1 |
\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{booktabs} |
10 |
%\usepackage{bibentry} |
11 |
%\usepackage{mathrsfs} |
12 |
\usepackage[ref]{overcite} |
13 |
\pagestyle{plain} \pagenumbering{arabic} \oddsidemargin 0.0cm |
14 |
\evensidemargin 0.0cm \topmargin -21pt \headsep 10pt \textheight |
15 |
9.0in \textwidth 6.5in \brokenpenalty=10000 |
16 |
|
17 |
% double space list of tables and figures |
18 |
\AtBeginDelayedFloats{\renewcommand{\baselinestretch}{1.66}} |
19 |
\setlength{\abovecaptionskip}{20 pt} |
20 |
\setlength{\belowcaptionskip}{30 pt} |
21 |
|
22 |
\renewcommand\citemid{\ } % no comma in optional referenc note |
23 |
|
24 |
\begin{document} |
25 |
|
26 |
\title{A gentler approach to RNEMD: Non-isotropic Velocity Scaling for computing Thermal Conductivity and Shear Viscosity} |
27 |
|
28 |
\author{Shenyu Kuang and J. Daniel |
29 |
Gezelter\footnote{Corresponding author. \ Electronic mail: gezelter@nd.edu} \\ |
30 |
Department of Chemistry and Biochemistry,\\ |
31 |
University of Notre Dame\\ |
32 |
Notre Dame, Indiana 46556} |
33 |
|
34 |
\date{\today} |
35 |
|
36 |
\maketitle |
37 |
|
38 |
\begin{doublespace} |
39 |
|
40 |
\begin{abstract} |
41 |
We present a new method for introducing stable non-equilibrium |
42 |
velocity and temperature distributions in molecular dynamics |
43 |
simulations of heterogeneous systems. This method extends some |
44 |
earlier Reverse Non-Equilibrium Molecular Dynamics (RNEMD) methods |
45 |
which use momentum exchange swapping moves that can create |
46 |
non-thermal velocity distributions (and which are difficult to use |
47 |
for interfacial calculations). By using non-isotropic velocity |
48 |
scaling (NIVS) on the molecules in specific regions of a system, it |
49 |
is possible to impose momentum or thermal flux between regions of a |
50 |
simulation and stable thermal and momentum gradients can then be |
51 |
established. The scaling method we have developed conserves the |
52 |
total linear momentum and total energy of the system. To test the |
53 |
methods, we have computed the thermal conductivity of model liquid |
54 |
and solid systems as well as the interfacial thermal conductivity of |
55 |
a metal-water interface. We find that the NIVS-RNEMD improves the |
56 |
problematic velocity distributions that develop in other RNEMD |
57 |
methods. |
58 |
\end{abstract} |
59 |
|
60 |
\newpage |
61 |
|
62 |
%\narrowtext |
63 |
|
64 |
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
65 |
% BODY OF TEXT |
66 |
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% |
67 |
|
68 |
\section{Introduction} |
69 |
The original formulation of Reverse Non-equilibrium Molecular Dynamics |
70 |
(RNEMD) obtains transport coefficients (thermal conductivity and shear |
71 |
viscosity) in a fluid by imposing an artificial momentum flux between |
72 |
two thin parallel slabs of material that are spatially separated in |
73 |
the simulation cell.\cite{MullerPlathe:1997xw,ISI:000080382700030} The |
74 |
artificial flux is typically created by periodically ``swapping'' |
75 |
either the entire momentum vector $\vec{p}$ or single components of |
76 |
this vector ($p_x$) between molecules in each of the two slabs. If |
77 |
the two slabs are separated along the $z$ coordinate, the imposed flux |
78 |
is either directional ($j_z(p_x)$) or isotropic ($J_z$), and the |
79 |
response of a simulated system to the imposed momentum flux will |
80 |
typically be a velocity or thermal gradient (Fig. \ref{thermalDemo}). |
81 |
The shear viscosity ($\eta$) and thermal conductivity ($\lambda$) are |
82 |
easily obtained by assuming linear response of the system, |
83 |
\begin{eqnarray} |
84 |
j_z(p_x) & = & -\eta \frac{\partial v_x}{\partial z}\\ |
85 |
J_z & = & \lambda \frac{\partial T}{\partial z} |
86 |
\end{eqnarray} |
87 |
RNEMD has been widely used to provide computational estimates of thermal |
88 |
conductivities and shear viscosities in a wide range of materials, |
89 |
from liquid copper to monatomic liquids to molecular fluids |
90 |
(e.g. ionic liquids).\cite{Bedrov:2000-1,Bedrov:2000,Muller-Plathe:2002,ISI:000184808400018,ISI:000231042800044,Maginn:2007,Muller-Plathe:2008,ISI:000258460400020,ISI:000258840700015,ISI:000261835100054} |
91 |
|
92 |
\begin{figure} |
93 |
\includegraphics[width=\linewidth]{thermalDemo} |
94 |
\caption{RNEMD methods impose an unphysical transfer of momentum or |
95 |
kinetic energy between a ``hot'' slab and a ``cold'' slab in the |
96 |
simulation box. The molecular system responds to this imposed flux |
97 |
by generating a momentum or temperature gradient. The slope of the |
98 |
gradient can then be used to compute transport properties (e.g. |
99 |
shear viscosity and thermal conductivity).} |
100 |
\label{thermalDemo} |
101 |
\end{figure} |
102 |
|
103 |
RNEMD is preferable in many ways to the forward NEMD |
104 |
methods\cite{ISI:A1988Q205300014,hess:209,Vasquez:2004fk,backer:154503,ISI:000266247600008} |
105 |
because it imposes what is typically difficult to measure (a flux or |
106 |
stress) and it is typically much easier to compute momentum gradients |
107 |
or strains (the response). For similar reasons, RNEMD is also |
108 |
preferable to slowly-converging equilibrium methods for measuring |
109 |
thermal conductivity and shear viscosity (using Green-Kubo |
110 |
relations\cite{daivis:541,mondello:9327} or the Helfand moment |
111 |
approach of Viscardy {\it et |
112 |
al.}\cite{Viscardy:2007bh,Viscardy:2007lq}) because these rely on |
113 |
computing difficult to measure quantities. |
114 |
|
115 |
Another attractive feature of RNEMD is that it conserves both total |
116 |
linear momentum and total energy during the swaps (as long as the two |
117 |
molecules have the same identity), so the swapped configurations are |
118 |
typically samples from the same manifold of states in the |
119 |
microcanonical ensemble. |
120 |
|
121 |
Recently, Tenney and Maginn\cite{Maginn:2010} have discovered |
122 |
some problems with the original RNEMD swap technique. Notably, large |
123 |
momentum fluxes (equivalent to frequent momentum swaps between the |
124 |
slabs) can result in ``notched'', ``peaked'' and generally non-thermal |
125 |
momentum distributions in the two slabs, as well as non-linear thermal |
126 |
and velocity distributions along the direction of the imposed flux |
127 |
($z$). Tenney and Maginn obtained reasonable limits on imposed flux |
128 |
and self-adjusting metrics for retaining the usability of the method. |
129 |
|
130 |
In this paper, we develop and test a method for non-isotropic velocity |
131 |
scaling (NIVS-RNEMD) which retains the desirable features of RNEMD |
132 |
(conservation of linear momentum and total energy, compatibility with |
133 |
periodic boundary conditions) while establishing true thermal |
134 |
distributions in each of the two slabs. In the next section, we |
135 |
present the method for determining the scaling constraints. We then |
136 |
test the method on both single component, multi-component, and |
137 |
non-isotropic mixtures and show that it is capable of providing |
138 |
reasonable estimates of the thermal conductivity and shear viscosity |
139 |
in these cases. |
140 |
|
141 |
\section{Methodology} |
142 |
We retain the basic idea of M\"{u}ller-Plathe's RNEMD method; the |
143 |
periodic system is partitioned into a series of thin slabs along one |
144 |
axis ($z$). One of the slabs at the end of the periodic box is |
145 |
designated the ``hot'' slab, while the slab in the center of the box |
146 |
is designated the ``cold'' slab. The artificial momentum flux will be |
147 |
established by transferring momentum from the cold slab and into the |
148 |
hot slab. |
149 |
|
150 |
Rather than using momentum swaps, we use a series of velocity scaling |
151 |
moves. For molecules $\{i\}$ located within the cold slab, |
152 |
\begin{equation} |
153 |
\vec{v}_i \leftarrow \left( \begin{array}{ccc} |
154 |
x & 0 & 0 \\ |
155 |
0 & y & 0 \\ |
156 |
0 & 0 & z \\ |
157 |
\end{array} \right) \cdot \vec{v}_i |
158 |
\end{equation} |
159 |
where ${x, y, z}$ are a set of 3 scaling variables for each of the |
160 |
three directions in the system. Likewise, the molecules $\{j\}$ |
161 |
located in the hot slab will see a concomitant scaling of velocities, |
162 |
\begin{equation} |
163 |
\vec{v}_j \leftarrow \left( \begin{array}{ccc} |
164 |
x^\prime & 0 & 0 \\ |
165 |
0 & y^\prime & 0 \\ |
166 |
0 & 0 & z^\prime \\ |
167 |
\end{array} \right) \cdot \vec{v}_j |
168 |
\end{equation} |
169 |
|
170 |
Conservation of linear momentum in each of the three directions |
171 |
($\alpha = x,y,z$) ties the values of the hot and cold scaling |
172 |
parameters together: |
173 |
\begin{equation} |
174 |
P_h^\alpha + P_c^\alpha = \alpha^\prime P_h^\alpha + \alpha P_c^\alpha |
175 |
\end{equation} |
176 |
where |
177 |
\begin{eqnarray} |
178 |
P_c^\alpha & = & \sum_{i = 1}^{N_c} m_i \left[\vec{v}_i\right]_\alpha \\ |
179 |
P_h^\alpha & = & \sum_{j = 1}^{N_h} m_j \left[\vec{v}_j\right]_\alpha |
180 |
\label{eq:momentumdef} |
181 |
\end{eqnarray} |
182 |
Therefore, for each of the three directions, the hot scaling |
183 |
parameters are a simple function of the cold scaling parameters and |
184 |
the instantaneous linear momentum in each of the two slabs. |
185 |
\begin{equation} |
186 |
\alpha^\prime = 1 + (1 - \alpha) p_\alpha |
187 |
\label{eq:hotcoldscaling} |
188 |
\end{equation} |
189 |
where |
190 |
\begin{equation} |
191 |
p_\alpha = \frac{P_c^\alpha}{P_h^\alpha} |
192 |
\end{equation} |
193 |
for convenience. |
194 |
|
195 |
Conservation of total energy also places constraints on the scaling: |
196 |
\begin{equation} |
197 |
\sum_{\alpha = x,y,z} K_h^\alpha + K_c^\alpha = \sum_{\alpha = x,y,z} |
198 |
\left(\alpha^\prime\right)^2 K_h^\alpha + \alpha^2 K_c^\alpha |
199 |
\end{equation} |
200 |
where the translational kinetic energies, $K_h^\alpha$ and |
201 |
$K_c^\alpha$, are computed for each of the three directions in a |
202 |
similar manner to the linear momenta (Eq. \ref{eq:momentumdef}). |
203 |
Substituting in the expressions for the hot scaling parameters |
204 |
($\alpha^\prime$) from Eq. (\ref{eq:hotcoldscaling}), we obtain the |
205 |
{\it constraint ellipsoid}: |
206 |
\begin{equation} |
207 |
\sum_{\alpha = x,y,z} a_\alpha \alpha^2 + b_\alpha \alpha + c_\alpha = 0 |
208 |
\label{eq:constraintEllipsoid} |
209 |
\end{equation} |
210 |
where the constants are obtained from the instantaneous values of the |
211 |
linear momenta and kinetic energies for the hot and cold slabs, |
212 |
\begin{eqnarray} |
213 |
a_\alpha & = &\left(K_c^\alpha + K_h^\alpha |
214 |
\left(p_\alpha\right)^2\right) \\ |
215 |
b_\alpha & = & -2 K_h^\alpha p_\alpha \left( 1 + p_\alpha\right) \\ |
216 |
c_\alpha & = & K_h^\alpha p_\alpha^2 + 2 K_h^\alpha p_\alpha - K_c^\alpha |
217 |
\label{eq:constraintEllipsoidConsts} |
218 |
\end{eqnarray} |
219 |
This ellipsoid (Eq. \ref{eq:constraintEllipsoid}) defines the set of |
220 |
cold slab scaling parameters which can be applied while preserving |
221 |
both linear momentum in all three directions as well as total kinetic |
222 |
energy. |
223 |
|
224 |
The goal of using velocity scaling variables is to transfer linear |
225 |
momentum or kinetic energy from the cold slab to the hot slab. If the |
226 |
hot and cold slabs are separated along the z-axis, the energy flux is |
227 |
given simply by the decrease in kinetic energy of the cold bin: |
228 |
\begin{equation} |
229 |
(1-x^2) K_c^x + (1-y^2) K_c^y + (1-z^2) K_c^z = J_z\Delta t |
230 |
\end{equation} |
231 |
The expression for the energy flux can be re-written as another |
232 |
ellipsoid centered on $(x,y,z) = 0$: |
233 |
\begin{equation} |
234 |
x^2 K_c^x + y^2 K_c^y + z^2 K_c^z = K_c^x + K_c^y + K_c^z - J_z\Delta t |
235 |
\label{eq:fluxEllipsoid} |
236 |
\end{equation} |
237 |
The spatial extent of the {\it thermal flux ellipsoid} is governed |
238 |
both by a targetted value, $J_z$ as well as the instantaneous values |
239 |
of the kinetic energy components in the cold bin. |
240 |
|
241 |
To satisfy an energetic flux as well as the conservation constraints, |
242 |
we must determine the points ${x,y,z}$ which lie on both the |
243 |
constraint ellipsoid (Eq. \ref{eq:constraintEllipsoid}) and the flux |
244 |
ellipsoid (Eq. \ref{eq:fluxEllipsoid}), i.e. the intersection of the |
245 |
two ellipsoids in 3-dimensional space. |
246 |
|
247 |
\begin{figure} |
248 |
\includegraphics[width=\linewidth]{ellipsoids} |
249 |
\caption{Illustration from the perspective of a space having cold |
250 |
slab scaling coefficients as its coordinates. Scaling points which |
251 |
maintain both constant energy and constant linear momentum of the |
252 |
system lie on the surface of the {\it constraint ellipsoid} while |
253 |
points which generate the target momentum flux lie on the surface of |
254 |
the {\it flux ellipsoid}. The velocity distributions in the cold bin |
255 |
are scaled by only those points which lie on both ellipsoids.} |
256 |
\label{ellipsoids} |
257 |
\end{figure} |
258 |
|
259 |
One may also define {\it momentum} flux (say along the $x$-direction) as: |
260 |
\begin{equation} |
261 |
(1-x) P_c^x = j_z(p_x)\Delta t |
262 |
\label{eq:fluxPlane} |
263 |
\end{equation} |
264 |
The above {\it momentum flux plane} is perpendicular to the $x$-axis, |
265 |
with its position governed both by a target value, $j_z(p_x)$ as well |
266 |
as the instantaneous value of the momentum along the $x$-direction. |
267 |
|
268 |
In order to satisfy a momentum flux as well as the conservation |
269 |
constraints, we must determine the points ${x,y,z}$ which lie on both |
270 |
the constraint ellipsoid (Eq. \ref{eq:constraintEllipsoid}) and the |
271 |
flux plane (Eq. \ref{eq:fluxPlane}), i.e. the intersection of an |
272 |
ellipsoid and a plane in 3-dimensional space. |
273 |
|
274 |
In both the momentum and energy flux scenarios, valid scaling |
275 |
parameters are arrived at by solving geometric intersection problems |
276 |
in $x, y, z$ space in order to obtain cold slab scaling parameters. |
277 |
Once the scaling variables for the cold slab are known, the hot slab |
278 |
scaling has also been determined. |
279 |
|
280 |
|
281 |
The following problem will be choosing an optimal set of scaling |
282 |
variables among the possible sets. Although this method is inherently |
283 |
non-isotropic, the goal is still to maintain the system as isotropic |
284 |
as possible. Under this consideration, one would like the kinetic |
285 |
energies in different directions could become as close as each other |
286 |
after each scaling. Simultaneously, one would also like each scaling |
287 |
as gentle as possible, i.e. ${x,y,z \rightarrow 1}$, in order to avoid |
288 |
large perturbation to the system. Therefore, one approach to obtain |
289 |
the scaling variables would be constructing an criteria function, with |
290 |
constraints as above equation sets, and solving the function's minimum |
291 |
by method like Lagrange multipliers. |
292 |
|
293 |
In order to save computation time, we have a different approach to a |
294 |
relatively good set of scaling variables with much less calculation |
295 |
than above. Here is the detail of our simplification of the problem. |
296 |
|
297 |
In the case of kinetic energy transfer, we impose another constraint |
298 |
${x = y}$, into the equation sets. Consequently, there are two |
299 |
variables left. And now one only needs to solve a set of two {\it |
300 |
ellipses equations}. This problem would be transformed into solving |
301 |
one quartic equation for one of the two variables. There are known |
302 |
generic methods that solve real roots of quartic equations. Then one |
303 |
can determine the other variable and obtain sets of scaling |
304 |
variables. Among these sets, one can apply the above criteria to |
305 |
choose the best set, while much faster with only a few sets to choose. |
306 |
|
307 |
In the case of momentum flux transfer, we impose another constraint to |
308 |
set the kinetic energy transfer as zero. In another word, we apply |
309 |
Eq. \ref{eq:fluxEllipsoid} and let ${J_z = 0}$. After that, with one |
310 |
variable fixed by Eq. \ref{eq:fluxPlane}, one now have a similar set |
311 |
of equations on the above kinetic energy transfer problem. Therefore, |
312 |
an approach similar to the above would be sufficient for this as well. |
313 |
|
314 |
\section{Computational Details} |
315 |
\subsection{Lennard-Jones Fluid} |
316 |
Our simulation consists of a series of systems. All of these |
317 |
simulations were run with the OpenMD simulation software |
318 |
package\cite{Meineke:2005gd} integrated with RNEMD codes. |
319 |
|
320 |
A Lennard-Jones fluid system was built and tested first. In order to |
321 |
compare our method with swapping RNEMD, a series of simulations were |
322 |
performed to calculate the shear viscosity and thermal conductivity of |
323 |
argon. 2592 atoms were in a orthorhombic cell, which was ${10.06\sigma |
324 |
\times 10.06\sigma \times 30.18\sigma}$ by size. The reduced density |
325 |
${\rho^* = \rho\sigma^3}$ was thus 0.85, which enabled direct |
326 |
comparison between our results and others. These simulations used |
327 |
velocity Verlet algorithm with reduced timestep ${\tau^* = |
328 |
4.6\times10^{-4}}$. |
329 |
|
330 |
For shear viscosity calculation, the reduced temperature was ${T^* = |
331 |
k_B T/\varepsilon = 0.72}$. Simulations were first equilibrated in canonical |
332 |
ensemble (NVT), then equilibrated in microcanonical ensemble |
333 |
(NVE). Establishing and stablizing momentum gradient were followed |
334 |
also in NVE ensemble. For the swapping part, Muller-Plathe's algorithm was |
335 |
adopted.\cite{ISI:000080382700030} The simulation box was under |
336 |
periodic boundary condition, and devided into ${N = 20}$ slabs. In each swap, |
337 |
the top slab ${(n = 1)}$ exchange the most negative $x$ momentum with the |
338 |
most positive $x$ momentum in the center slab ${(n = N/2 + 1)}$. Referred |
339 |
to Tenney {\it et al.}\cite{Maginn:2010}, a series of swapping |
340 |
frequency were chosen. According to each result from swapping |
341 |
RNEMD, scaling RNEMD simulations were run with the target momentum |
342 |
flux set to produce a similar momentum flux, and consequently shear |
343 |
rate. Furthermore, various scaling frequencies can be tested for one |
344 |
single swapping rate. To test the temperature homogeneity in our |
345 |
system of swapping and scaling methods, temperatures of different |
346 |
dimensions in all the slabs were observed. Most of the simulations |
347 |
include $10^5$ steps of equilibration without imposing momentum flux, |
348 |
$10^5$ steps of stablization with imposing unphysical momentum |
349 |
transfer, and $10^6$ steps of data collection under RNEMD. For |
350 |
relatively high momentum flux simulations, ${5\times10^5}$ step data |
351 |
collection is sufficient. For some low momentum flux simulations, |
352 |
${2\times10^6}$ steps were necessary. |
353 |
|
354 |
After each simulation, the shear viscosity was calculated in reduced |
355 |
unit. The momentum flux was calculated with total unphysical |
356 |
transferred momentum ${P_x}$ and data collection time $t$: |
357 |
\begin{equation} |
358 |
j_z(p_x) = \frac{P_x}{2 t L_x L_y} |
359 |
\end{equation} |
360 |
where $L_x$ and $L_y$ denotes $x$ and $y$ lengths of the simulation |
361 |
box, and physical momentum transfer occurs in two ways due to our |
362 |
periodic boundary condition settings. And the velocity gradient |
363 |
${\langle \partial v_x /\partial z \rangle}$ can be obtained by a |
364 |
linear regression of the velocity profile. From the shear viscosity |
365 |
$\eta$ calculated with the above parameters, one can further convert |
366 |
it into reduced unit ${\eta^* = \eta \sigma^2 (\varepsilon m)^{-1/2}}$. |
367 |
|
368 |
For thermal conductivity calculations, simulations were first run under |
369 |
reduced temperature ${\langle T^*\rangle = 0.72}$ in NVE |
370 |
ensemble. Muller-Plathe's algorithm was adopted in the swapping |
371 |
method. Under identical simulation box parameters with our shear |
372 |
viscosity calculations, in each swap, the top slab exchanges all three |
373 |
translational momentum components of the molecule with least kinetic |
374 |
energy with the same components of the molecule in the center slab |
375 |
with most kinetic energy, unless this ``coldest'' molecule in the |
376 |
``hot'' slab is still ``hotter'' than the ``hottest'' molecule in the |
377 |
``cold'' slab. According to swapping RNEMD results, target energy flux |
378 |
for scaling RNEMD simulations can be set. Also, various scaling |
379 |
frequencies can be tested for one target energy flux. To compare the |
380 |
performance between swapping and scaling method, distributions of |
381 |
velocity and speed in different slabs were observed. |
382 |
|
383 |
For each swapping rate, thermal conductivity was calculated in reduced |
384 |
unit. The energy flux was calculated similarly to the momentum flux, |
385 |
with total unphysical transferred energy ${E_{total}}$ and data collection |
386 |
time $t$: |
387 |
\begin{equation} |
388 |
J_z = \frac{E_{total}}{2 t L_x L_y} |
389 |
\end{equation} |
390 |
And the temperature gradient ${\langle\partial T/\partial z\rangle}$ |
391 |
can be obtained by a linear regression of the temperature |
392 |
profile. From the thermal conductivity $\lambda$ calculated, one can |
393 |
further convert it into reduced unit ${\lambda^*=\lambda \sigma^2 |
394 |
m^{1/2} k_B^{-1}\varepsilon^{-1/2}}$. |
395 |
|
396 |
\subsection{ Water / Metal Thermal Conductivity} |
397 |
Another series of our simulation is the calculation of interfacial |
398 |
thermal conductivity of a Au/H$_2$O system. Respective calculations of |
399 |
liquid water (Extended Simple Point Charge model) and crystal gold |
400 |
thermal conductivity were performed and compared with current results |
401 |
to ensure the validity of NIVS-RNEMD. After that, a mixture system was |
402 |
simulated. |
403 |
|
404 |
For thermal conductivity calculation of bulk water, a simulation box |
405 |
consisting of 1000 molecules were first equilibrated under ambient |
406 |
pressure and temperature conditions using NPT ensemble, followed by |
407 |
equilibration in fixed volume (NVT). The system was then equilibrated in |
408 |
microcanonical ensemble (NVE). Also in NVE ensemble, establishing a |
409 |
stable thermal gradient was followed. The simulation box was under |
410 |
periodic boundary condition and devided into 10 slabs. Data collection |
411 |
process was similar to Lennard-Jones fluid system. |
412 |
|
413 |
Thermal conductivity calculation of bulk crystal gold used a similar |
414 |
protocol. Two types of force field parameters, Embedded Atom Method |
415 |
(EAM) and Quantum Sutten-Chen (QSC) force field were used |
416 |
respectively. The face-centered cubic crystal simulation box consists of |
417 |
2880 Au atoms. The lattice was first allowed volume change to relax |
418 |
under ambient temperature and pressure. Equilibrations in canonical and |
419 |
microcanonical ensemble were followed in order. With the simulation |
420 |
lattice devided evenly into 10 slabs, different thermal gradients were |
421 |
established by applying a set of target thermal transfer flux. Data of |
422 |
the series of thermal gradients was collected for calculation. |
423 |
|
424 |
After simulations of bulk water and crystal gold, a mixture system was |
425 |
constructed, consisting of 1188 Au atoms and 1862 H$_2$O |
426 |
molecules. Spohr potential was adopted in depicting the interaction |
427 |
between metal atom and water molecule.\cite{ISI:000167766600035} A |
428 |
similar protocol of equilibration was followed. Several thermal |
429 |
gradients was built under different target thermal flux. It was found |
430 |
out that compared to our previous simulation systems, the two phases |
431 |
could have large temperature difference even under a relatively low |
432 |
thermal flux. Therefore, under our low flux conditions, it is assumed |
433 |
that the metal and water phases have respectively homogeneous |
434 |
temperature, excluding the surface regions. In calculating the |
435 |
interfacial thermal conductivity $G$, this assumptioin was applied and |
436 |
thus our formula becomes: |
437 |
|
438 |
\begin{equation} |
439 |
G = \frac{E_{total}}{2 t L_x L_y \left( \langle T_{gold}\rangle - |
440 |
\langle T_{water}\rangle \right)} |
441 |
\label{interfaceCalc} |
442 |
\end{equation} |
443 |
where ${E_{total}}$ is the imposed unphysical kinetic energy transfer |
444 |
and ${\langle T_{gold}\rangle}$ and ${\langle T_{water}\rangle}$ are the |
445 |
average observed temperature of gold and water phases respectively. |
446 |
|
447 |
\section{Results And Discussions} |
448 |
\subsection{Thermal Conductivity} |
449 |
\subsubsection{Lennard-Jones Fluid} |
450 |
Our thermal conductivity calculations show that scaling method results |
451 |
agree with swapping method. Four different exchange intervals were |
452 |
tested (Table \ref{thermalLJRes}) using swapping method. With a fixed |
453 |
10fs exchange interval, target exchange kinetic energy was set to |
454 |
produce equivalent kinetic energy flux as in swapping method. And |
455 |
similar thermal gradients were observed with similar thermal flux in |
456 |
two simulation methods (Figure \ref{thermalGrad}). |
457 |
|
458 |
\begin{table*} |
459 |
\begin{minipage}{\linewidth} |
460 |
\begin{center} |
461 |
|
462 |
\caption{Calculation results for thermal conductivity of Lennard-Jones |
463 |
fluid at ${\langle T^* \rangle = 0.72}$ and ${\rho^* = 0.85}$, with |
464 |
swap and scale methods at various kinetic energy exchange rates. Results |
465 |
in reduced unit. Errors of calculations in parentheses.} |
466 |
|
467 |
\begin{tabular}{ccc} |
468 |
\hline |
469 |
(Equilvalent) Exchange Interval (fs) & $\lambda^*_{swap}$ & |
470 |
$\lambda^*_{scale}$\\ |
471 |
\hline |
472 |
250 & 7.03(0.34) & 7.30(0.10)\\ |
473 |
500 & 7.03(0.14) & 6.95(0.09)\\ |
474 |
1000 & 6.91(0.42) & 7.19(0.07)\\ |
475 |
2000 & 7.52(0.15) & 7.19(0.28)\\ |
476 |
\hline |
477 |
\end{tabular} |
478 |
\label{thermalLJRes} |
479 |
\end{center} |
480 |
\end{minipage} |
481 |
\end{table*} |
482 |
|
483 |
\begin{figure} |
484 |
\includegraphics[width=\linewidth]{thermalGrad} |
485 |
\caption{NIVS-RNEMD method introduced similar temperature gradients |
486 |
compared to ``swapping'' method under various kinetic energy flux in |
487 |
thermal conductivity simulations.} |
488 |
\label{thermalGrad} |
489 |
\end{figure} |
490 |
|
491 |
During these simulations, molecule velocities were recorded in 1000 of |
492 |
all the snapshots of one single data collection process. These |
493 |
velocity data were used to produce histograms of velocity and speed |
494 |
distribution in different slabs. From these histograms, it is observed |
495 |
that under relatively high unphysical kinetic energy flux, speed and |
496 |
velocity distribution of molecules in slabs where swapping occured |
497 |
could deviate from Maxwell-Boltzmann distribution. Figure |
498 |
\ref{thermalHist} a) illustrates how these distributions deviate from an |
499 |
ideal distribution. In high temperature slab, probability density in |
500 |
low speed is confidently smaller than ideal curve fit; in low |
501 |
temperature slab, probability density in high speed is smaller than |
502 |
ideal, while larger than ideal in low speed. This phenomenon is more |
503 |
obvious in our high swapping rate simulations. And this deviation |
504 |
could also leads to deviation of distribution of velocity in various |
505 |
dimensions. One feature of these deviated distribution is that in high |
506 |
temperature slab, the ideal Gaussian peak was changed into a |
507 |
relatively flat plateau; while in low temperature slab, that peak |
508 |
appears sharper. This problem is rooted in the mechanism of the |
509 |
swapping method. Continually depleting low (high) speed particles in |
510 |
the high (low) temperature slab could not be complemented by |
511 |
diffusions of low (high) speed particles from neighbor slabs, unless |
512 |
in suffciently low swapping rate. Simutaneously, surplus low speed |
513 |
particles in the low temperature slab do not have sufficient time to |
514 |
diffuse to neighbor slabs. However, thermal exchange rate should reach |
515 |
a minimum level to produce an observable thermal gradient under noise |
516 |
interference. Consequently, swapping RNEMD has a relatively narrow |
517 |
choice of swapping rate to satisfy these above restrictions. |
518 |
|
519 |
Comparatively, NIVS-RNEMD has a speed distribution closer to the ideal |
520 |
curve fit (Figure \ref{thermalHist} b). Essentially, after scaling, a |
521 |
Gaussian distribution function would remain Gaussian. Although a |
522 |
single scaling is non-isotropic in all three dimensions, our scaling |
523 |
coefficient criteria could help maintian the scaling region as |
524 |
isotropic as possible. On the other hand, scaling coefficients are |
525 |
preferred to be as close to 1 as possible, which also helps minimize |
526 |
the difference among different dimensions. This is possible if scaling |
527 |
interval and one-time thermal transfer energy are well |
528 |
chosen. Consequently, NIVS-RNEMD is able to impose an unphysical |
529 |
thermal flux as the previous RNEMD method without large perturbation |
530 |
to the distribution of velocity and speed in the exchange regions. |
531 |
|
532 |
\begin{figure} |
533 |
\includegraphics[width=\linewidth]{thermalHist} |
534 |
\caption{Speed distribution for thermal conductivity using a) |
535 |
``swapping'' and b) NIVS- RNEMD methods. Shown is from the |
536 |
simulations with an exchange or equilvalent exchange interval of 250 |
537 |
fs. In circled areas, distributions from ``swapping'' RNEMD |
538 |
simulation have deviation from ideal Maxwell-Boltzmann distribution |
539 |
(curves fit for each distribution).} |
540 |
\label{thermalHist} |
541 |
\end{figure} |
542 |
|
543 |
\subsubsection{SPC/E Water} |
544 |
Our results of SPC/E water thermal conductivity are comparable to |
545 |
Bedrov {\it et al.}\cite{Bedrov:2000}, which employed the |
546 |
previous swapping RNEMD method for their calculation. Bedrov {\it et |
547 |
al.}\cite{Bedrov:2000} argued that exchange of the molecule |
548 |
center-of-mass velocities instead of single atom velocities in a |
549 |
molecule conserves the total kinetic energy and linear momentum. This |
550 |
principle is adopted in our simulations. Scaling is applied to the |
551 |
velocities of the rigid bodies of SPC/E model water molecules, instead |
552 |
of each hydrogen and oxygen atoms in relevant water molecules. As |
553 |
shown in Figure \ref{spceGrad}, temperature gradients were established |
554 |
similar to their system. However, the average temperature of our |
555 |
system is 300K, while theirs is 318K, which would be attributed for |
556 |
part of the difference between the final calculation results (Table |
557 |
\ref{spceThermal}). Both methods yields values in agreement with |
558 |
experiment. And this shows the applicability of our method to |
559 |
multi-atom molecular system. |
560 |
|
561 |
\begin{figure} |
562 |
\includegraphics[width=\linewidth]{spceGrad} |
563 |
\caption{Temperature gradients in SPC/E water thermal conductivity |
564 |
simulations.} |
565 |
\label{spceGrad} |
566 |
\end{figure} |
567 |
|
568 |
\begin{table*} |
569 |
\begin{minipage}{\linewidth} |
570 |
\begin{center} |
571 |
|
572 |
\caption{Calculation results for thermal conductivity of SPC/E water |
573 |
at ${\langle T\rangle}$ = 300K at various thermal exchange rates. Errors of |
574 |
calculations in parentheses. } |
575 |
|
576 |
\begin{tabular}{cccc} |
577 |
\hline |
578 |
$\langle dT/dz\rangle$(K/\AA) & & $\lambda$(W/m/K) & \\ |
579 |
& This work & Previous simulations\cite{Bedrov:2000} & |
580 |
Experiment$^a$\\ |
581 |
\hline |
582 |
0.38 & 0.816(0.044) & & 0.64\\ |
583 |
0.81 & 0.770(0.008) & 0.784\\ |
584 |
1.54 & 0.813(0.007) & 0.730\\ |
585 |
\hline |
586 |
\end{tabular} |
587 |
\label{spceThermal} |
588 |
\end{center} |
589 |
\end{minipage} |
590 |
\end{table*} |
591 |
|
592 |
\subsubsection{Crystal Gold} |
593 |
Our results of gold thermal conductivity using two force fields are |
594 |
shown in Table \ref{AuThermal}. In these calculations,the end and |
595 |
middle slabs were excluded in thermal gradient regession and only used |
596 |
as heat source and drain in the systems. Our yielded values using EAM |
597 |
force field are slightly larger than those using QSC force |
598 |
field. However, both series are significantly smaller than |
599 |
experimental value by a factor of more than 200. It has been verified |
600 |
that this difference is mainly attributed to the lack of electron |
601 |
interaction representation in these force field parameters. Richardson |
602 |
{\it et al.}\cite{Clancy:1992} used EAM force field parameters in |
603 |
their metal thermal conductivity calculations. The Non-Equilibrium MD |
604 |
method they employed in their simulations produced comparable results |
605 |
to ours. As Zhang {\it et al.}\cite{ISI:000231042800044} stated, |
606 |
thermal conductivity values are influenced mainly by force |
607 |
field. Another factor that affects the calculation results could be |
608 |
the density of the metal. After equilibration under |
609 |
isobaric-isothermal conditions, our crystall simulation cell expanded |
610 |
by the order of 1\%. Under longer lattice constant than default, |
611 |
lower thermal conductance would be expected. Therefore, it is still |
612 |
confident to conclude that NIVS-RNEMD is applicable to metal force |
613 |
field system. |
614 |
|
615 |
\begin{table*} |
616 |
\begin{minipage}{\linewidth} |
617 |
\begin{center} |
618 |
|
619 |
\caption{Calculation results for thermal conductivity of crystal gold |
620 |
using different force fields at ${\langle T\rangle}$ = 300K at |
621 |
various thermal exchange rates. Errors of calculations in parentheses.} |
622 |
|
623 |
\begin{tabular}{ccc} |
624 |
\hline |
625 |
Force Field Used & $\langle dT/dz\rangle$(K/\AA) & $\lambda$(W/m/K)\\ |
626 |
\hline |
627 |
& 1.44 & 1.10(0.01)\\ |
628 |
QSC & 2.86 & 1.08(0.02)\\ |
629 |
& 5.14 & 1.15(0.01)\\ |
630 |
\hline |
631 |
& 1.24 & 1.24(0.06)\\ |
632 |
EAM & 2.06 & 1.37(0.04)\\ |
633 |
& 2.55 & 1.41(0.03)\\ |
634 |
\hline |
635 |
\end{tabular} |
636 |
\label{AuThermal} |
637 |
\end{center} |
638 |
\end{minipage} |
639 |
\end{table*} |
640 |
|
641 |
|
642 |
\subsection{Interfaciel Thermal Conductivity} |
643 |
After simulations of homogeneous water and gold systems using |
644 |
NIVS-RNEMD method were proved valid, calculation of gold/water |
645 |
interfacial thermal conductivity was followed. It is found out that |
646 |
the low interfacial conductance is probably due to the hydrophobic |
647 |
surface in our system. Figure \ref{interface} (a) demonstrates mass |
648 |
density change along $z$-axis, which is perpendicular to the |
649 |
gold/water interface. It is observed that water density significantly |
650 |
decreases when approaching the surface. Under this low thermal |
651 |
conductance, both gold and water phase have sufficient time to |
652 |
eliminate temperature difference inside respectively (Figure |
653 |
\ref{interface} b). With indistinguishable temperature difference |
654 |
within respective phase, it is valid to assume that the temperature |
655 |
difference between gold and water on surface would be approximately |
656 |
the same as the difference between the gold and water phase. This |
657 |
assumption enables convenient calculation of $G$ using |
658 |
Eq. \ref{interfaceCalc} instead of measuring temperatures of thin |
659 |
layer of water and gold close enough to surface, which would have |
660 |
greater fluctuation and lower accuracy. Reported results (Table |
661 |
\ref{interfaceRes}) are of two orders of magnitude smaller than our |
662 |
calculations on homogeneous systems, and thus have larger relative |
663 |
errors than our calculation results on homogeneous systems. |
664 |
|
665 |
\begin{figure} |
666 |
\includegraphics[width=\linewidth]{interface} |
667 |
\caption{Simulation results for Gold/Water interfacial thermal |
668 |
conductivity: (a) Significant water density decrease is observed on |
669 |
crystalline gold surface, which indicates low surface contact and |
670 |
leads to low thermal conductance. (b) Temperature profiles for a |
671 |
series of simulations. Temperatures of different slabs in the same |
672 |
phase show no significant differences.} |
673 |
\label{interface} |
674 |
\end{figure} |
675 |
|
676 |
\begin{table*} |
677 |
\begin{minipage}{\linewidth} |
678 |
\begin{center} |
679 |
|
680 |
\caption{Calculation results for interfacial thermal conductivity |
681 |
at ${\langle T\rangle \sim}$ 300K at various thermal exchange |
682 |
rates. Errors of calculations in parentheses. } |
683 |
|
684 |
\begin{tabular}{cccc} |
685 |
\hline |
686 |
$J_z$(MW/m$^2$) & $T_{gold}$ & $T_{water}$ & $G$(MW/m$^2$/K)\\ |
687 |
\hline |
688 |
98.0 & 355.2 & 295.8 & 1.65(0.21) \\ |
689 |
78.8 & 343.8 & 298.0 & 1.72(0.32) \\ |
690 |
73.6 & 344.3 & 298.0 & 1.59(0.24) \\ |
691 |
49.2 & 330.1 & 300.4 & 1.65(0.35) \\ |
692 |
\hline |
693 |
\end{tabular} |
694 |
\label{interfaceRes} |
695 |
\end{center} |
696 |
\end{minipage} |
697 |
\end{table*} |
698 |
|
699 |
\subsection{Shear Viscosity} |
700 |
Our calculations (Table \ref{shearRate}) shows that scale RNEMD method |
701 |
produced comparable shear viscosity to swap RNEMD method. In Table |
702 |
\ref{shearRate}, the names of the calculated samples are devided into |
703 |
two parts. The first number refers to total slabs in one simulation |
704 |
box. The second number refers to the swapping interval in swap method, or |
705 |
in scale method the equilvalent swapping interval that the same |
706 |
momentum flux would theoretically result in swap method. All the scale |
707 |
method results were from simulations that had a scaling interval of 10 |
708 |
time steps. The average molecular momentum gradients of these samples |
709 |
are shown in Figure \ref{shear} (a) and (b). |
710 |
|
711 |
\begin{table*} |
712 |
\begin{minipage}{\linewidth} |
713 |
\begin{center} |
714 |
|
715 |
\caption{Calculation results for shear viscosity of Lennard-Jones |
716 |
fluid at ${T^* = 0.72}$ and ${\rho^* = 0.85}$, with swap and scale |
717 |
methods at various momentum exchange rates. Results in reduced |
718 |
unit. Errors of calculations in parentheses. } |
719 |
|
720 |
\begin{tabular}{ccc} |
721 |
\hline |
722 |
(Equilvalent) Exchange Interval (fs) & $\eta^*_{swap}$ & |
723 |
$\eta^*_{scale}$\\ |
724 |
\hline |
725 |
500 & 3.64(0.05) & 3.76(0.09)\\ |
726 |
1000 & 3.52(0.16) & 3.66(0.06)\\ |
727 |
2000 & 3.72(0.05) & 3.32(0.18)\\ |
728 |
2500 & 3.42(0.06) & 3.43(0.08)\\ |
729 |
\hline |
730 |
\end{tabular} |
731 |
\label{shearRate} |
732 |
\end{center} |
733 |
\end{minipage} |
734 |
\end{table*} |
735 |
|
736 |
\begin{figure} |
737 |
\includegraphics[width=\linewidth]{shear} |
738 |
\caption{Average momentum gradients in shear viscosity simulations, |
739 |
using (a) ``swapping'' method and (b) NIVS-RNEMD method |
740 |
respectively. (c) Temperature difference among x and y, z dimensions |
741 |
observed when using NIVS-RNEMD with equivalent exchange interval of |
742 |
500 fs.} |
743 |
\label{shear} |
744 |
\end{figure} |
745 |
|
746 |
However, observations of temperatures along three dimensions show that |
747 |
inhomogeneity occurs in scaling RNEMD simulations, particularly in the |
748 |
two slabs which were scaled. Figure \ref{shear} (c) indicate that with |
749 |
relatively large imposed momentum flux, the temperature difference among $x$ |
750 |
and the other two dimensions was significant. This would result from the |
751 |
algorithm of scaling method. From Eq. \ref{eq:fluxPlane}, after |
752 |
momentum gradient is set up, $P_c^x$ would be roughly stable |
753 |
($<0$). Consequently, scaling factor $x$ would most probably larger |
754 |
than 1. Therefore, the kinetic energy in $x$-dimension $K_c^x$ would |
755 |
keep increase after most scaling steps. And if there is not enough time |
756 |
for the kinetic energy to exchange among different dimensions and |
757 |
different slabs, the system would finally build up temperature |
758 |
(kinetic energy) difference among the three dimensions. Also, between |
759 |
$y$ and $z$ dimensions in the scaled slabs, temperatures of $z$-axis |
760 |
are closer to neighbor slabs. This is due to momentum transfer along |
761 |
$z$ dimension between slabs. |
762 |
|
763 |
Although results between scaling and swapping methods are comparable, |
764 |
the inherent temperature inhomogeneity even in relatively low imposed |
765 |
exchange momentum flux simulations makes scaling RNEMD method less |
766 |
attractive than swapping RNEMD in shear viscosity calculation. |
767 |
|
768 |
\section{Conclusions} |
769 |
NIVS-RNEMD simulation method is developed and tested on various |
770 |
systems. Simulation results demonstrate its validity in thermal |
771 |
conductivity calculations, from Lennard-Jones fluid to multi-atom |
772 |
molecule like water and metal crystals. NIVS-RNEMD improves |
773 |
non-Boltzmann-Maxwell distributions, which exist in previous RNEMD |
774 |
methods. Furthermore, it develops a valid means for unphysical thermal |
775 |
transfer between different species of molecules, and thus extends its |
776 |
applicability to interfacial systems. Our calculation of gold/water |
777 |
interfacial thermal conductivity demonstrates this advantage over |
778 |
previous RNEMD methods. NIVS-RNEMD has also limited application on |
779 |
shear viscosity calculations, but could cause temperature difference |
780 |
among different dimensions under high momentum flux. Modification is |
781 |
necessary to extend the applicability of NIVS-RNEMD in shear viscosity |
782 |
calculations. |
783 |
|
784 |
\section{Acknowledgments} |
785 |
Support for this project was provided by the National Science |
786 |
Foundation under grant CHE-0848243. Computational time was provided by |
787 |
the Center for Research Computing (CRC) at the University of Notre |
788 |
Dame. \newpage |
789 |
|
790 |
\bibliographystyle{aip} |
791 |
\bibliography{nivsRnemd} |
792 |
|
793 |
\end{doublespace} |
794 |
\end{document} |
795 |
|