 3. Time-resolved surface-enhanced Raman spectra from asingle molecule characterization of R6G, recorded at 1 s intervals(Reproduced from Figure 5 in ref 4). These SERS signals aremeasured at room temperature from a single adsorbed molecule on aAg colloidal substrate in water. For more experimental details, refer toref 4.

The Journal of Physical Chemistry C Article

DOI: 10.1021/acs.jpcc.7b09441J. Phys. Chem. C 2018, 122, 465−473

469

two modes having similar wavenumbers and Raman intensities,such as the modes at 1173 and 1200 cm−1.Among the nine modes, we observe that the inclusion of the

metal surface only improves the experimental agreement formode A, in comparison with a simpler isolated moleculecalculation. Strikingly, modes C, E, and F are all in betteragreement for the isolated molecule calculation. A carefulexamination of the “ethylamine down” configuration indicatesthat the xanthene plane and its angle with respect to the phenylring are significantly distorted. The distortion is due to the vander Waals attraction between this plane and the silversurface,21 together with steric repulsion between the surfaceand ethylamine and phenyl groups; see Figures 4 and 5c. We

consider that this distortion could be at the source of thisinconsistency and, likely, it could prevent us from getting goodagreement with the experimental data, in comparison with theisolated configuration. Such a consideration is consistent withthe computed Raman spectrum of a much more severelydistorted binding state which was labeled as “twisted flip” inour previous work21 and results from a flip of the “ethylaminedown” configuration. As is shown in Figure 2d, this

configuration leads to worse experimental agreement, incomparison with the “ethylamine down” configuration. Itcompletely misses the modes B and F, broadens mode E, anddoes not give better predictions for the other modes. The badperformance of the “twisted flip” configuration confirms thehypothesis that reducing the internal distortion in the moleculeis an important feature to determine the adsorption state. Italso demonstrates that twisted flip configurations are to beexcluded, in agreement with its lower adsorption energy (2.00eV for the twisted flip configuration vs 2.18 eV for the“ethylamine down” configuration, as computed at the DFT/PBE+TS level of theory in ref 21).Taking the distortion into account, it is worthwhile to realize

that the ethylamine groups connect to the xanthene plane by aσ bond in sp3 geometry. Each ethylamine group is thus able torotate around this bond without experiencing a significantenergy barrier.49,50 This observation suggests a new possibleconfiguration of adsorbed R6G in which the ethylamine groupscould point in the opposite direction: “ethylamine up” (seeFigure 5d). The distortion in the xanthene plane is expected tobe significantly reduced for this configuration, and thismotivates us to study the adsorption properties of this newconfiguration, following our trial and error procedure.As expected, the computed adsorption energy of the

“ethylamine up” configuration is larger than the “ethylaminedown” configuration by 0.33 eV; see Figure 5. We used thesame procedure as given in our previous work to compute theadsorption energy.21 We note that twisting the ethylaminegroups up even lowered the energy of the molecule in gasphase by 0.002 eV. By decomposing the adsorption energyusing the method adopted in our previous study,21 it isrevealed that the “ethylamine up” configuration reduces theinternal distortion by 0.24 eV and increases the interactionwith the surface by 0.09 eV, in comparison with the“ethylamine down” configuration. The distortion in thexanthene plane is highly reduced as can be observed in Figures4 and 5. Therefore, this new configuration is a more probablebinding geometry.Looking at the computed Raman spectrum, we observe that

the “ethylamine up” configuration matches with “ethylaminedown” in all the highlighted modes, except for some minorshifts (<2 cm−1) that are seen for modes D, E, F, and I, withrespect to the experiments. The prediction at mode I isimproved from 619 to 614 cm−1 which is a perfect agreementwith the experimental result in Figure 2a. This change from619 cm−1 to a lower wavenumber shows the right trend incomparison with the “ethylamine up” configuration, sinceMode I is observed at mode frequencies equal to or less than615 cm−1 in most experiments.10,12,14−16 We show in theSupporting Information (Figure S1) that the twist of theethylamine group does not lead to any shift in normal modeenergies of the isolated molecule. This leads to the conclusionthat both the twisted ethylamine groups and the considerationof the metal/molecule interaction are responsible for the moreaccurate prediction of modes D, E, F, and I. Importantly, welose a bit of accuracy for mode A by 3 cm−1, in comparison tothe experimental spectrum. This is the only mode that is inpoorer agreement with experiment in this structure, and wenote that the metal-induced renormalization of this mode isstill present for this “ethylamine up” configuration.Considering our characterization of the simulated Raman

spectrum of the “ethylamine up” configuration, together withits higher adsorption energy, our calculations suggest that the

Figure 4. A side view of R6G is presented for isolated, ethylaminedown and ethylamine up configurations. The distortion can be seen inthe xanthene plane, as highlighted for the ethylamine downconfiguration. It can also be seen that the nitrogen atom lies out ofthe xanthene plane for the ethylamine down configuration, due to thesteric repulsion with the metal surface. We note that the last methylgroup of the ethylamine has been removed for clarity, in theethylamine down representation.

Figure 5. Optimized geometries are shown here for (a) isolated R6G,(b) isolated R6G with ethylamine groups twisted up, (c) ethylaminedown adsorbed R6G on Ag slab where isolated R6G was used as thestarting geometry, (d) ethylamine up adsorbed R6G on Ag slab whereisolated R6G with ethylamine groups twisted up was used as thestarting geometry. The numbers in green below (a) and (b) areenergies of isolated R6G calculated using the PBE level DFTformalism. The numbers in red below (c) and (d) are adsorptionenergies. The numbers in blue and black that come after the rednumbers are molecule−substrate interaction energy and molecularrelaxation energy, respectively.

The Journal of Physical Chemistry C Article

DOI: 10.1021/acs.jpcc.7b09441J. Phys. Chem. C 2018, 122, 465−473

470

R6G molecules are mainly adsorbed on the silver surface in aparallel configuration with the tail of the phenyl ring pointingout from the surface. In this case, there is very limiteddistortion, which is a configuration that cannot be achieved for“ethylamine down”. This directly suggests that the molecule isadsorbed with the ethylamine group pointing up from thesilver surface.To be complete, our discussion also shows that it is a

challenging problem to determine the adsorption geometryfrom a comparison between a simulated and measured Ramanspectra. Indeed, we observe that the correction given by theinclusion of the metal surface, despite generating a much largercomputational cost, remains limited. Only three peaks (A, D,and I), among the nine peaks we identified, are improved bymore than 2 cm−1, in comparison with the isolated moleculeand the experimental SERS signal. Moreover, the agreementfor modes C and E is actually better in the isolated moleculeconfiguration.This shows that significant inaccuracies are still present in

our study, and therefore, we consider that our assignedgeometry constitutes a reasonable suggestion, but with roomfor improvement. We believe that this limited success is due toseveral shortcomings of the computational methodology. First,the relative mode intensities that we compute are not inquantitative agreement, which could lead to a modemisidentification, as we suggest being the case for mode C.Second, the silver surface is kept frozen during our normalmodes calculation, which cannot account for modes thatdelocalize on both the metal and the adsorbent. Third,semilocal exchange-correlation functionals possess notoriouserrors, which could be corrected using more sophisticatedhybrid functionals but at greater cost.51 Fourth, a couplingbetween the modes, which goes beyond the harmonicapproximation, might further modify a few peak positions.52,53

However, at the present stage of ab initio computationaltechniques, the correction of each of these shortcomingsassociates with a computational cost that is very likely to beprohibitive. This issue confers a high interest in parametrizedor empirical quantum chemistry methods.54

To finish this discussion, we stress that, even though thereare expected quantitative shortcomings, the fact that thesurface−molecule interaction leads to a significant distortion ofthe molecule, and that it should be observable in the Ramanspectrum, constitutes a qualitative trend, which we consider iswell captured by our level of DFT calculations.

■ CONCLUSIONTo summarize, we propose in this paper that the R6Gmolecule-substrate system probed in the SERS experiment canbe characterized by combining a periodic slab model withexisting methodology for calculating SERS spectra. We findthat inclusion of the slab does not drastically improve thepredicted Raman spectra compared to the previous isolatedmolecular models. We interpret this as a direct demonstrationthat the R6G molecules probed in the SERS experimentpossess very limited distortion from the gas phase structure.This observation can be further used as a guideline towardidentification of the adsorption geometry.Following this reasoning, we conclude that the R6G

molecule adsorbs on the silver surface with the xantheneplane lying parallel and both the tail of the phenyl group andthe ethylamine groups pointing up from the surface. This is onthe basis of a better agreement between simulated and

experimental Raman spectra, but is also evidenced by a highercomputed adsorption energy. We believe that this reasoningcan be applied to other systems of interest, which paves theway to directly probe the adsorption geometry of the reactantin SM-SERS experiments. We note, however, that it constitutesa complex challenge, which was here greatly facilitated by thefact that we based our reasoning on a low temperature andhigh-vacuum experimental SERS acquisition. We believe thatanother factor that could facilitate this comparison would be totether the molecule to the metal substrate, with a chemicalgroup.16

While some mode energies can be improved afterconsideration of the substrate, our approach possesses severalshortcomings, such as the lack of the interaction with thesubstrate when calculating the polarizability derivatives.Therefore, the relative intensities of a few peaks may bequestionable, after comparison with the experimental spec-trum. A possibility to overcome this weakness might be givenby the replacement of our second computational step byanother computational scheme that would include theinteraction with the plasmonic surface.27,55

■ ASSOCIATED CONTENT*S Supporting InformationThe Supporting Information is available free of charge on theACS Publications website at DOI: 10.1021/acs.jpcc.7b09441.

Computational details; computed Raman spectra ofisolated R6G in “ethylamine down” and “ethylamine up”configurations; modes assignments for characteristicmodes of R6G; data set employed for determiningrescaling factor (PDF)

■ AUTHOR INFORMATIONCorresponding Authors*E-mail: [email protected] (C.V.D.).*E-mail: [email protected] (M.A.R.).ORCIDColin Van Dyck: 0000-0003-2853-3821Richard P. Van Duyne: 0000-0001-8861-2228George C. Schatz: 0000-0001-5837-4740Author Contributions∥C.V.D. and B.F. contributed equally to this work. All authorshave given approval to the final version of the manuscript.NotesThe authors declare no competing financial interest.

■ ACKNOWLEDGMENTSWe thank Dr. Fredy Aquino for the help with the AOResponsemodule of the NWChem package. This work was supported bythe Air Force Office of Scientific Research MURI project(FA9550-14-1-0003). C.V.D. thanks Dr. Lindsey Madison foruseful discussion and the support by the National Institute forNanotechnology, which is operated as a partnership betweenthe National Research Council, Canada, the University ofAlberta, and the Government of Alberta. We gratefullyacknowledge the computational resources from the Questhigh performance computing facility at Northwestern Uni-versity and the Extreme Science and Engineering DiscoveryEnvironment (XSEDE) program which is supported byNational Science Foundation grant number ACI-1053575.We also acknowledge the Center for Nanoscale Materials

The Journal of Physical Chemistry C Article

DOI: 10.1021/acs.jpcc.7b09441J. Phys. Chem. C 2018, 122, 465−473

471

(Argonne National Lab), an Office of Science user facility,supported by the U.S. Department of Energy, Office ofScience, Office of Basic Energy Sciences, under Contract No.DE-AC02-06CH11357.

■ REFERENCES(1) Jeanmaire, D. L.; Van Duyne, R. P. Surface RamanSpectroelectrochemistry. J. Electroanal. Chem. Interfacial Electrochem.1977, 84, 1−20.(2) Fleischmann, M.; Hendra, P. J.; McQuillan, A. J. Raman Spectraof Pyridine Adsorbed at a Silver Electrode. Chem. Phys. Lett. 1974, 26,163−166.(3) Moskovits, M. Surface-Enhanced Spectroscopy. Rev. Mod. Phys.1985, 57, 783−826.(4) Nie, S.; Emory, S. R. Probing Single Molecules and SingleNanoparticles by Surface-Enhanced Raman Scattering. Science 1997,275, 1102.(5) Kneipp, K.; Wang, Y.; Kneipp, H.; Perelman, L. T.; Itzkan, I.;Dasari, R. R.; Feld, M. S. Single Molecule Detection Using Surface-Enhanced Raman Scattering (Sers). Phys. Rev. Lett. 1997, 78, 1667−1670.(6) Hildebrandt, P.; Stockburger, M. Surface-Enhanced ResonanceRaman Spectroscopy of Rhodamine 6g Adsorbed on Colloidal Silver.J. Phys. Chem. 1984, 88, 5935−5944.(7) Michaels, A. M.; Nirmal, M.; Brus, L. E. Surface EnhancedRaman Spectroscopy of Individual Rhodamine 6g Molecules on LargeAg Nanocrystals. J. Am. Chem. Soc. 1999, 121, 9932−9939.(8) Michaels, A. M.; Jiang; Brus, L. Ag Nanocrystal Junctions as theSite for Surface-Enhanced Raman Scattering of Single Rhodamine 6gMolecules. J. Phys. Chem. B 2000, 104, 11965−11971.(9) Hayazawa, N.; Inouye, Y.; Sekkat, Z.; Kawata, S. Near-FieldRaman Scattering Enhanced by a Metallized Tip. Chem. Phys. Lett.2001, 335, 369−374.(10) Anger, P.; Feltz, A.; Berghaus, T.; Meixner, A. J. Near-Field andConfocal Surface-Enhanced Resonance Raman Spectroscopy atCryogenic Temperatures. J. Microsc. 2003, 209, 162−166.(11) Dieringer, J. A.; Lettan, R. B.; Scheidt, K. A.; Van Duyne, R. P.A Frequency Domain Existence Proof of Single-Molecule Surface-Enhanced Raman Spectroscopy. J. Am. Chem. Soc. 2007, 129, 16249−16256.(12) Shim, S.; Stuart, C. M.; Mathies, R. A. Resonance RamanCross-Sections and Vibronic Analysis of Rhodamine 6g fromBroadband Stimulated Raman Spectroscopy. ChemPhysChem 2008,9, 697−699.(13) Dieringer, J. A.; Wustholz, K. L.; Masiello, D. J.; Camden, J. P.;Kleinman, S. L.; Schatz, G. C.; Van Duyne, R. P. Surface-EnhancedRaman Excitation Spectroscopy of a Single Rhodamine 6g Molecule.J. Am. Chem. Soc. 2009, 131, 849−854.(14) Klingsporn, J. M.; Jiang, N.; Pozzi, E. A.; Sonntag, M. D.;Chulhai, D.; Seideman, T.; Jensen, L.; Hersam, M. C.; Duyne, R. P. V.Intramolecular Insight Into Adsorbate−Substrate Interactions ViaLow-Temperature, Ultrahigh-Vacuum Tip-Enhanced Raman Spec-troscopy. J. Am. Chem. Soc. 2014, 136, 3881−3887.(15) Zaleski, S.; Cardinal, M. F.; Klingsporn, J. M.; Van Duyne, R. P.Observing Single, Heterogeneous, One-Electron Transfer Reactions. J.Phys. Chem. C 2015, 119, 28226−28234.(16) Zaleski, S.; Cardinal, M. F.; Chulhai, D. V.; Wilson, A. J.;Willets, K. A.; Jensen, L.; Van Duyne, R. P. Toward MonitoringElectrochemical Reactions with Dual-Wavelength Sers: Character-ization of Rhodamine 6g (R6g) Neutral Radical Species and CovalentTethering of R6g to Silver Nanoparticles. J. Phys. Chem. C 2016, 120,24982−24991.(17) Watanabe, H.; Hayazawa, N.; Inouye, Y.; Kawata, S. DftVibrational Calculations of Rhodamine 6g Adsorbed on Silver:Analysis of Tip-Enhanced Raman Spectroscopy. J. Phys. Chem. B2005, 109, 5012−5020.

(18) Jensen, L.; Schatz, G. C. Resonance Raman Scattering ofRhodamine 6g as Calculated Using Time-Dependent DensityFunctional Theory. J. Phys. Chem. A 2006, 110, 5973−5977.(19) Zhao, J.; Jensen, L.; Sung, J.; Zou, S.; Schatz, G. C.; Van Duyne,R. P. Interaction of Plasmon and Molecular Resonances forRhodamine 6g Adsorbed on Silver Nanoparticles. J. Am. Chem. Soc.2007, 129, 7647−7656.(20) Guthmuller, J.; Champagne, B. Resonance Raman Scattering ofRhodamine 6g as Calculated by Time-Dependent Density FunctionalTheory: Vibronic and Solvent Effects. J. Phys. Chem. A 2008, 112,3215−3223.(21) Fu, B.; Van Dyck, C.; Zaleski, S.; Van Duyne, R. P.; Ratner, M.A. Single Molecule Electrochemistry: Impact of Surface SiteHeterogeneity. J. Phys. Chem. C 2016, 120, 27241−27249.(22) Jensen, L.; Autschbach, J.; Schatz, G. C. Finite Lifetime Effectson the Polarizability within Time-Dependent Density-FunctionalTheory. J. Chem. Phys. 2005, 122, 224115.(23) Aquino, F. W.; Schatz, G. C. Time-Dependent DensityFunctional Methods for Raman Spectra in Open-Shell Systems. J.Phys. Chem. A 2014, 118, 517−525.(24) Neugebauer, J.; Reiher, M.; Kind, C.; Hess, B. A. QuantumChemical Calculation of Vibrational Spectra of Large MoleculesRaman and Ir Spectra for Buckminsterfullerene. J. Comput. Chem.2002, 23, 895−910.(25) Sonntag, M. D.; Chulhai, D.; Seideman, T.; Jensen, L.; VanDuyne, R. P. The Origin of Relative Intensity Fluctuations in Single-Molecule Tip-Enhanced Raman Spectroscopy. J. Am. Chem. Soc.2013, 135, 17187−17192.(26) Pozzi, E. A.; Sonntag, M. D.; Jiang, N.; Chiang, N.; Seideman,T.; Hersam, M. C.; Van Duyne, R. P. Ultrahigh Vacuum Tip-Enhanced Raman Spectroscopy with Picosecond Excitation. J. Phys.Chem. Lett. 2014, 5, 2657−2661.(27) Zhao; Jensen, L.; Schatz, G. C. Pyridine−Ag20 Cluster: AModel System for Studying Surface-Enhanced Raman Scattering. J.Am. Chem. Soc. 2006, 128, 2911−2919.(28) Zhao, L. L.; Jensen, L.; Schatz, G. C. Surface-Enhanced RamanScattering of Pyrazine at the Junction Between Two Ag20Nanoclusters. Nano Lett. 2006, 6, 1229−1234.(29) Morton, S. M.; Jensen, L. Understanding the Molecule−SurfaceChemical Coupling in Sers. J. Am. Chem. Soc. 2009, 131, 4090−4098.(30) Moore, J. E.; Morton, S. M.; Jensen, L. Importance of CorrectlyDescribing Charge-Transfer Excitations for Understanding theChemical Effect in Sers. J. Phys. Chem. Lett. 2012, 3, 2470−2475.(31) Mullin, J. M.; Autschbach, J.; Schatz, G. C. Time-DependentDensity Functional Methods for Surface Enhanced Raman Scattering(Sers) Studies. Comput. Theor. Chem. 2012, 987, 32−41.(32) Plieth, W. J. Electrochemical Properties of Small Clusters ofMetal Atoms and Their Role in the Surface Enhanced RamanScattering. J. Phys. Chem. 1982, 86, 3166−3170.(33) Ivanova, O. S.; Zamborini, F. P. Size-Dependent Electro-chemical Oxidation of Silver Nanoparticles. J. Am. Chem. Soc. 2010,132, 70−72.(34) Hansen, K. H.; Worren, T.; Stempel, S.; Lægsgaard, E.; Baumer,M.; Freund, H. J.; Besenbacher, F.; Stensgaard, I. PalladiumNanocrystals on Al2O3: Structure and Adhesion Energy. Phys. Rev.Lett. 1999, 83, 4120−4123.(35) Hansen, T. W.; Wagner, J. B.; Hansen, P. L.; Dahl, S.; Topsøe,H.; Jacobsen, C. J. H. Atomic-Resolution in Situ TransmissionElectron Microscopy of a Promoter of a Heterogeneous Catalyst.Science 2001, 294, 1508−1510.(36) Hansen, P. L.; Wagner, J. B.; Helveg, S.; Rostrup-Nielsen, J. R.;Clausen, B. S.; Topsøe, H. Atom-Resolved Imaging of Dynamic ShapeChanges in Supported Copper Nanocrystals. Science 2002, 295,2053−2055.(37) Marks, L. D. Experimental Studies of Small Particle Structures.Rep. Prog. Phys. 1994, 57, 603.(38) Sun, Y.; Xia, Y. Shape-Controlled Synthesis of Gold and SilverNanoparticles. Science 2002, 298, 2176.

The Journal of Physical Chemistry C Article

DOI: 10.1021/acs.jpcc.7b09441J. Phys. Chem. C 2018, 122, 465−473

472

(39) Barmparis, G. D.; Lodziana, Z.; Lopez, N.; Remediakis, I. N.Nanoparticle Shapes by Using Wulff Constructions and First-Principles Calculations. Beilstein J. Nanotechnol. 2015, 6, 361−368.(40) Kresse, G.; Furthmuller, J. Efficiency of Ab-Initio Total EnergyCalculations for Metals and Semiconductors Using a Plane-WaveBasis Set. Comput. Mater. Sci. 1996, 6, 15−50.(41) Kresse, G.; Furthmuller, J. Efficient Iterative Schemes for AbInitio Total-Energy Calculations Using a Plane-Wave Basis Set. Phys.Rev. B: Condens. Matter Mater. Phys. 1996, 54, 11169−11186.(42) Perdew, J. P.; Burke, K.; Ernzerhof, M. Generalized GradientApproximation Made Simple. Phys. Rev. Lett. 1996, 77, 3865−3868.(43) Tkatchenko, A.; Scheffler, M. Accurate Molecular Van DerWaals Interactions from Ground-State Electron Density and Free-Atom Reference Data. Phys. Rev. Lett. 2009, 102, 073005.(44) Scott, A. P.; Radom, L. Harmonic Vibrational Frequencies: AnEvaluation of Hartree−Fock, Møller−Plesset, Quadratic Configu-ration Interaction, Density Functional Theory, and SemiempiricalScale Factors. J. Phys. Chem. 1996, 100, 16502−16513.(45) Andersson, M. P.; Uvdal, P. New Scale Factors for HarmonicVibrational Frequencies Using the B3lyp Density Functional Methodwith the Triple-Ζ Basis Set 6-311+G(D,P). J. Phys. Chem. A 2005,109, 2937−2941.(46) Alecu, I. M.; Zheng, J.; Zhao, Y.; Truhlar, D. G. ComputationalThermochemistry: Scale Factor Databases and Scale Factors forVibrational Frequencies Obtained from Electronic Model Chem-istries. J. Chem. Theory Comput. 2010, 6, 2872−2887.(47) Valiev, M.; Bylaska, E. J.; Govind, N.; Kowalski, K.; Straatsma,T. P.; Van Dam, H. J. J.; Wang, D.; Nieplocha, J.; Apra, E.; Windus, T.L.; et al. Nwchem: A Comprehensive and Scalable Open-SourceSolution for Large Scale Molecular Simulations. Comput. Phys.Commun. 2010, 181, 1477−1489.(48) Jensen, L.; Autschbach, J.; Schatz, G. C. Finite Lifetime Effectson the Polarizability within Time-Dependent Density-FunctionalTheory. J. Chem. Phys. 2005, 122, 224115.(49) Pople, J. A.; Gordon, M. Molecular Orbital Theory of theElectronic Structure of Organic Compounds. I. Substituent Effectsand Dipole Moments. J. Am. Chem. Soc. 1967, 89, 4253−4261.(50) Durig, J. R.; Li, Y. S. Raman Spectra of Gases. XVIII. InternalRotational Motions in Ethylamine and Ethylamine-D2. J. Chem. Phys.1975, 63, 4110−4113.(51) Silverstein, D. W.; Jensen, L. Understanding the ResonanceRaman Scattering of Donor−Acceptor Complexes Using Long-RangeCorrected DFT. J. Chem. Theory Comput. 2010, 6, 2845−2855.(52) Ratner, M. A.; Gerber, R. B. Excited Vibrational States ofPolyatomic Molecules: The Semiclassical Self-Consistent FieldApproach. J. Phys. Chem. 1986, 90, 20−30.(53) Nagalakshmi, V.; Lakshminarayana, V.; Sumithra, G.; DurgaPrasad, M. Durga Prasad, M. Coupled Cluster Description ofAnharmonic Molecular Vibrations. Application to O3 and So2.Chem. Phys. Lett. 1994, 217, 279−282.(54) Gieseking, R. L.; Ratner, M. A.; Schatz, G. C. SemiempiricalModeling of Electrochemical Charge Transfer. Faraday Discuss. 2017,199, 547−563.(55) Gieseking, R. L.; Ratner, M. A.; Schatz, G. C. SemiempiricalModeling of Ag Nanoclusters: New Parameters for Optical PropertyStudies Enable Determination of Double Excitation Contributions toPlasmonic Excitation. J. Phys. Chem. A 2016, 120, 4542−4549.

The Journal of Physical Chemistry C Article

DOI: 10.1021/acs.jpcc.7b09441J. Phys. Chem. C 2018, 122, 465−473

473