Open Access Article
Zachary K.
Goldsmith
*,
Marcos F.
Calegari Andrade
and
Annabella
Selloni
*
Department of Chemistry, Princeton University, Princeton, NJ 08544, USA. E-mail: zkg@princeton.edu; aselloni@princeton.edu
First published on 18th March 2021
Electrode–water interfaces under voltage bias demonstrate anomalous electrostatic and structural properties that are influential in their catalytic and technological applications. Mean-field and empirical models of the electrical double layer (EDL) that forms in response to an applied potential do not capture the heterogeneity that polarizable, liquid-phase water molecules engender. To illustrate the inhomogeneous nature of the electrochemical interface, Born–Oppenheimer ab initio molecular dynamics calculations of electrified Au(111) slabs interfaced with liquid water were performed using a combined explicit–implicit solvent approach. The excess charges localized on the model electrode were held constant and the electrode potentials were computed at frequent simulation times. The electrode potential in each trajectory fluctuated with changes in the atomic structure, and the trajectory-averaged potentials converged and yielded a physically reasonable differential capacitance for the system. The effects of the average applied voltages, both positive and negative, on the structural, hydrogen bonding, dynamical, and vibrational properties of water were characterized and compared to literature where applicable. Controlled-potential simulations of the interfacial solvent dynamics provide a framework for further investigation of more complex or reactive species in the EDL and broadly for understanding electrochemical interfaces in situ.
Experimental studies of the structure of the electrode–water interface, often by X-ray absorption spectroscopy (XAS), have underscored its anomalous structure and hydrogen bonding patterns.9,10 Specifically, water forms at least one structural layer against common electrodes such as Ag and Au with densities higher than bulk and with dangling OH bonds at both positive and negative voltage biases.11 Directly studying electrochemical interfaces in situ has been largely accomplished via surface-specific vibrational spectroscopies such as surface-enhanced infrared absorption (SEIRAS),12,13 plasmon-enhanced Raman spectroscopy,14 and sum-frequency generation (SFG).15,16 These techniques have elucidated dependencies of the vibrational spectra of water at the interface on the voltage applied, consequences of the strong electric fields in the EDL.17 Nonetheless, experimental probes of the structure and composition of the EDL are challenging to perform, incomplete in scope, and would benefit from improvements in atomistic modeling.
Realistic simulation of the EDL and processes therein is an outstanding challenge for theory and computation as well. Classical MD and Monte Carlo simulations of water at electrified surfaces18–20 have been instructive but are limited in their microscopic resolution. Notably, empirical approaches lack instantaneous polarization of species in the interfacial electric field that may give way to novel properties and/or reactivity. Quantum chemical simulations, most practically density functional theory (DFT), are capable of describing molecular polarization and bond-breaking, albeit at greater computational cost. DFT with a posteriori corrections for the thermodynamic effects of the electrode potential21,22 have been useful in elaborating catalytic mechanisms, but neglect specific effects of the strong, induced electric fields at the interface.
Recent advances in periodic DFT with continuum solvation,23–26 self-consistent analytical EDL models,27,28 as well as formally grand-canonical approaches29,30 have enabled studies of more sophisticated systems that include a controlled potential. Applications of such implicit and/or hybrid solvation methods have included potentiostatic simulations along key nuclear coordinates associated with heterogeneous catalysis31–34 and surface-enhanced spectroscopy.35,36 Dynamical ab initio studies including voltage bias have mostly considered just the electrode37 or limited numbers of molecules38–40 and/or monolayers.14,41–44
In this work, we performed ab initio molecular dynamics of water at a Au electrode interface with fixed amounts of charge on the Au electrode as a means of modulating the electrode potential. The electrode potentials were measured at frequent times for each simulation and their converged average values yielded a physically reasonable differential capacitance for the system. The recovery of the system's electronic properties enabled analyses of the structure and dynamics of interfacial water under quantifiable potentials. In particular, we elucidated the effect of voltage on the density of water with respect to surface normal as well as the characteristic orientations of water molecules near the electrode as a function of the voltage bias. Furthermore, our computed vibrational densities of states of the interfacial water under bias corroborated phenomena previously reported by surface-enhanced spectroscopies. Moreover, we realized ab initio-quality simulations of an electrode under quantifiable voltage biases in contact with several layers of explicit solvent, a long outstanding benchmark for computation. We anticipate that atomistic modeling of electrochemical interfaces will be a powerful tool for understanding how chemical mechanisms are shaped by the polarization and statistical dynamics of the EDL.
The inactive back side of the Au(111) slab was in contact with vacuum (ε = 1) and the surface atoms on this side were held fixed in the direction perpendicular to the surface (z). A region of dielectric continuum representative of bulk water (ε = 78.3) was imposed beyond the region of explicit water molecules. This hybrid explicit/implicit approach enabled us to use a smaller system in expensive electronic structure calculations than a fully explicit system. More critically, this hybrid approach made it possible to establish both a region of “bulk solution” and a vacuum on the inactive side of the Au slab. The presence of the vacuum ensured the selective localization of excess charge at the interface with water.33 The continuum solvent was included using the Environ24,46 plug-in to Quantum-ESPRESSO,47,48 which solves the Poisson equation for the dielectric continuum in contact with the atomistic system self-consistently with the Kohn–Sham equation. Specifically, we employed a “solvent-aware”49 implementation of the self-consistent continuum solvation method24 and solved the relevant Poisson equation using periodic boundary conditions only in the surface-parallel dimensions to avoid solvation interactions between unit cells. A plane of charge equal and opposite that on the electrode was positioned within the region of implicit water to maintain the overall neutrality of the periodic cell. In addition, a dipole correction was employed in the vacant region near the boundary of the cell50 to negate spurious electrostatic interactions via periodicity.
Additionally, a planar step function potential51 was positioned at a distance from the Au surface that ensured a total density of 1 g mL−1 of the 56 water molecules. The step function, which in practice was error function-shaped, had a height of 0.25 Ry. To maintain 3D periodicity, this region was terminated near the top of the unit cell and beyond the position of the counter-charge plane by a step function of equal size and opposite directionality. These features are illustrated in Fig. 1.
We carried out five BOMD trajectories of our model electrochemical half-cell, each with a different amount of charge localized at the Au surface. The total electrode charges per unit cell σ used were −0.4e, −0.2e, 0e, +0.2e, and +0.4e, where e is the elementary charge. These amounted to at most ±0.025e per surface Au atom, if distributed evenly. We demonstrate below that these σ represented modest applied potentials smaller in magnitude than ±0.5 V vs. PZC. The BOMD runs were performed with a timestep of ∼0.48 fs, which was made possible by using the mass of D for all H atoms, and at 330 K, as is typical for AIMD simulations of bulk liquid water.58 The results below regarding hydrogen bonding and diffusivity are indicative that water was sufficiently liquid in our simulations. The physical Au and O masses were used otherwise. We employed an NVT ensemble with a velocity-rescaling thermostat with a tolerance of ±100 K. Each trajectory spanned over 10 ps of integration time, which we will later show was sufficient to recover a physical differential capacitance for the system. The initial coordinates were generated via a geometry minimization in a classical force field57 with no excess charges followed by partial DFT-level geometry optimization at each trajectory's respective electrode charge. The 3D electrostatic potential Φ, as obtained by solving the Kohn–Sham equation self-consistently, was recorded every 100 time steps (∼48 fs) of each trajectory.
![]() | (1) |
. Then, we took as ΦS the value of
deep in the continuum solvent region of the cell,25,36,61 where
is unchanging as a function of z (Fig. 1B). To measure the metal potential, we macroscopically-averaged
in the Au slab using the optimized layer spacing of 2.4 Å and took as μM the average of this quantity over the back 1 Å of the slab. Utilizing the partially-frozen back layer provided a more stable metal potential (Fig. 1B), although alternative definitions for ΦM did not qualitatively change our results (see Fig. S3†). The resulting values of Φe presented below represent the difference between the quantities ΦM and ΦS for each trajectory.
For the purposes of analyzing the effects of electrode potential on various properties of water we referred potentials to the system's PZC, i.e. σ = 0 e. The referenced potentials ΔΦe are expressed as,
| ΔΦe = Φabs[σ] − Φabs[σ = 0] = Φe[σ] − Φe[σ = 0] | (2) |
In this expression it is evident that we do not need the bulk metal's Fermi energy to compute the referenced electrode potentials. While certain implementations of e.g. Poisson–Boltzmann boundary conditions may make it possible to obtain the absolute potentials from ab initio calculations of small systems,27,41,62 the PZC-referenced potentials will suffice to quantify the effects of electrode polarization on water. In the present application, the referenced electrode potentials associated with each σ will emerge as the trajectory averages of the Φe, referred to hereafter as φ and computed as,
| φ = Δ〈Φe〉 | (3) |
In the following section we will first present the dynamics and convergence of each trajectory's Φe before subsequently analyzing the structure, dynamics, and vibrational spectra of interfacial water as a function of φ, the trajectory-averaged potentials versus PZC.
for each electrode charge as well as the full trajectory averages are shown in Fig. 2. The electrostatic potential wells observed include one for each of the layers of the Au slab and at least one characteristic of a structural layer of water at the interface. In the region of dielectric continuum, to the right of the potential energy step in Fig. 2,
is unchanging in z, making the calculation of ΦS straightforward. The metal potentials as measured at the back of the slab were relatively stable as a function of time, as shown in Fig. S2 and S3.† In the two trajectories with negative electrode charges, and therefore positive counter charges, we observed electrostatic potential wells near the boundary of the region of explicit water. While these are indicative of a spurious interaction between some water molecules and the counter charge plane, the below analyses demonstrate that the system's electrical properties and more immediate interfacial water layers were nonetheless physically realistic.
At each electrode charge,
fluctuated as a consequence of the changing atomic structure during the molecular dynamics, as depicted in Fig. 2. Moreover, for these fixed σ trajectories, motion of both the Au and water atoms yielded changes to the electrode potentials Φe over time, plotted in Fig. 3A. The values of Φe for the five trajectories largely preserve the intuitive ordering according to the electrode charges throughout the dynamics. After at most 5–6 ps of dynamics, the cumulative time averages 〈Φe〉 (shown in Fig. 3B) were well converged. Although the 10 ps trajectories were short relative to molecular dynamics studies of interfacial water,63 the converged electrical properties of the system allowed for a qualitative investigation into water's structure and dynamics under quantifiable voltage bias.
The relationship between the surface charges σ and the full trajectory-averaged electrode potentials versus PZC φ was found to be nearly perfectly linear, as depicted in Fig. 4. We ascribe the observed symmetry of the potentials around PZC to the lack of electrolyte ions in our system. The inverse slope of the linear fit to these five data points represents the differential capacitance Cd of the Au(111)/water system, 14.3 ± 2.2 μF cm−2. This Cd is within the rather large range of experimentally measured values, 11–40 μF cm−2,64–66 which depend significantly on experimental conditions including but not limited to the concentration and identity of the electrolyte ions. In the absence of ions, we are satisfied to recover a reasonable differential capacitance for Au(111)/water towards the bottom of the measured range of Cd.67 Similar values of Cd have been achieved for Au(111)/water with a fully implicit solvent with specifically tuned parameters25 and with a monolayer of water.33 Our results recover this macroscopic electrical property from short AIMD trajectories and an effectively fully explicit solvent.
The structural similarities of the first water layers at PZC and at φ > 0 V have been previously observed by XAS.11 Furthermore, the position and maximum of the first density peak at PZC is in agreement with recent AIMD studies of Au(111)/water.68,69 An earlier AIMD study of this system at PZC predicted a narrowly resolved first water layer with a density maximum of ∼4 times that of bulk water,11 although this work was performed with PBE water at 300 K, which is now known to likely be over-structured.70
For the purposes of further analyses, we designated four characteristic layers of water for each trajectory, the approximate positions of which are delineated in Fig. 5. The layer boundaries were chosen at local density minima and differed slightly for each φ. Hereafter we will refer to water in these spatial regions by the numbers shown in Fig. 5, with layer 1 being the most immediate interfacial layer, followed by layer 2. Layer 3 is yet further from the surface and appears in terms of the density to be most similar to bulk water. We will neglect layer 4 in the subsequent analyses given its aforementioned unphysical behavior.
θ) at non-zero φ in Fig. 6A demonstrate that when the Au surface was charged positively the bisectors were selectively oriented away from the surface (O atoms oriented towards the surface) and when the surface was negatively charged the water dipoles were selectively oriented towards the surface (O atoms oriented away from the surface). These results follow intuitively from electrostatic attraction of O to the positively charged surface and of H/D to the negatively charged surface. Nonetheless, these maxima appear not at cos
θ = ±1 but at intermediate positive and negative values, i.e. at these modest voltages water's geometric dipoles are not entirely oriented to the electric fields. We also observed non-negligible probabilities of water molecules being oriented counter to the prevailing electrostatics for both positive and negative φ. At PZC, we observed a bimodal distribution of water bisectors with more being oriented generally away from the surface by a small margin.71 This is in subtle contrast to literature characterizations of room temperature water at the Au(111) interface as being similar to a 2D ice, in which the bisectors would be predominantly parallel to the surface and P(cos
θ) would be maximal near cos
θ = 0.11 In general, the bisector orientation statistics in layer 1 appear to smoothly change as a function of φ.
The histogram of the cosine of the angles formed between all water OD vectors and a unit vector parallel to the z axis in layer 1 is given in Fig. 6B. These plots demonstrate local maxima at −1 for the negative φ and at +1 for the positive φ, i.e. OD bonds pointed perpendicularly towards and away from the surface, respectively, in accordance with the bond's geometric dipole. At each non-zero φ we additionally observe significant frequencies near cos
θ = 0, indicative of a significant proportion of OD bonds being nearly parallel to the surface in layer 1.
The complimentary observations of water bisectors not being entirely oriented along the applied fields and the significant proportion of OD bonds being roughly parallel to the surface suggest a general characterization of water's structure in layer 1: water's geometric dipole being partially oriented according to the local field while sustaining hydrogen bonding interactions with neighboring water molecules. Nonetheless, interfacial water at room temperature still exhibits considerable disorder within this paradigm at the modest values of φ studied.
This characterization is further supported by an analysis of the hydrogen bonding properties of water in our simulations that found on average at least 1 and up to more than 2 intra-layer hydrogen bonds among layer 1 water at every potential (Fig. S5†). In addition, our calculations suggest fewer hydrogen bonds formed by water in layer 1 at negative φ than at PZC and positive φ, on average (Fig. S5†). This is consistent with the observation of more dangling OH bonds at negative potentials than at positive potentials by XAS.11 The hydrogen bonding properties of water at an Au(111) interface at PZC are in accordance with a previous BOMD study.71 The average number of hydrogen bonds in layer 3, the most bulk-like water, ranged from 3.42–3.58 (Fig. S7†). This compares favorably to the experimental value for bulk water at 330 K, 3.58.72 Comparable BOMD simulations of bulk water estimate this value to be approximately 3.85, 3.77, and 3.60 for PBE, PBE + TS, and SCAN, respectively.58
The biasing voltage affects the orientation of water in sub-interfacial layers 2 and 3 as well, as shown in Fig. S8 and S9.† We observed little attenuation of the bisector orientation preferences of water in layers 2 and 3 at non-zero φ. In the case of the PZC trajectory, water no longer shows any significant orientation pattern beyond layer 1.
For water in layer 1, we observed an increase in the D‖ at negative φ relative to both PZC and to bulk D2O. At PZC and at positive φ we observed little difference within error of D‖ relative to bulk D2O, further evidence contrary to layer 1 water having ice-like properties. For layer 2, we found a minimum D‖ at PZC, with D‖ at PZC being below, but within the error, of bulk D2O. Furthermore, D‖ at both positive and negative potentials in layer 2 was found to be higher than that of bulk D2O. In layer 3 we observed a similar response of D‖ to φ as in layer 2. Note that in layer 3, the most bulk-like water in our small simulation, D‖ at PZC was nearly identical to that of bulk D2O (Fig. 7).
The observation of increased parallel diffusivity of water in layer 1 at negative φ can be rationalized by the smaller average number of hydrogen bonds observed under these conditions (Fig. S5–S7†). Conversely, the total average numbers of hydrogen bonds were roughly the same for the trajectories of φ ≥ 0, for which the layer 1 parallel diffusion coefficients were similar. The increased D‖ in layers 2 and 3 at φ ≠ 0 are more difficult to rationalize on the basis of hydrogen bonding, given that their total numbers and ratios of intra- and inter-layer hydrogen bonding are not very different from PZC in these layers (Fig. S5–S7†). However, there is literature precedent for water's diffusion being faster in small, oscillatory external electric fields,73 which may describe the sub-interfacial water layer. Recent AIMD simulations demonstrated that such electric fields induced frequent rearrangements of the hydrogen bonding network and consequentially increased water's diffusivity.73 We also note that D‖ in each layer is generally lowest at PZC, the only trajectory in which the vdW correction was included, despite the inclusion of vdW corrections in AIMD being known to increase the diffusivity of bulk water.58
![]() | ||
| Fig. 8 Vibrational density of states of D2O in layers 1 (top), 2 (middle), and 3 (bottom) at every φ. Spectra for each φ are vertically offset by arbitrary amounts in each plot for visual clarity. | ||
In the VDOS spectra for layer 1 water (top panel of Fig. 8) there is a clear dependence on φ. Both positive and negative φ have the effect of blue shifting the peaks associated with the δDOD and νOD vibrations. This phenomenon has been observed experimentally by SEIRAS.12 Moreover, the blue shifts of the two characteristic signatures with both positive and negative φ are consistent with the orientation of water's geometric dipoles with the induced fields vis-à-vis the Stark effect.75 This qualitative conclusion holds despite the vdW correction (only employed at PZC) being known to blue shift both the δDOD and νOD peaks.76,77
Furthermore, the δDOD peaks of layer 1 water appear at lower frequencies relative to those of water in layers 2 and 3 (Fig. S19†). This is suggestive of a weaker hydrogen bonding network78 in layer 1, consistent with our observation of fewer overall hydrogen bonds formed in the interfacial layer (Fig. S5–S7†). In addition to the behavior of the peak positions, the integrated intensities of the δDOD peaks demonstrate a minimum at φ = 0 V (Fig. S21†), consistent with SEIRAS data.12 Further analyses of the VDOS spectra, including the spectra that emerge from the velocities only in z, are presented in the ESI.†
The VDOS spectra for layer 2 water (middle panel of Fig. 8) are significantly less sensitive to φ than are those in layer 1. Nonetheless, we observe in layer 2 small potential dependencies of the νOD peak (Fig. S17†). It is evident from the bottom panel of Fig. 8 and S18† that in layer 3, which is comprised of water even further from the electrode surface, the effects of φ on the VDOS spectra are mostly negligible. This observation is consistent with the electric field associated with the applied potential being nearly entirely screened by the first two layers of water.
That the BOMD trajectories reliably quantified the potential φ enabled structural and dynamic analyses of water with respect to both φ and distance from the electrode surface. Our calculations demonstrated that water forms at least one structural layer with a density greater than that of bulk against the Au(111) electrode, and more than one such layer at PZC and at positive φ. In addition, an analysis of the orientations of water characterize a general structure of water under bias in layer 1 in which one OD bond is generally oriented towards or away from the electrode depending on the sign of the applied potential whereas the other OD bond is nearly parallel to the surface. We observed that the D‖ of layer 1 water is enhanced only at potentials negative of the PZC, and that D‖ increases for all non-zero values of φ in the second and third layers of water. Lastly, the VDOS were most significantly potential-dependent in layer 1 of water, in which the δDOD and νOD peaks exhibited blue shifts for all trajectories with φ ≠ 0. The δDOD peak positions supported the observation of a weakened hydrogen bonding network at the interface.
This work represents a step towards describing in situ aqueous electrochemistry with ab initio molecular simulation. Water comprises a significant portion of the EDL even for concentrated aqueous electrolyte solutions, necessitating that we intimately understand how its properties are altered by applied voltage at equilibrium and beyond. Looking ahead, the inclusion of ions that constitute the EDL as well as reliable descriptions of driven, faradaic charge transfer processes would vastly enhance the scope of these simulations and offer valuable microscopic insights into electrochemistry and its diverse applications.
Footnote |
| † Electronic supplementary information (ESI) available. See DOI: 10.1039/d1sc00354b |
| This journal is © The Royal Society of Chemistry 2021 |