ViewVC Help
View File | Revision Log | Show Annotations | View Changeset | Root Listing
root/group/stokes/stokes.tex
Revision: 3772
Committed: Tue Dec 6 18:05:05 2011 UTC (12 years, 9 months ago) by skuang
Content type: application/x-tex
File size: 44652 byte(s)
Log Message:
add a plot and edit methodology

File Contents

# User Rev Content
1 gezelter 3769 \documentclass[11pt]{article}
2     \usepackage{amsmath}
3     \usepackage{amssymb}
4     \usepackage{setspace}
5     \usepackage{endfloat}
6     \usepackage{caption}
7     %\usepackage{tabularx}
8     \usepackage{graphicx}
9     \usepackage{multirow}
10     %\usepackage{booktabs}
11     %\usepackage{bibentry}
12     %\usepackage{mathrsfs}
13     %\usepackage[ref]{overcite}
14     \usepackage[square, comma, sort&compress]{natbib}
15     \usepackage{url}
16     \pagestyle{plain} \pagenumbering{arabic} \oddsidemargin 0.0cm
17     \evensidemargin 0.0cm \topmargin -21pt \headsep 10pt \textheight
18     9.0in \textwidth 6.5in \brokenpenalty=10000
19    
20     % double space list of tables and figures
21     \AtBeginDelayedFloats{\renewcommand{\baselinestretch}{1.66}}
22     \setlength{\abovecaptionskip}{20 pt}
23     \setlength{\belowcaptionskip}{30 pt}
24    
25     %\renewcommand\citemid{\ } % no comma in optional reference note
26     \bibpunct{[}{]}{,}{n}{}{;}
27     \bibliographystyle{achemso}
28    
29     \begin{document}
30    
31 skuang 3770 \title{ENTER TITLE HERE}
32 gezelter 3769
33     \author{Shenyu Kuang and J. Daniel
34     Gezelter\footnote{Corresponding author. \ Electronic mail: gezelter@nd.edu} \\
35     Department of Chemistry and Biochemistry,\\
36     University of Notre Dame\\
37     Notre Dame, Indiana 46556}
38    
39     \date{\today}
40    
41     \maketitle
42    
43     \begin{doublespace}
44    
45     \begin{abstract}
46 skuang 3770 REPLACE ABSTRACT HERE
47 gezelter 3769 With the Non-Isotropic Velocity Scaling (NIVS) approach to Reverse
48     Non-Equilibrium Molecular Dynamics (RNEMD) it is possible to impose
49     an unphysical thermal flux between different regions of
50     inhomogeneous systems such as solid / liquid interfaces. We have
51     applied NIVS to compute the interfacial thermal conductance at a
52     metal / organic solvent interface that has been chemically capped by
53     butanethiol molecules. Our calculations suggest that coupling
54     between the metal and liquid phases is enhanced by the capping
55     agents, leading to a greatly enhanced conductivity at the interface.
56     Specifically, the chemical bond between the metal and the capping
57     agent introduces a vibrational overlap that is not present without
58     the capping agent, and the overlap between the vibrational spectra
59     (metal to cap, cap to solvent) provides a mechanism for rapid
60     thermal transport across the interface. Our calculations also
61     suggest that this is a non-monotonic function of the fractional
62     coverage of the surface, as moderate coverages allow diffusive heat
63     transport of solvent molecules that have been in close contact with
64     the capping agent.
65    
66     \end{abstract}
67    
68     \newpage
69    
70     %\narrowtext
71    
72     %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
73     % BODY OF TEXT
74     %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
75    
76     \section{Introduction}
77 skuang 3771 [REFINE LATER, ADD MORE REF.S]
78     Imposed-flux methods in Molecular Dynamics (MD)
79     simulations\cite{MullerPlathe:1997xw} can establish steady state
80     systems with a set applied flux vs a corresponding gradient that can
81     be measured. These methods does not need many trajectories to provide
82     information of transport properties of a given system. Thus, they are
83     utilized in computing thermal and mechanical transfer of homogeneous
84     or bulk systems as well as heterogeneous systems such as liquid-solid
85     interfaces.\cite{kuang:AuThl}
86 skuang 3770
87 skuang 3771 The Reverse Non-Equilibrium MD (RNEMD) methods adopt constraints that
88     satisfy linear momentum and total energy conservation of a system when
89     imposing fluxes in a simulation. Thus they are compatible with various
90     ensembles, including the micro-canonical (NVE) ensemble, without the
91     need of an external thermostat. The original approaches by
92     M\"{u}ller-Plathe {\it et
93     al.}\cite{MullerPlathe:1997xw,ISI:000080382700030} utilize simple
94     momentum swapping for generating energy/momentum fluxes, which is also
95     compatible with particles of different identities. Although simple to
96     implement in a simulation, this approach can create nonthermal
97     velocity distributions, as discovered by Tenney and
98     Maginn\cite{Maginn:2010}. Furthermore, this approach to kinetic energy
99     transfer between particles of different identities is less efficient
100     when the mass difference between the particles becomes significant,
101     which also limits its application on heterogeneous interfacial
102     systems.
103 skuang 3770
104 skuang 3771 Recently, we developed a different approach, using Non-Isotropic
105     Velocity Scaling (NIVS) \cite{kuang:164101} algorithm to impose
106     fluxes. Compared to the momentum swapping move, it scales the velocity
107     vectors in two separate regions of a simulated system with respective
108     diagonal scaling matrices. These matrices are determined by solving a
109     set of equations including linear momentum and kinetic energy
110     conservation constraints and target flux satisfaction. This method is
111     able to effectively impose a wide range of kinetic energy fluxes
112     without obvious perturbation to the velocity distributions of the
113     simulated systems, regardless of the presence of heterogeneous
114     interfaces. We have successfully applied this approach in studying the
115     interfacial thermal conductance at metal-solvent
116     interfaces.\cite{kuang:AuThl}
117 gezelter 3769
118 skuang 3771 However, the NIVS approach limits its application in imposing momentum
119     fluxes. Temperature anisotropy can happen under high momentum fluxes,
120     due to the nature of the algorithm. Thus, combining thermal and
121     momentum flux is also difficult to implement with this
122     approach. However, such combination may provide a means to simulate
123     thermal/momentum gradient coupled processes such as freeze
124     desalination. Therefore, developing novel approaches to extend the
125     application of imposed-flux method is desired.
126 gezelter 3769
127 skuang 3771 In this paper, we improve the NIVS method and propose a novel approach
128     to impose fluxes. This approach separate the means of applying
129     momentum and thermal flux with operations in one time step and thus is
130     able to simutaneously impose thermal and momentum flux. Furthermore,
131     the approach retains desirable features of previous RNEMD approaches
132     and is simpler to implement compared to the NIVS method. In what
133     follows, we first present the method to implement the method in a
134     simulation. Then we compare the method on bulk fluids to previous
135     methods. Also, interfacial frictions are computed for a series of
136     interfaces.
137 gezelter 3769
138     \section{Methodology}
139 skuang 3770 Similar to the NIVS methodology,\cite{kuang:164101} we consider a
140     periodic system divided into a series of slabs along a certain axis
141     (e.g. $z$). The unphysical thermal and/or momentum flux is designated
142     from the center slab to one of the end slabs, and thus the center slab
143     would have a lower temperature than the end slab (unless the thermal
144     flux is negative). Therefore, the center slab is denoted as ``$c$''
145     while the end slab as ``$h$''.
146    
147     To impose these fluxes, we periodically apply separate operations to
148     velocities of particles {$i$} within the center slab and of particles
149     {$j$} within the end slab:
150     \begin{eqnarray}
151     \vec{v}_i & \leftarrow & c\cdot\left(\vec{v}_i - \langle\vec{v}_c
152     \rangle\right) + \left(\langle\vec{v}_c\rangle + \vec{a}_c\right) \\
153     \vec{v}_j & \leftarrow & h\cdot\left(\vec{v}_j - \langle\vec{v}_h
154     \rangle\right) + \left(\langle\vec{v}_h\rangle + \vec{a}_h\right)
155     \end{eqnarray}
156     where $\langle\vec{v}_c\rangle$ and $\langle\vec{v}_h\rangle$ denotes
157     the instantaneous bulk velocity of slabs $c$ and $h$ respectively
158     before an operation occurs. When a momentum flux $\vec{j}_z(\vec{p})$
159     presents, these bulk velocities would have a corresponding change
160     ($\vec{a}_c$ and $\vec{a}_h$ respectively) according to Newton's
161     second law:
162     \begin{eqnarray}
163     M_c \vec{a}_c & = & -\vec{j}_z(\vec{p}) \Delta t \\
164     M_h \vec{a}_h & = & \vec{j}_z(\vec{p}) \Delta t
165     \end{eqnarray}
166     where
167     \begin{eqnarray}
168     M_c & = & \sum_{i = 1}^{N_c} m_i \\
169     M_h & = & \sum_{j = 1}^{N_h} m_j
170     \end{eqnarray}
171     and $\Delta t$ is the interval between two operations.
172    
173     The above operations conserve the linear momentum of a periodic
174     system. To satisfy total energy conservation as well as to impose a
175     thermal flux $J_z$, one would have
176 skuang 3771 [SUPPORT INFO MIGHT BE NECESSARY TO PUT EXTRA MATH IN]
177 skuang 3770 \begin{eqnarray}
178     K_c - J_z\Delta t & = & c^2 (K_c - \frac{1}{2}M_c \langle\vec{v}_c
179     \rangle^2) + \frac{1}{2}M_c (\langle \vec{v}_c \rangle + \vec{a}_c)^2 \\
180     K_h + J_z\Delta t & = & h^2 (K_h - \frac{1}{2}M_h \langle\vec{v}_h
181     \rangle^2) + \frac{1}{2}M_h (\langle \vec{v}_h \rangle + \vec{a}_h)^2
182     \end{eqnarray}
183     where $K_c$ and $K_h$ denotes translational kinetic energy of slabs
184     $c$ and $h$ respectively before an operation occurs. These
185     translational kinetic energy conservation equations are sufficient to
186     ensure total energy conservation, as the operations applied do not
187     change the potential energy of a system, given that the potential
188     energy does not depend on particle velocity.
189    
190     The above sets of equations are sufficient to determine the velocity
191     scaling coefficients ($c$ and $h$) as well as $\vec{a}_c$ and
192     $\vec{a}_h$. Note that two roots of $c$ and $h$ exist
193     respectively. However, to avoid dramatic perturbations to a system,
194 skuang 3772 the positive roots (which are closer to 1) are chosen. Figure
195     \ref{method} illustrates the implementation of this algorithm in an
196     individual step.
197 skuang 3770
198 skuang 3772 \begin{figure}
199     \includegraphics[width=\linewidth]{method}
200     \caption{Illustration of the implementation of the algorithm in a
201     single step. Starting from an ideal velocity distribution, the
202     transformation is used to apply both thermal and momentum flux from
203     the ``c'' slab to the ``h'' slab. As the figure shows, the thermal
204     distributions preserve after this operation.}
205     \label{method}
206     \end{figure}
207    
208 skuang 3770 By implementing these operations at a certain frequency, a steady
209     thermal and/or momentum flux can be applied and the corresponding
210     temperature and/or momentum gradients can be established.
211    
212 skuang 3771 This approach is more computationaly efficient compared to the
213     previous NIVS method, in that only quadratic equations are involved,
214     while the NIVS method needs to solve a quartic equations. Furthermore,
215     the method implements isotropic scaling of velocities in respective
216     slabs, unlike the NIVS, where an extra criteria function is necessary
217     to choose a set of coefficients that performs the most isotropic
218     scaling. More importantly, separating the momentum flux imposing from
219     velocity scaling avoids the underlying cause that NIVS produced
220     thermal anisotropy when applying a momentum flux.
221     %NEW METHOD DOESN'T CAUSE UNDESIRED CONCOMITENT MOMENTUM FLUX WHEN
222     %IMPOSING A THERMAL FLUX
223 gezelter 3769
224 skuang 3772 The advantages of the approach over the original momentum swapping
225     approach lies in its nature to preserve a Gaussian
226     distribution. Because the momentum swapping tends to render a
227     nonthermal distribution, when the imposed flux is relatively large,
228     diffusion of the neighboring slabs could no longer remedy this effect,
229     and nonthermal distributions would be observed. Results in later
230     section will illustrate this effect.
231 gezelter 3769
232 skuang 3772 \section{Computational Details}
233 gezelter 3769
234 skuang 3772
235 gezelter 3769 \subsection{Simulation Protocol}
236     The NIVS algorithm has been implemented in our MD simulation code,
237     OpenMD\cite{Meineke:2005gd,openmd}, and was used for our simulations.
238     Metal slabs of 6 or 11 layers of Au atoms were first equilibrated
239     under atmospheric pressure (1 atm) and 200K. After equilibration,
240     butanethiol capping agents were placed at three-fold hollow sites on
241     the Au(111) surfaces. These sites are either {\it fcc} or {\it
242     hcp} sites, although Hase {\it et al.} found that they are
243     equivalent in a heat transfer process,\cite{hase:2010} so we did not
244     distinguish between these sites in our study. The maximum butanethiol
245     capacity on Au surface is $1/3$ of the total number of surface Au
246     atoms, and the packing forms a $(\sqrt{3}\times\sqrt{3})R30^\circ$
247     structure\cite{doi:10.1021/ja00008a001,doi:10.1021/cr9801317}. A
248     series of lower coverages was also prepared by eliminating
249     butanethiols from the higher coverage surface in a regular manner. The
250     lower coverages were prepared in order to study the relation between
251     coverage and interfacial conductance.
252    
253     The capping agent molecules were allowed to migrate during the
254     simulations. They distributed themselves uniformly and sampled a
255     number of three-fold sites throughout out study. Therefore, the
256     initial configuration does not noticeably affect the sampling of a
257     variety of configurations of the same coverage, and the final
258     conductance measurement would be an average effect of these
259     configurations explored in the simulations.
260    
261     After the modified Au-butanethiol surface systems were equilibrated in
262     the canonical (NVT) ensemble, organic solvent molecules were packed in
263     the previously empty part of the simulation cells.\cite{packmol} Two
264     solvents were investigated, one which has little vibrational overlap
265     with the alkanethiol and which has a planar shape (toluene), and one
266     which has similar vibrational frequencies to the capping agent and
267     chain-like shape ({\it n}-hexane).
268    
269     The simulation cells were not particularly extensive along the
270     $z$-axis, as a very long length scale for the thermal gradient may
271     cause excessively hot or cold temperatures in the middle of the
272     solvent region and lead to undesired phenomena such as solvent boiling
273     or freezing when a thermal flux is applied. Conversely, too few
274     solvent molecules would change the normal behavior of the liquid
275     phase. Therefore, our $N_{solvent}$ values were chosen to ensure that
276     these extreme cases did not happen to our simulations. The spacing
277     between periodic images of the gold interfaces is $45 \sim 75$\AA in
278     our simulations.
279    
280     The initial configurations generated are further equilibrated with the
281     $x$ and $y$ dimensions fixed, only allowing the $z$-length scale to
282     change. This is to ensure that the equilibration of liquid phase does
283     not affect the metal's crystalline structure. Comparisons were made
284     with simulations that allowed changes of $L_x$ and $L_y$ during NPT
285     equilibration. No substantial changes in the box geometry were noticed
286     in these simulations. After ensuring the liquid phase reaches
287     equilibrium at atmospheric pressure (1 atm), further equilibration was
288     carried out under canonical (NVT) and microcanonical (NVE) ensembles.
289    
290     After the systems reach equilibrium, NIVS was used to impose an
291     unphysical thermal flux between the metal and the liquid phases. Most
292     of our simulations were done under an average temperature of
293     $\sim$200K. Therefore, thermal flux usually came from the metal to the
294     liquid so that the liquid has a higher temperature and would not
295     freeze due to lowered temperatures. After this induced temperature
296     gradient had stabilized, the temperature profile of the simulation cell
297     was recorded. To do this, the simulation cell is divided evenly into
298     $N$ slabs along the $z$-axis. The average temperatures of each slab
299     are recorded for 1$\sim$2 ns. When the slab width $d$ of each slab is
300     the same, the derivatives of $T$ with respect to slab number $n$ can
301     be directly used for $G^\prime$ calculations: \begin{equation}
302     G^\prime = |J_z|\Big|\frac{\partial^2 T}{\partial z^2}\Big|
303     \Big/\left(\frac{\partial T}{\partial z}\right)^2
304     = |J_z|\Big|\frac{1}{d^2}\frac{\partial^2 T}{\partial n^2}\Big|
305     \Big/\left(\frac{1}{d}\frac{\partial T}{\partial n}\right)^2
306     = |J_z|\Big|\frac{\partial^2 T}{\partial n^2}\Big|
307     \Big/\left(\frac{\partial T}{\partial n}\right)^2
308     \label{derivativeG2}
309     \end{equation}
310     The absolute values in Eq. \ref{derivativeG2} appear because the
311     direction of the flux $\vec{J}$ is in an opposing direction on either
312     side of the metal slab.
313    
314     All of the above simulation procedures use a time step of 1 fs. Each
315     equilibration stage took a minimum of 100 ps, although in some cases,
316     longer equilibration stages were utilized.
317    
318     \subsection{Force Field Parameters}
319     Our simulations include a number of chemically distinct components.
320     Figure \ref{demoMol} demonstrates the sites defined for both
321     United-Atom and All-Atom models of the organic solvent and capping
322     agents in our simulations. Force field parameters are needed for
323     interactions both between the same type of particles and between
324     particles of different species.
325    
326     \begin{figure}
327     \includegraphics[width=\linewidth]{structures}
328     \caption{Structures of the capping agent and solvents utilized in
329     these simulations. The chemically-distinct sites (a-e) are expanded
330     in terms of constituent atoms for both United Atom (UA) and All Atom
331     (AA) force fields. Most parameters are from References
332     \protect\cite{TraPPE-UA.alkanes,TraPPE-UA.alkylbenzenes,TraPPE-UA.thiols}
333     (UA) and \protect\cite{OPLSAA} (AA). Cross-interactions with the Au
334     atoms are given in Table 1 in the supporting information.}
335     \label{demoMol}
336     \end{figure}
337    
338     The Au-Au interactions in metal lattice slab is described by the
339     quantum Sutton-Chen (QSC) formulation.\cite{PhysRevB.59.3527} The QSC
340     potentials include zero-point quantum corrections and are
341     reparametrized for accurate surface energies compared to the
342     Sutton-Chen potentials.\cite{Chen90}
343    
344     For the two solvent molecules, {\it n}-hexane and toluene, two
345     different atomistic models were utilized. Both solvents were modeled
346     using United-Atom (UA) and All-Atom (AA) models. The TraPPE-UA
347     parameters\cite{TraPPE-UA.alkanes,TraPPE-UA.alkylbenzenes} are used
348     for our UA solvent molecules. In these models, sites are located at
349     the carbon centers for alkyl groups. Bonding interactions, including
350     bond stretches and bends and torsions, were used for intra-molecular
351     sites closer than 3 bonds. For non-bonded interactions, Lennard-Jones
352     potentials are used.
353    
354     By eliminating explicit hydrogen atoms, the TraPPE-UA models are
355     simple and computationally efficient, while maintaining good accuracy.
356     However, the TraPPE-UA model for alkanes is known to predict a slightly
357     lower boiling point than experimental values. This is one of the
358     reasons we used a lower average temperature (200K) for our
359     simulations. If heat is transferred to the liquid phase during the
360     NIVS simulation, the liquid in the hot slab can actually be
361     substantially warmer than the mean temperature in the simulation. The
362     lower mean temperatures therefore prevent solvent boiling.
363    
364     For UA-toluene, the non-bonded potentials between intermolecular sites
365     have a similar Lennard-Jones formulation. The toluene molecules were
366     treated as a single rigid body, so there was no need for
367     intramolecular interactions (including bonds, bends, or torsions) in
368     this solvent model.
369    
370     Besides the TraPPE-UA models, AA models for both organic solvents are
371     included in our studies as well. The OPLS-AA\cite{OPLSAA} force fields
372     were used. For hexane, additional explicit hydrogen sites were
373     included. Besides bonding and non-bonded site-site interactions,
374     partial charges and the electrostatic interactions were added to each
375     CT and HC site. For toluene, a flexible model for the toluene molecule
376     was utilized which included bond, bend, torsion, and inversion
377     potentials to enforce ring planarity.
378    
379     The butanethiol capping agent in our simulations, were also modeled
380     with both UA and AA model. The TraPPE-UA force field includes
381     parameters for thiol molecules\cite{TraPPE-UA.thiols} and are used for
382     UA butanethiol model in our simulations. The OPLS-AA also provides
383     parameters for alkyl thiols. However, alkyl thiols adsorbed on Au(111)
384     surfaces do not have the hydrogen atom bonded to sulfur. To derive
385     suitable parameters for butanethiol adsorbed on Au(111) surfaces, we
386     adopt the S parameters from Luedtke and Landman\cite{landman:1998} and
387     modify the parameters for the CTS atom to maintain charge neutrality
388     in the molecule. Note that the model choice (UA or AA) for the capping
389     agent can be different from the solvent. Regardless of model choice,
390     the force field parameters for interactions between capping agent and
391     solvent can be derived using Lorentz-Berthelot Mixing Rule:
392     \begin{eqnarray}
393     \sigma_{ij} & = & \frac{1}{2} \left(\sigma_{ii} + \sigma_{jj}\right) \\
394     \epsilon_{ij} & = & \sqrt{\epsilon_{ii}\epsilon_{jj}}
395     \end{eqnarray}
396    
397     To describe the interactions between metal (Au) and non-metal atoms,
398     we refer to an adsorption study of alkyl thiols on gold surfaces by
399     Vlugt {\it et al.}\cite{vlugt:cpc2007154} They fitted an effective
400     Lennard-Jones form of potential parameters for the interaction between
401     Au and pseudo-atoms CH$_x$ and S based on a well-established and
402     widely-used effective potential of Hautman and Klein for the Au(111)
403     surface.\cite{hautman:4994} As our simulations require the gold slab
404     to be flexible to accommodate thermal excitation, the pair-wise form
405     of potentials they developed was used for our study.
406    
407     The potentials developed from {\it ab initio} calculations by Leng
408     {\it et al.}\cite{doi:10.1021/jp034405s} are adopted for the
409     interactions between Au and aromatic C/H atoms in toluene. However,
410     the Lennard-Jones parameters between Au and other types of particles,
411     (e.g. AA alkanes) have not yet been established. For these
412     interactions, the Lorentz-Berthelot mixing rule can be used to derive
413     effective single-atom LJ parameters for the metal using the fit values
414     for toluene. These are then used to construct reasonable mixing
415     parameters for the interactions between the gold and other atoms.
416     Table 1 in the supporting information summarizes the
417     ``metal/non-metal'' parameters utilized in our simulations.
418    
419     \section{Results}
420 skuang 3770 [L-J COMPARED TO RENMD NIVS; WATER COMPARED TO RNEMD NIVS;
421     SLIP BOUNDARY VS STICK BOUNDARY; ICE-WATER INTERFACES]
422    
423 gezelter 3769 There are many factors contributing to the measured interfacial
424     conductance; some of these factors are physically motivated
425     (e.g. coverage of the surface by the capping agent coverage and
426     solvent identity), while some are governed by parameters of the
427     methodology (e.g. applied flux and the formulas used to obtain the
428     conductance). In this section we discuss the major physical and
429     calculational effects on the computed conductivity.
430    
431     \subsection{Effects due to capping agent coverage}
432    
433     A series of different initial conditions with a range of surface
434     coverages was prepared and solvated with various with both of the
435     solvent molecules. These systems were then equilibrated and their
436     interfacial thermal conductivity was measured with the NIVS
437     algorithm. Figure \ref{coverage} demonstrates the trend of conductance
438     with respect to surface coverage.
439    
440     \begin{figure}
441     \includegraphics[width=\linewidth]{coverage}
442     \caption{The interfacial thermal conductivity ($G$) has a
443     non-monotonic dependence on the degree of surface capping. This
444     data is for the Au(111) / butanethiol / solvent interface with
445     various UA force fields at $\langle T\rangle \sim $200K.}
446     \label{coverage}
447     \end{figure}
448    
449     In partially covered surfaces, the derivative definition for
450     $G^\prime$ (Eq. \ref{derivativeG}) becomes difficult to apply, as the
451     location of maximum change of $\lambda$ becomes washed out. The
452     discrete definition (Eq. \ref{discreteG}) is easier to apply, as the
453     Gibbs dividing surface is still well-defined. Therefore, $G$ (not
454     $G^\prime$) was used in this section.
455    
456     From Figure \ref{coverage}, one can see the significance of the
457     presence of capping agents. When even a small fraction of the Au(111)
458     surface sites are covered with butanethiols, the conductivity exhibits
459     an enhancement by at least a factor of 3. Capping agents are clearly
460     playing a major role in thermal transport at metal / organic solvent
461     surfaces.
462    
463     We note a non-monotonic behavior in the interfacial conductance as a
464     function of surface coverage. The maximum conductance (largest $G$)
465     happens when the surfaces are about 75\% covered with butanethiol
466     caps. The reason for this behavior is not entirely clear. One
467     explanation is that incomplete butanethiol coverage allows small gaps
468     between butanethiols to form. These gaps can be filled by transient
469     solvent molecules. These solvent molecules couple very strongly with
470     the hot capping agent molecules near the surface, and can then carry
471     away (diffusively) the excess thermal energy from the surface.
472    
473     There appears to be a competition between the conduction of the
474     thermal energy away from the surface by the capping agents (enhanced
475     by greater coverage) and the coupling of the capping agents with the
476     solvent (enhanced by interdigitation at lower coverages). This
477     competition would lead to the non-monotonic coverage behavior observed
478     here.
479    
480     Results for rigid body toluene solvent, as well as the UA hexane, are
481     within the ranges expected from prior experimental
482     work.\cite{Wilson:2002uq,cahill:793,PhysRevB.80.195406} This suggests
483     that explicit hydrogen atoms might not be required for modeling
484     thermal transport in these systems. C-H vibrational modes do not see
485     significant excited state population at low temperatures, and are not
486     likely to carry lower frequency excitations from the solid layer into
487     the bulk liquid.
488    
489     The toluene solvent does not exhibit the same behavior as hexane in
490     that $G$ remains at approximately the same magnitude when the capping
491     coverage increases from 25\% to 75\%. Toluene, as a rigid planar
492     molecule, cannot occupy the relatively small gaps between the capping
493     agents as easily as the chain-like {\it n}-hexane. The effect of
494     solvent coupling to the capping agent is therefore weaker in toluene
495     except at the very lowest coverage levels. This effect counters the
496     coverage-dependent conduction of heat away from the metal surface,
497     leading to a much flatter $G$ vs. coverage trend than is observed in
498     {\it n}-hexane.
499    
500     \subsection{Effects due to Solvent \& Solvent Models}
501     In addition to UA solvent and capping agent models, AA models have
502     also been included in our simulations. In most of this work, the same
503     (UA or AA) model for solvent and capping agent was used, but it is
504     also possible to utilize different models for different components.
505     We have also included isotopic substitutions (Hydrogen to Deuterium)
506     to decrease the explicit vibrational overlap between solvent and
507     capping agent. Table \ref{modelTest} summarizes the results of these
508     studies.
509    
510     \begin{table*}
511     \begin{minipage}{\linewidth}
512     \begin{center}
513    
514     \caption{Computed interfacial thermal conductance ($G$ and
515     $G^\prime$) values for interfaces using various models for
516     solvent and capping agent (or without capping agent) at
517     $\langle T\rangle\sim$200K. Here ``D'' stands for deuterated
518     solvent or capping agent molecules. Error estimates are
519     indicated in parentheses.}
520    
521     \begin{tabular}{llccc}
522     \hline\hline
523     Butanethiol model & Solvent & $G$ & $G^\prime$ \\
524     (or bare surface) & model & \multicolumn{2}{c}{(MW/m$^2$/K)} \\
525     \hline
526     UA & UA hexane & 131(9) & 87(10) \\
527     & UA hexane(D) & 153(5) & 136(13) \\
528     & AA hexane & 131(6) & 122(10) \\
529     & UA toluene & 187(16) & 151(11) \\
530     & AA toluene & 200(36) & 149(53) \\
531     \hline
532     AA & UA hexane & 116(9) & 129(8) \\
533     & AA hexane & 442(14) & 356(31) \\
534     & AA hexane(D) & 222(12) & 234(54) \\
535     & UA toluene & 125(25) & 97(60) \\
536     & AA toluene & 487(56) & 290(42) \\
537     \hline
538     AA(D) & UA hexane & 158(25) & 172(4) \\
539     & AA hexane & 243(29) & 191(11) \\
540     & AA toluene & 364(36) & 322(67) \\
541     \hline
542     bare & UA hexane & 46.5(3.2) & 49.4(4.5) \\
543     & UA hexane(D) & 43.9(4.6) & 43.0(2.0) \\
544     & AA hexane & 31.0(1.4) & 29.4(1.3) \\
545     & UA toluene & 70.1(1.3) & 65.8(0.5) \\
546     \hline\hline
547     \end{tabular}
548     \label{modelTest}
549     \end{center}
550     \end{minipage}
551     \end{table*}
552    
553     To facilitate direct comparison between force fields, systems with the
554     same capping agent and solvent were prepared with the same length
555     scales for the simulation cells.
556    
557     On bare metal / solvent surfaces, different force field models for
558     hexane yield similar results for both $G$ and $G^\prime$, and these
559     two definitions agree with each other very well. This is primarily an
560     indicator of weak interactions between the metal and the solvent.
561    
562     For the fully-covered surfaces, the choice of force field for the
563     capping agent and solvent has a large impact on the calculated values
564     of $G$ and $G^\prime$. The AA thiol to AA solvent conductivities are
565     much larger than their UA to UA counterparts, and these values exceed
566     the experimental estimates by a large measure. The AA force field
567     allows significant energy to go into C-H (or C-D) stretching modes,
568     and since these modes are high frequency, this non-quantum behavior is
569     likely responsible for the overestimate of the conductivity. Compared
570     to the AA model, the UA model yields more reasonable conductivity
571     values with much higher computational efficiency.
572    
573     \subsubsection{Are electronic excitations in the metal important?}
574     Because they lack electronic excitations, the QSC and related embedded
575     atom method (EAM) models for gold are known to predict unreasonably
576     low values for bulk conductivity
577     ($\lambda$).\cite{kuang:164101,ISI:000207079300006,Clancy:1992} If the
578     conductance between the phases ($G$) is governed primarily by phonon
579     excitation (and not electronic degrees of freedom), one would expect a
580     classical model to capture most of the interfacial thermal
581     conductance. Our results for $G$ and $G^\prime$ indicate that this is
582     indeed the case, and suggest that the modeling of interfacial thermal
583     transport depends primarily on the description of the interactions
584     between the various components at the interface. When the metal is
585     chemically capped, the primary barrier to thermal conductivity appears
586     to be the interface between the capping agent and the surrounding
587     solvent, so the excitations in the metal have little impact on the
588     value of $G$.
589    
590     \subsection{Effects due to methodology and simulation parameters}
591    
592     We have varied the parameters of the simulations in order to
593     investigate how these factors would affect the computation of $G$. Of
594     particular interest are: 1) the length scale for the applied thermal
595     gradient (modified by increasing the amount of solvent in the system),
596     2) the sign and magnitude of the applied thermal flux, 3) the average
597     temperature of the simulation (which alters the solvent density during
598     equilibration), and 4) the definition of the interfacial conductance
599     (Eqs. (\ref{discreteG}) or (\ref{derivativeG})) used in the
600     calculation.
601    
602     Systems of different lengths were prepared by altering the number of
603     solvent molecules and extending the length of the box along the $z$
604     axis to accomodate the extra solvent. Equilibration at the same
605     temperature and pressure conditions led to nearly identical surface
606     areas ($L_x$ and $L_y$) available to the metal and capping agent,
607     while the extra solvent served mainly to lengthen the axis that was
608     used to apply the thermal flux. For a given value of the applied
609     flux, the different $z$ length scale has only a weak effect on the
610     computed conductivities.
611    
612     \subsubsection{Effects of applied flux}
613     The NIVS algorithm allows changes in both the sign and magnitude of
614     the applied flux. It is possible to reverse the direction of heat
615     flow simply by changing the sign of the flux, and thermal gradients
616     which would be difficult to obtain experimentally ($5$ K/\AA) can be
617     easily simulated. However, the magnitude of the applied flux is not
618     arbitrary if one aims to obtain a stable and reliable thermal gradient.
619     A temperature gradient can be lost in the noise if $|J_z|$ is too
620     small, and excessive $|J_z|$ values can cause phase transitions if the
621     extremes of the simulation cell become widely separated in
622     temperature. Also, if $|J_z|$ is too large for the bulk conductivity
623     of the materials, the thermal gradient will never reach a stable
624     state.
625    
626     Within a reasonable range of $J_z$ values, we were able to study how
627     $G$ changes as a function of this flux. In what follows, we use
628     positive $J_z$ values to denote the case where energy is being
629     transferred by the method from the metal phase and into the liquid.
630     The resulting gradient therefore has a higher temperature in the
631     liquid phase. Negative flux values reverse this transfer, and result
632     in higher temperature metal phases. The conductance measured under
633     different applied $J_z$ values is listed in Tables 2 and 3 in the
634     supporting information. These results do not indicate that $G$ depends
635     strongly on $J_z$ within this flux range. The linear response of flux
636     to thermal gradient simplifies our investigations in that we can rely
637     on $G$ measurement with only a small number $J_z$ values.
638    
639     The sign of $J_z$ is a different matter, however, as this can alter
640     the temperature on the two sides of the interface. The average
641     temperature values reported are for the entire system, and not for the
642     liquid phase, so at a given $\langle T \rangle$, the system with
643     positive $J_z$ has a warmer liquid phase. This means that if the
644     liquid carries thermal energy via diffusive transport, {\it positive}
645     $J_z$ values will result in increased molecular motion on the liquid
646     side of the interface, and this will increase the measured
647     conductivity.
648    
649     \subsubsection{Effects due to average temperature}
650    
651     We also studied the effect of average system temperature on the
652     interfacial conductance. The simulations are first equilibrated in
653     the NPT ensemble at 1 atm. The TraPPE-UA model for hexane tends to
654     predict a lower boiling point (and liquid state density) than
655     experiments. This lower-density liquid phase leads to reduced contact
656     between the hexane and butanethiol, and this accounts for our
657     observation of lower conductance at higher temperatures. In raising
658     the average temperature from 200K to 250K, the density drop of
659     $\sim$20\% in the solvent phase leads to a $\sim$40\% drop in the
660     conductance.
661    
662     Similar behavior is observed in the TraPPE-UA model for toluene,
663     although this model has better agreement with the experimental
664     densities of toluene. The expansion of the toluene liquid phase is
665     not as significant as that of the hexane (8.3\% over 100K), and this
666     limits the effect to $\sim$20\% drop in thermal conductivity.
667    
668     Although we have not mapped out the behavior at a large number of
669     temperatures, is clear that there will be a strong temperature
670     dependence in the interfacial conductance when the physical properties
671     of one side of the interface (notably the density) change rapidly as a
672     function of temperature.
673    
674     Besides the lower interfacial thermal conductance, surfaces at
675     relatively high temperatures are susceptible to reconstructions,
676     particularly when butanethiols fully cover the Au(111) surface. These
677     reconstructions include surface Au atoms which migrate outward to the
678     S atom layer, and butanethiol molecules which embed into the surface
679     Au layer. The driving force for this behavior is the strong Au-S
680     interactions which are modeled here with a deep Lennard-Jones
681     potential. This phenomenon agrees with reconstructions that have been
682     experimentally
683     observed.\cite{doi:10.1021/j100035a033,doi:10.1021/la026493y}. Vlugt
684     {\it et al.} kept their Au(111) slab rigid so that their simulations
685     could reach 300K without surface
686     reconstructions.\cite{vlugt:cpc2007154} Since surface reconstructions
687     blur the interface, the measurement of $G$ becomes more difficult to
688     conduct at higher temperatures. For this reason, most of our
689     measurements are undertaken at $\langle T\rangle\sim$200K where
690     reconstruction is minimized.
691    
692     However, when the surface is not completely covered by butanethiols,
693     the simulated system appears to be more resistent to the
694     reconstruction. Our Au / butanethiol / toluene system had the Au(111)
695     surfaces 90\% covered by butanethiols, but did not see this above
696     phenomena even at $\langle T\rangle\sim$300K. That said, we did
697     observe butanethiols migrating to neighboring three-fold sites during
698     a simulation. Since the interface persisted in these simulations, we
699     were able to obtain $G$'s for these interfaces even at a relatively
700     high temperature without being affected by surface reconstructions.
701    
702     \section{Discussion}
703 skuang 3770 [COMBINE W. RESULTS]
704 gezelter 3769 The primary result of this work is that the capping agent acts as an
705     efficient thermal coupler between solid and solvent phases. One of
706     the ways the capping agent can carry out this role is to down-shift
707     between the phonon vibrations in the solid (which carry the heat from
708     the gold) and the molecular vibrations in the liquid (which carry some
709     of the heat in the solvent).
710    
711     To investigate the mechanism of interfacial thermal conductance, the
712     vibrational power spectrum was computed. Power spectra were taken for
713     individual components in different simulations. To obtain these
714     spectra, simulations were run after equilibration in the
715     microcanonical (NVE) ensemble and without a thermal
716     gradient. Snapshots of configurations were collected at a frequency
717     that is higher than that of the fastest vibrations occurring in the
718     simulations. With these configurations, the velocity auto-correlation
719     functions can be computed:
720     \begin{equation}
721     C_A (t) = \langle\vec{v}_A (t)\cdot\vec{v}_A (0)\rangle
722     \label{vCorr}
723     \end{equation}
724     The power spectrum is constructed via a Fourier transform of the
725     symmetrized velocity autocorrelation function,
726     \begin{equation}
727     \hat{f}(\omega) =
728     \int_{-\infty}^{\infty} C_A (t) e^{-2\pi it\omega}\,dt
729     \label{fourier}
730     \end{equation}
731    
732     \subsection{The role of specific vibrations}
733     The vibrational spectra for gold slabs in different environments are
734     shown as in Figure \ref{specAu}. Regardless of the presence of
735     solvent, the gold surfaces which are covered by butanethiol molecules
736     exhibit an additional peak observed at a frequency of
737     $\sim$165cm$^{-1}$. We attribute this peak to the S-Au bonding
738     vibration. This vibration enables efficient thermal coupling of the
739     surface Au layer to the capping agents. Therefore, in our simulations,
740     the Au / S interfaces do not appear to be the primary barrier to
741     thermal transport when compared with the butanethiol / solvent
742     interfaces. This supports the results of Luo {\it et
743     al.}\cite{Luo20101}, who reported $G$ for Au-SAM junctions roughly
744     twice as large as what we have computed for the thiol-liquid
745     interfaces.
746    
747     \begin{figure}
748     \includegraphics[width=\linewidth]{vibration}
749     \caption{The vibrational power spectrum for thiol-capped gold has an
750     additional vibrational peak at $\sim $165cm$^{-1}$. Bare gold
751     surfaces (both with and without a solvent over-layer) are missing
752     this peak. A similar peak at $\sim $165cm$^{-1}$ also appears in
753     the vibrational power spectrum for the butanethiol capping agents.}
754     \label{specAu}
755     \end{figure}
756    
757     Also in this figure, we show the vibrational power spectrum for the
758     bound butanethiol molecules, which also exhibits the same
759     $\sim$165cm$^{-1}$ peak.
760    
761     \subsection{Overlap of power spectra}
762     A comparison of the results obtained from the two different organic
763     solvents can also provide useful information of the interfacial
764     thermal transport process. In particular, the vibrational overlap
765     between the butanethiol and the organic solvents suggests a highly
766     efficient thermal exchange between these components. Very high
767     thermal conductivity was observed when AA models were used and C-H
768     vibrations were treated classically. The presence of extra degrees of
769     freedom in the AA force field yields higher heat exchange rates
770     between the two phases and results in a much higher conductivity than
771     in the UA force field. The all-atom classical models include high
772     frequency modes which should be unpopulated at our relatively low
773     temperatures. This artifact is likely the cause of the high thermal
774     conductance in all-atom MD simulations.
775    
776     The similarity in the vibrational modes available to solvent and
777     capping agent can be reduced by deuterating one of the two components
778     (Fig. \ref{aahxntln}). Once either the hexanes or the butanethiols
779     are deuterated, one can observe a significantly lower $G$ and
780     $G^\prime$ values (Table \ref{modelTest}).
781    
782     \begin{figure}
783     \includegraphics[width=\linewidth]{aahxntln}
784     \caption{Spectra obtained for all-atom (AA) Au / butanethiol / solvent
785     systems. When butanethiol is deuterated (lower left), its
786     vibrational overlap with hexane decreases significantly. Since
787     aromatic molecules and the butanethiol are vibrationally dissimilar,
788     the change is not as dramatic when toluene is the solvent (right).}
789     \label{aahxntln}
790     \end{figure}
791    
792     For the Au / butanethiol / toluene interfaces, having the AA
793     butanethiol deuterated did not yield a significant change in the
794     measured conductance. Compared to the C-H vibrational overlap between
795     hexane and butanethiol, both of which have alkyl chains, the overlap
796     between toluene and butanethiol is not as significant and thus does
797     not contribute as much to the heat exchange process.
798    
799     Previous observations of Zhang {\it et al.}\cite{hase:2010} indicate
800     that the {\it intra}molecular heat transport due to alkylthiols is
801     highly efficient. Combining our observations with those of Zhang {\it
802     et al.}, it appears that butanethiol acts as a channel to expedite
803     heat flow from the gold surface and into the alkyl chain. The
804     vibrational coupling between the metal and the liquid phase can
805     therefore be enhanced with the presence of suitable capping agents.
806    
807     Deuterated models in the UA force field did not decouple the thermal
808     transport as well as in the AA force field. The UA models, even
809     though they have eliminated the high frequency C-H vibrational
810     overlap, still have significant overlap in the lower-frequency
811     portions of the infrared spectra (Figure \ref{uahxnua}). Deuterating
812     the UA models did not decouple the low frequency region enough to
813     produce an observable difference for the results of $G$ (Table
814     \ref{modelTest}).
815    
816     \begin{figure}
817     \includegraphics[width=\linewidth]{uahxnua}
818     \caption{Vibrational power spectra for UA models for the butanethiol
819     and hexane solvent (upper panel) show the high degree of overlap
820     between these two molecules, particularly at lower frequencies.
821     Deuterating a UA model for the solvent (lower panel) does not
822     decouple the two spectra to the same degree as in the AA force
823     field (see Fig \ref{aahxntln}).}
824     \label{uahxnua}
825     \end{figure}
826    
827     \section{Conclusions}
828     The NIVS algorithm has been applied to simulations of
829     butanethiol-capped Au(111) surfaces in the presence of organic
830     solvents. This algorithm allows the application of unphysical thermal
831     flux to transfer heat between the metal and the liquid phase. With the
832     flux applied, we were able to measure the corresponding thermal
833     gradients and to obtain interfacial thermal conductivities. Under
834     steady states, 2-3 ns trajectory simulations are sufficient for
835     computation of this quantity.
836    
837     Our simulations have seen significant conductance enhancement in the
838     presence of capping agent, compared with the bare gold / liquid
839     interfaces. The vibrational coupling between the metal and the liquid
840     phase is enhanced by a chemically-bonded capping agent. Furthermore,
841     the coverage percentage of the capping agent plays an important role
842     in the interfacial thermal transport process. Moderately low coverages
843     allow higher contact between capping agent and solvent, and thus could
844     further enhance the heat transfer process, giving a non-monotonic
845     behavior of conductance with increasing coverage.
846    
847     Our results, particularly using the UA models, agree well with
848     available experimental data. The AA models tend to overestimate the
849     interfacial thermal conductance in that the classically treated C-H
850     vibrations become too easily populated. Compared to the AA models, the
851     UA models have higher computational efficiency with satisfactory
852     accuracy, and thus are preferable in modeling interfacial thermal
853     transport.
854    
855     Of the two definitions for $G$, the discrete form
856     (Eq. \ref{discreteG}) was easier to use and gives out relatively
857     consistent results, while the derivative form (Eq. \ref{derivativeG})
858     is not as versatile. Although $G^\prime$ gives out comparable results
859     and follows similar trend with $G$ when measuring close to fully
860     covered or bare surfaces, the spatial resolution of $T$ profile
861     required for the use of a derivative form is limited by the number of
862     bins and the sampling required to obtain thermal gradient information.
863    
864     Vlugt {\it et al.} have investigated the surface thiol structures for
865     nanocrystalline gold and pointed out that they differ from those of
866     the Au(111) surface.\cite{landman:1998,vlugt:cpc2007154} This
867     difference could also cause differences in the interfacial thermal
868     transport behavior. To investigate this problem, one would need an
869     effective method for applying thermal gradients in non-planar
870     (i.e. spherical) geometries.
871    
872     \section{Acknowledgments}
873     Support for this project was provided by the National Science
874     Foundation under grant CHE-0848243. Computational time was provided by
875     the Center for Research Computing (CRC) at the University of Notre
876     Dame.
877    
878     \newpage
879    
880 skuang 3770 \bibliography{stokes}
881 gezelter 3769
882     \end{doublespace}
883     \end{document}
884