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 |
88 |
thermal conductivities and shear viscosities in a wide range of |
89 |
materials, from liquid copper to both monatomic and molecular fluids |
90 |
(e.g. ionic |
91 |
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} |
92 |
|
93 |
\begin{figure} |
94 |
\includegraphics[width=\linewidth]{thermalDemo} |
95 |
\caption{RNEMD methods impose an unphysical transfer of momentum or |
96 |
kinetic energy between a ``hot'' slab and a ``cold'' slab in the |
97 |
simulation box. The molecular system responds to this imposed flux |
98 |
by generating a momentum or temperature gradient. The slope of the |
99 |
gradient can then be used to compute transport properties (e.g. |
100 |
shear viscosity and thermal conductivity).} |
101 |
\label{thermalDemo} |
102 |
\end{figure} |
103 |
|
104 |
RNEMD is preferable in many ways to the forward NEMD |
105 |
methods\cite{ISI:A1988Q205300014,hess:209,Vasquez:2004fk,backer:154503,ISI:000266247600008} |
106 |
because it imposes what is typically difficult to measure (a flux or |
107 |
stress) and it is typically much easier to compute the response |
108 |
(momentum gradients or strains. For similar reasons, RNEMD is also |
109 |
preferable to slowly-converging equilibrium methods for measuring |
110 |
thermal conductivity and shear viscosity (using Green-Kubo |
111 |
relations\cite{daivis:541,mondello:9327} or the Helfand moment |
112 |
approach of Viscardy {\it et |
113 |
al.}\cite{Viscardy:2007bh,Viscardy:2007lq}) because these rely on |
114 |
computing difficult to measure quantities. |
115 |
|
116 |
Another attractive feature of RNEMD is that it conserves both total |
117 |
linear momentum and total energy during the swaps (as long as the two |
118 |
molecules have the same identity), so the swapped configurations are |
119 |
typically samples from the same manifold of states in the |
120 |
microcanonical ensemble. |
121 |
|
122 |
Recently, Tenney and Maginn\cite{Maginn:2010} have discovered |
123 |
some problems with the original RNEMD swap technique. Notably, large |
124 |
momentum fluxes (equivalent to frequent momentum swaps between the |
125 |
slabs) can result in ``notched'', ``peaked'' and generally non-thermal |
126 |
momentum distributions in the two slabs, as well as non-linear thermal |
127 |
and velocity distributions along the direction of the imposed flux |
128 |
($z$). Tenney and Maginn obtained reasonable limits on imposed flux |
129 |
and self-adjusting metrics for retaining the usability of the method. |
130 |
|
131 |
In this paper, we develop and test a method for non-isotropic velocity |
132 |
scaling (NIVS) which retains the desirable features of RNEMD |
133 |
(conservation of linear momentum and total energy, compatibility with |
134 |
periodic boundary conditions) while establishing true thermal |
135 |
distributions in each of the two slabs. In the next section, we |
136 |
present the method for determining the scaling constraints. We then |
137 |
test the method on both liquids and solids as well as a non-isotropic |
138 |
liquid-solid interface and show that it is capable of providing |
139 |
reasonable estimates of the thermal conductivity and shear viscosity |
140 |
in all of these cases. |
141 |
|
142 |
\section{Methodology} |
143 |
We retain the basic idea of M\"{u}ller-Plathe's RNEMD method; the |
144 |
periodic system is partitioned into a series of thin slabs along one |
145 |
axis ($z$). One of the slabs at the end of the periodic box is |
146 |
designated the ``hot'' slab, while the slab in the center of the box |
147 |
is designated the ``cold'' slab. The artificial momentum flux will be |
148 |
established by transferring momentum from the cold slab and into the |
149 |
hot slab. |
150 |
|
151 |
Rather than using momentum swaps, we use a series of velocity scaling |
152 |
moves. For molecules $\{i\}$ located within the cold slab, |
153 |
\begin{equation} |
154 |
\vec{v}_i \leftarrow \left( \begin{array}{ccc} |
155 |
x & 0 & 0 \\ |
156 |
0 & y & 0 \\ |
157 |
0 & 0 & z \\ |
158 |
\end{array} \right) \cdot \vec{v}_i |
159 |
\end{equation} |
160 |
where ${x, y, z}$ are a set of 3 velocity-scaling variables for each |
161 |
of the three directions in the system. Likewise, the molecules |
162 |
$\{j\}$ located in the hot slab will see a concomitant scaling of |
163 |
velocities, |
164 |
\begin{equation} |
165 |
\vec{v}_j \leftarrow \left( \begin{array}{ccc} |
166 |
x^\prime & 0 & 0 \\ |
167 |
0 & y^\prime & 0 \\ |
168 |
0 & 0 & z^\prime \\ |
169 |
\end{array} \right) \cdot \vec{v}_j |
170 |
\end{equation} |
171 |
|
172 |
Conservation of linear momentum in each of the three directions |
173 |
($\alpha = x,y,z$) ties the values of the hot and cold scaling |
174 |
parameters together: |
175 |
\begin{equation} |
176 |
P_h^\alpha + P_c^\alpha = \alpha^\prime P_h^\alpha + \alpha P_c^\alpha |
177 |
\end{equation} |
178 |
where |
179 |
\begin{eqnarray} |
180 |
P_c^\alpha & = & \sum_{i = 1}^{N_c} m_i \left[\vec{v}_i\right]_\alpha \\ |
181 |
P_h^\alpha & = & \sum_{j = 1}^{N_h} m_j \left[\vec{v}_j\right]_\alpha |
182 |
\label{eq:momentumdef} |
183 |
\end{eqnarray} |
184 |
Therefore, for each of the three directions, the hot scaling |
185 |
parameters are a simple function of the cold scaling parameters and |
186 |
the instantaneous linear momentum in each of the two slabs. |
187 |
\begin{equation} |
188 |
\alpha^\prime = 1 + (1 - \alpha) p_\alpha |
189 |
\label{eq:hotcoldscaling} |
190 |
\end{equation} |
191 |
where |
192 |
\begin{equation} |
193 |
p_\alpha = \frac{P_c^\alpha}{P_h^\alpha} |
194 |
\end{equation} |
195 |
for convenience. |
196 |
|
197 |
Conservation of total energy also places constraints on the scaling: |
198 |
\begin{equation} |
199 |
\sum_{\alpha = x,y,z} K_h^\alpha + K_c^\alpha = \sum_{\alpha = x,y,z} |
200 |
\left(\alpha^\prime\right)^2 K_h^\alpha + \alpha^2 K_c^\alpha |
201 |
\end{equation} |
202 |
where the translational kinetic energies, $K_h^\alpha$ and |
203 |
$K_c^\alpha$, are computed for each of the three directions in a |
204 |
similar manner to the linear momenta (Eq. \ref{eq:momentumdef}). |
205 |
Substituting in the expressions for the hot scaling parameters |
206 |
($\alpha^\prime$) from Eq. (\ref{eq:hotcoldscaling}), we obtain the |
207 |
{\it constraint ellipsoid}: |
208 |
\begin{equation} |
209 |
\sum_{\alpha = x,y,z} \left( a_\alpha \alpha^2 + b_\alpha \alpha + |
210 |
c_\alpha \right) = 0 |
211 |
\label{eq:constraintEllipsoid} |
212 |
\end{equation} |
213 |
where the constants are obtained from the instantaneous values of the |
214 |
linear momenta and kinetic energies for the hot and cold slabs, |
215 |
\begin{eqnarray} |
216 |
a_\alpha & = &\left(K_c^\alpha + K_h^\alpha |
217 |
\left(p_\alpha\right)^2\right) \\ |
218 |
b_\alpha & = & -2 K_h^\alpha p_\alpha \left( 1 + p_\alpha\right) \\ |
219 |
c_\alpha & = & K_h^\alpha p_\alpha^2 + 2 K_h^\alpha p_\alpha - K_c^\alpha |
220 |
\label{eq:constraintEllipsoidConsts} |
221 |
\end{eqnarray} |
222 |
This ellipsoid (Eq. \ref{eq:constraintEllipsoid}) defines the set of |
223 |
cold slab scaling parameters which, when applied, preserve the linear |
224 |
momentum of the system in all three directions as well as total |
225 |
kinetic energy. |
226 |
|
227 |
The goal of using these velocity scaling variables is to transfer |
228 |
linear momentum or kinetic energy from the cold slab to the hot slab. |
229 |
If the hot and cold slabs are separated along the z-axis, the energy |
230 |
flux is given simply by the decrease in kinetic energy of the cold |
231 |
bin: |
232 |
\begin{equation} |
233 |
(1-x^2) K_c^x + (1-y^2) K_c^y + (1-z^2) K_c^z = J_z\Delta t |
234 |
\end{equation} |
235 |
The expression for the energy flux can be re-written as another |
236 |
ellipsoid centered on $(x,y,z) = 0$: |
237 |
\begin{equation} |
238 |
\sum_{\alpha = x,y,z} K_c^\alpha \alpha^2 = \sum_{\alpha = x,y,z} |
239 |
K_c^\alpha -J_z \Delta t |
240 |
\label{eq:fluxEllipsoid} |
241 |
\end{equation} |
242 |
The spatial extent of the {\it thermal flux ellipsoid} is governed |
243 |
both by the target flux, $J_z$ as well as the instantaneous values of |
244 |
the kinetic energy components in the cold bin. |
245 |
|
246 |
To satisfy an energetic flux as well as the conservation constraints, |
247 |
we must determine the points ${x,y,z}$ that lie on both the constraint |
248 |
ellipsoid (Eq. \ref{eq:constraintEllipsoid}) and the flux ellipsoid |
249 |
(Eq. \ref{eq:fluxEllipsoid}), i.e. the intersection of the two |
250 |
ellipsoids in 3-dimensional space. |
251 |
|
252 |
\begin{figure} |
253 |
\includegraphics[width=\linewidth]{ellipsoids} |
254 |
\caption{Velocity scaling coefficients which maintain both constant |
255 |
energy and constant linear momentum of the system lie on the surface |
256 |
of the {\it constraint ellipsoid} while points which generate the |
257 |
target momentum flux lie on the surface of the {\it flux ellipsoid}. |
258 |
The velocity distributions in the cold bin are scaled by only those |
259 |
points which lie on both ellipsoids.} |
260 |
\label{ellipsoids} |
261 |
\end{figure} |
262 |
|
263 |
Since ellipsoids can be expressed as polynomials up to second order in |
264 |
each of the three coordinates, finding the the intersection points of |
265 |
two ellipsoids is isomorphic to finding the roots a polynomial of |
266 |
degree 16. There are a number of polynomial root-finding methods in |
267 |
the literature, [CITATIONS NEEDED] but numerically finding the roots |
268 |
of high-degree polynomials is generally an ill-conditioned |
269 |
problem.[CITATION NEEDED] One way around this is to try to maintain |
270 |
velocity scalings that are {\it as isotropic as possible}. To do |
271 |
this, we impose $x=y$, and to treat both the constraint and flux |
272 |
ellipsoids as 2-dimensional ellipses. In reduced dimensionality, the |
273 |
intersecting-ellipse problem reduces to finding the roots of |
274 |
polynomials of degree 4. |
275 |
|
276 |
Depending on the target flux and current velocity distributions, the |
277 |
ellipsoids can have between 0 and 4 intersection points. If there are |
278 |
no intersection points, it is not possible to satisfy the constraints |
279 |
while performing a non-equilibrium scaling move, and no change is made |
280 |
to the dynamics. |
281 |
|
282 |
With multiple intersection points, any of the scaling points will |
283 |
conserve the linear momentum and kinetic energy of the system and will |
284 |
generate the correct target flux. Although this method is inherently |
285 |
non-isotropic, the goal is still to maintain the system as close to an |
286 |
isotropic fluid as possible. With this in mind, we would like the |
287 |
kinetic energies in the three different directions could become as |
288 |
close as each other as possible after each scaling. Simultaneously, |
289 |
one would also like each scaling as gentle as possible, i.e. ${x,y,z |
290 |
\rightarrow 1}$, in order to avoid large perturbation to the system. |
291 |
To do this, we pick the intersection point which maintains the scaling |
292 |
variables ${x=y, z}$ as well as the ratio of kinetic energies |
293 |
${K_c^z/K_c^x, K_c^z/K_c^y}$ as close as possible to 1. |
294 |
|
295 |
After the valid scaling parameters are arrived at by solving geometric |
296 |
intersection problems in $x, y, z$ space in order to obtain cold slab |
297 |
scaling parameters, Eq. (\ref{eq:hotcoldscaling}) is used to |
298 |
determine the conjugate hot slab scaling variables. |
299 |
|
300 |
\subsection{Introducing shear stress via velocity scaling} |
301 |
Rather than using this method to induce a thermal flux, it is possible |
302 |
to use the random fluctuations of the average momentum in each of the |
303 |
bins to induce a momentum flux. Doing this repeatedly will create a |
304 |
shear stress on the system which will respond with an easily-measured |
305 |
strain. The momentum flux (say along the $x$-direction) may be |
306 |
defined as: |
307 |
\begin{equation} |
308 |
(1-x) P_c^x = j_z(p_x)\Delta t |
309 |
\label{eq:fluxPlane} |
310 |
\end{equation} |
311 |
This {\it momentum flux plane} is perpendicular to the $x$-axis, with |
312 |
its position governed both by a target value, $j_z(p_x)$ as well as |
313 |
the instantaneous value of the momentum along the $x$-direction. |
314 |
|
315 |
In order to satisfy a momentum flux as well as the conservation |
316 |
constraints, we must determine the points ${x,y,z}$ which lie on both |
317 |
the constraint ellipsoid (Eq. \ref{eq:constraintEllipsoid}) and the |
318 |
flux plane (Eq. \ref{eq:fluxPlane}), i.e. the intersection of an |
319 |
ellipsoid and a plane in 3-dimensional space. |
320 |
|
321 |
In the case of momentum flux transfer, we also impose another |
322 |
constraint to set the kinetic energy transfer as zero. In another |
323 |
word, we apply Eq. \ref{eq:fluxEllipsoid} and let ${J_z = 0}$. With |
324 |
one variable fixed by Eq. \ref{eq:fluxPlane}, one now have a similar |
325 |
set of quartic equations to the above kinetic energy transfer problem. |
326 |
|
327 |
\section{Computational Details} |
328 |
|
329 |
We have implemented this methodology in our molecular dynamics |
330 |
code,\cite{Meineke:2005gd} by performing the NIVS scaling moves after |
331 |
each MD step. We have tested it for a variety of different |
332 |
situations, including homogeneous fluids (Lennard-Jones and SPC/E |
333 |
water), crystalline solids (EAM and Sutton-Chen models for Gold), and |
334 |
heterogeneous interfaces (QSC gold - SPC/E water). The last of these |
335 |
systems would have been very difficult to study using previous RNEMD |
336 |
methods, but using velocity scaling moves, we can even obtain |
337 |
estimates of the interfacial thermal conductivity. |
338 |
|
339 |
\subsection{Lennard-Jones Fluid} |
340 |
|
341 |
2592 Lennard-Jones atoms were placed in an orthorhombic cell |
342 |
${10.06\sigma \times 10.06\sigma \times 30.18\sigma}$ on a side. The |
343 |
reduced density ${\rho^* = \rho\sigma^3}$ was thus 0.85, which enabled |
344 |
direct comparison between our results and previous methods. These |
345 |
simulations were carried out with a reduced timestep ${\tau^* = |
346 |
4.6\times10^{-4}}$. For the shear viscosity calculation, the mean |
347 |
temperature was ${T^* = k_B T/\varepsilon = 0.72}$. Simulations were |
348 |
first thermalized in canonical ensemble (NVT), then equilibrated in |
349 |
microcanonical ensemble (NVE) before introducing any non-equilibrium |
350 |
method. |
351 |
|
352 |
We have compared the momentum gradients established using our method |
353 |
to those obtained using the original M\"{u}ller-Plathe swapping |
354 |
algorithm.\cite{ISI:000080382700030} In both cases, the simulation box |
355 |
was divided into ${N = 20}$ slabs. In the swapping algorithm, the top |
356 |
slab $(n = 1)$ exchanges the most negative $x$ momentum with the most |
357 |
positive $x$ momentum in the center slab $(n = N/2 + 1)$. The rate at |
358 |
which the swapping moves are carried out defines the momentum or |
359 |
thermal flux between the two slabs. In their work, Tenney {\it et |
360 |
al.}\cite{Maginn:2010} found problematic behavior with large swap |
361 |
frequencies. |
362 |
|
363 |
According to each result from swapping RNEMD, scaling RNEMD |
364 |
simulations were run with the target momentum flux set to produce a |
365 |
similar momentum flux, and consequently shear rate. Furthermore, |
366 |
various scaling frequencies can be tested for one single swapping |
367 |
rate. To test the temperature homogeneity in our system of swapping |
368 |
and scaling methods, temperatures of different dimensions in all the |
369 |
slabs were observed. Most of the simulations include $10^5$ steps of |
370 |
equilibration without imposing momentum flux, $10^5$ steps of |
371 |
stablization with imposing unphysical momentum transfer, and $10^6$ |
372 |
steps of data collection under RNEMD. For relatively high momentum |
373 |
flux simulations, ${5\times10^5}$ step data collection is sufficient. |
374 |
For some low momentum flux simulations, ${2\times10^6}$ steps were |
375 |
necessary. |
376 |
|
377 |
After each simulation, the shear viscosity was calculated in reduced |
378 |
unit. The momentum flux was calculated with total unphysical |
379 |
transferred momentum ${P_x}$ and data collection time $t$: |
380 |
\begin{equation} |
381 |
j_z(p_x) = \frac{P_x}{2 t L_x L_y} |
382 |
\end{equation} |
383 |
where $L_x$ and $L_y$ denotes $x$ and $y$ lengths of the simulation |
384 |
box, and physical momentum transfer occurs in two ways due to our |
385 |
periodic boundary condition settings. And the velocity gradient |
386 |
${\langle \partial v_x /\partial z \rangle}$ can be obtained by a |
387 |
linear regression of the velocity profile. From the shear viscosity |
388 |
$\eta$ calculated with the above parameters, one can further convert |
389 |
it into reduced unit ${\eta^* = \eta \sigma^2 (\varepsilon m)^{-1/2}}$. |
390 |
|
391 |
For thermal conductivity calculations, simulations were first run under |
392 |
reduced temperature ${\langle T^*\rangle = 0.72}$ in NVE |
393 |
ensemble. Muller-Plathe's algorithm was adopted in the swapping |
394 |
method. Under identical simulation box parameters with our shear |
395 |
viscosity calculations, in each swap, the top slab exchanges all three |
396 |
translational momentum components of the molecule with least kinetic |
397 |
energy with the same components of the molecule in the center slab |
398 |
with most kinetic energy, unless this ``coldest'' molecule in the |
399 |
``hot'' slab is still ``hotter'' than the ``hottest'' molecule in the |
400 |
``cold'' slab. According to swapping RNEMD results, target energy flux |
401 |
for scaling RNEMD simulations can be set. Also, various scaling |
402 |
frequencies can be tested for one target energy flux. To compare the |
403 |
performance between swapping and scaling method, distributions of |
404 |
velocity and speed in different slabs were observed. |
405 |
|
406 |
For each swapping rate, thermal conductivity was calculated in reduced |
407 |
unit. The energy flux was calculated similarly to the momentum flux, |
408 |
with total unphysical transferred energy ${E_{total}}$ and data collection |
409 |
time $t$: |
410 |
\begin{equation} |
411 |
J_z = \frac{E_{total}}{2 t L_x L_y} |
412 |
\end{equation} |
413 |
And the temperature gradient ${\langle\partial T/\partial z\rangle}$ |
414 |
can be obtained by a linear regression of the temperature |
415 |
profile. From the thermal conductivity $\lambda$ calculated, one can |
416 |
further convert it into reduced unit ${\lambda^*=\lambda \sigma^2 |
417 |
m^{1/2} k_B^{-1}\varepsilon^{-1/2}}$. |
418 |
|
419 |
\subsection{ Water / Metal Thermal Conductivity} |
420 |
Another series of our simulation is the calculation of interfacial |
421 |
thermal conductivity of a Au/H$_2$O system. Respective calculations of |
422 |
liquid water (Extended Simple Point Charge model) and crystal gold |
423 |
thermal conductivity were performed and compared with current results |
424 |
to ensure the validity of NIVS-RNEMD. After that, a mixture system was |
425 |
simulated. |
426 |
|
427 |
For thermal conductivity calculation of bulk water, a simulation box |
428 |
consisting of 1000 molecules were first equilibrated under ambient |
429 |
pressure and temperature conditions using NPT ensemble, followed by |
430 |
equilibration in fixed volume (NVT). The system was then equilibrated in |
431 |
microcanonical ensemble (NVE). Also in NVE ensemble, establishing a |
432 |
stable thermal gradient was followed. The simulation box was under |
433 |
periodic boundary condition and devided into 10 slabs. Data collection |
434 |
process was similar to Lennard-Jones fluid system. |
435 |
|
436 |
Thermal conductivity calculation of bulk crystal gold used a similar |
437 |
protocol. Two types of force field parameters, Embedded Atom Method |
438 |
(EAM) and Quantum Sutten-Chen (QSC) force field were used |
439 |
respectively. The face-centered cubic crystal simulation box consists of |
440 |
2880 Au atoms. The lattice was first allowed volume change to relax |
441 |
under ambient temperature and pressure. Equilibrations in canonical and |
442 |
microcanonical ensemble were followed in order. With the simulation |
443 |
lattice devided evenly into 10 slabs, different thermal gradients were |
444 |
established by applying a set of target thermal transfer flux. Data of |
445 |
the series of thermal gradients was collected for calculation. |
446 |
|
447 |
After simulations of bulk water and crystal gold, a mixture system was |
448 |
constructed, consisting of 1188 Au atoms and 1862 H$_2$O |
449 |
molecules. Spohr potential was adopted in depicting the interaction |
450 |
between metal atom and water molecule.\cite{ISI:000167766600035} A |
451 |
similar protocol of equilibration was followed. Several thermal |
452 |
gradients was built under different target thermal flux. It was found |
453 |
out that compared to our previous simulation systems, the two phases |
454 |
could have large temperature difference even under a relatively low |
455 |
thermal flux. Therefore, under our low flux conditions, it is assumed |
456 |
that the metal and water phases have respectively homogeneous |
457 |
temperature, excluding the surface regions. In calculating the |
458 |
interfacial thermal conductivity $G$, this assumptioin was applied and |
459 |
thus our formula becomes: |
460 |
|
461 |
\begin{equation} |
462 |
G = \frac{E_{total}}{2 t L_x L_y \left( \langle T_{gold}\rangle - |
463 |
\langle T_{water}\rangle \right)} |
464 |
\label{interfaceCalc} |
465 |
\end{equation} |
466 |
where ${E_{total}}$ is the imposed unphysical kinetic energy transfer |
467 |
and ${\langle T_{gold}\rangle}$ and ${\langle T_{water}\rangle}$ are the |
468 |
average observed temperature of gold and water phases respectively. |
469 |
|
470 |
\section{Results And Discussions} |
471 |
\subsection{Thermal Conductivity} |
472 |
\subsubsection{Lennard-Jones Fluid} |
473 |
Our thermal conductivity calculations show that scaling method results |
474 |
agree with swapping method. Four different exchange intervals were |
475 |
tested (Table \ref{thermalLJRes}) using swapping method. With a fixed |
476 |
10fs exchange interval, target exchange kinetic energy was set to |
477 |
produce equivalent kinetic energy flux as in swapping method. And |
478 |
similar thermal gradients were observed with similar thermal flux in |
479 |
two simulation methods (Figure \ref{thermalGrad}). |
480 |
|
481 |
\begin{table*} |
482 |
\begin{minipage}{\linewidth} |
483 |
\begin{center} |
484 |
|
485 |
\caption{Calculation results for thermal conductivity of Lennard-Jones |
486 |
fluid at ${\langle T^* \rangle = 0.72}$ and ${\rho^* = 0.85}$, with |
487 |
swap and scale methods at various kinetic energy exchange rates. Results |
488 |
in reduced unit. Errors of calculations in parentheses.} |
489 |
|
490 |
\begin{tabular}{ccccc} |
491 |
Swapping method & & & NIVS-RNEMD\\ |
492 |
\hline |
493 |
Swap Interval (fs) & $\lambda^*_{swap}$ & & Equilvalent $J_z^*$ & |
494 |
$\lambda^*_{scale}$\\ |
495 |
\hline |
496 |
250 & 7.03(0.34) & & 0.16 & 7.30(0.10)\\ |
497 |
500 & 7.03(0.14) & & 0.09 & 6.95(0.09)\\ |
498 |
1000 & 6.91(0.42) & & 0.047 & 7.19(0.07)\\ |
499 |
2000 & 7.52(0.15) & & 0.024 & 7.19(0.28)\\ |
500 |
\hline |
501 |
\end{tabular} |
502 |
\label{thermalLJRes} |
503 |
\end{center} |
504 |
\end{minipage} |
505 |
\end{table*} |
506 |
|
507 |
\begin{figure} |
508 |
\includegraphics[width=\linewidth]{thermalGrad} |
509 |
\caption{NIVS-RNEMD method introduced similar temperature gradients |
510 |
compared to ``swapping'' method under various kinetic energy flux in |
511 |
thermal conductivity simulations.} |
512 |
\label{thermalGrad} |
513 |
\end{figure} |
514 |
|
515 |
During these simulations, molecule velocities were recorded in 1000 of |
516 |
all the snapshots of one single data collection process. These |
517 |
velocity data were used to produce histograms of velocity and speed |
518 |
distribution in different slabs. From these histograms, it is observed |
519 |
that under relatively high unphysical kinetic energy flux, speed and |
520 |
velocity distribution of molecules in slabs where swapping occured |
521 |
could deviate from Maxwell-Boltzmann distribution. Figure |
522 |
\ref{thermalHist} a) illustrates how these distributions deviate from an |
523 |
ideal distribution. In high temperature slab, probability density in |
524 |
low speed is confidently smaller than ideal curve fit; in low |
525 |
temperature slab, probability density in high speed is smaller than |
526 |
ideal, while larger than ideal in low speed. This phenomenon is more |
527 |
obvious in our high swapping rate simulations. And this deviation |
528 |
could also leads to deviation of distribution of velocity in various |
529 |
dimensions. One feature of these deviated distribution is that in high |
530 |
temperature slab, the ideal Gaussian peak was changed into a |
531 |
relatively flat plateau; while in low temperature slab, that peak |
532 |
appears sharper. This problem is rooted in the mechanism of the |
533 |
swapping method. Continually depleting low (high) speed particles in |
534 |
the high (low) temperature slab could not be complemented by |
535 |
diffusions of low (high) speed particles from neighbor slabs, unless |
536 |
in suffciently low swapping rate. Simutaneously, surplus low speed |
537 |
particles in the low temperature slab do not have sufficient time to |
538 |
diffuse to neighbor slabs. However, thermal exchange rate should reach |
539 |
a minimum level to produce an observable thermal gradient under noise |
540 |
interference. Consequently, swapping RNEMD has a relatively narrow |
541 |
choice of swapping rate to satisfy these above restrictions. |
542 |
|
543 |
Comparatively, NIVS-RNEMD has a speed distribution closer to the ideal |
544 |
curve fit (Figure \ref{thermalHist} b). Essentially, after scaling, a |
545 |
Gaussian distribution function would remain Gaussian. Although a |
546 |
single scaling is non-isotropic in all three dimensions, our scaling |
547 |
coefficient criteria could help maintian the scaling region as |
548 |
isotropic as possible. On the other hand, scaling coefficients are |
549 |
preferred to be as close to 1 as possible, which also helps minimize |
550 |
the difference among different dimensions. This is possible if scaling |
551 |
interval and one-time thermal transfer energy are well |
552 |
chosen. Consequently, NIVS-RNEMD is able to impose an unphysical |
553 |
thermal flux as the previous RNEMD method without large perturbation |
554 |
to the distribution of velocity and speed in the exchange regions. |
555 |
|
556 |
\begin{figure} |
557 |
\includegraphics[width=\linewidth]{thermalHist} |
558 |
\caption{Speed distribution for thermal conductivity using a) |
559 |
``swapping'' and b) NIVS- RNEMD methods. Shown is from the |
560 |
simulations with an exchange or equilvalent exchange interval of 250 |
561 |
fs. In circled areas, distributions from ``swapping'' RNEMD |
562 |
simulation have deviation from ideal Maxwell-Boltzmann distribution |
563 |
(curves fit for each distribution).} |
564 |
\label{thermalHist} |
565 |
\end{figure} |
566 |
|
567 |
\subsubsection{SPC/E Water} |
568 |
Our results of SPC/E water thermal conductivity are comparable to |
569 |
Bedrov {\it et al.}\cite{Bedrov:2000}, which employed the |
570 |
previous swapping RNEMD method for their calculation. Bedrov {\it et |
571 |
al.}\cite{Bedrov:2000} argued that exchange of the molecule |
572 |
center-of-mass velocities instead of single atom velocities in a |
573 |
molecule conserves the total kinetic energy and linear momentum. This |
574 |
principle is adopted in our simulations. Scaling is applied to the |
575 |
velocities of the rigid bodies of SPC/E model water molecules, instead |
576 |
of each hydrogen and oxygen atoms in relevant water molecules. As |
577 |
shown in Figure \ref{spceGrad}, temperature gradients were established |
578 |
similar to their system. However, the average temperature of our |
579 |
system is 300K, while theirs is 318K, which would be attributed for |
580 |
part of the difference between the final calculation results (Table |
581 |
\ref{spceThermal}). Both methods yields values in agreement with |
582 |
experiment. And this shows the applicability of our method to |
583 |
multi-atom molecular system. |
584 |
|
585 |
\begin{figure} |
586 |
\includegraphics[width=\linewidth]{spceGrad} |
587 |
\caption{Temperature gradients in SPC/E water thermal conductivity |
588 |
simulations.} |
589 |
\label{spceGrad} |
590 |
\end{figure} |
591 |
|
592 |
\begin{table*} |
593 |
\begin{minipage}{\linewidth} |
594 |
\begin{center} |
595 |
|
596 |
\caption{Calculation results for thermal conductivity of SPC/E water |
597 |
at ${\langle T\rangle}$ = 300K at various thermal exchange rates. Errors of |
598 |
calculations in parentheses. } |
599 |
|
600 |
\begin{tabular}{cccc} |
601 |
\hline |
602 |
$\langle dT/dz\rangle$(K/\AA) & & $\lambda$(W/m/K) & \\ |
603 |
& This work & Previous simulations\cite{Bedrov:2000} & |
604 |
Experiment[CITATION]\\ |
605 |
\hline |
606 |
0.38 & 0.816(0.044) & & 0.64\\ |
607 |
0.81 & 0.770(0.008) & 0.784\\ |
608 |
1.54 & 0.813(0.007) & 0.730\\ |
609 |
\hline |
610 |
\end{tabular} |
611 |
\label{spceThermal} |
612 |
\end{center} |
613 |
\end{minipage} |
614 |
\end{table*} |
615 |
|
616 |
\subsubsection{Crystal Gold} |
617 |
Our results of gold thermal conductivity using two force fields are |
618 |
shown in Table \ref{AuThermal}. In these calculations,the end and |
619 |
middle slabs were excluded in thermal gradient regession and only used |
620 |
as heat source and drain in the systems. Our yielded values using EAM |
621 |
force field are slightly larger than those using QSC force |
622 |
field. However, both series are significantly smaller than |
623 |
experimental value by a factor of more than 200. It has been verified |
624 |
that this difference is mainly attributed to the lack of electron |
625 |
interaction representation in these force field parameters. Richardson |
626 |
{\it et al.}\cite{Clancy:1992} used EAM force field parameters in |
627 |
their metal thermal conductivity calculations. The Non-Equilibrium MD |
628 |
method they employed in their simulations produced comparable results |
629 |
to ours. As Zhang {\it et al.}\cite{ISI:000231042800044} stated, |
630 |
thermal conductivity values are influenced mainly by force |
631 |
field. Another factor that affects the calculation results could be |
632 |
the density of the metal. After equilibration under |
633 |
isobaric-isothermal conditions, our crystall simulation cell expanded |
634 |
by the order of 1\%. Under larger lattice constant than default, |
635 |
lower thermal conductance were expected and observed. Furthermore, |
636 |
from our simulations, larger thermal gradients tend to produce larger |
637 |
results. Consequently, having a higher temperature gradient (3.6 |
638 |
K/\AA), Richardson {\it et al.} could obtain a larger |
639 |
result. Therefore, it is still confident to apply NIVS-RNEMD to metal |
640 |
force field systems. |
641 |
|
642 |
\begin{table*} |
643 |
\begin{minipage}{\linewidth} |
644 |
\begin{center} |
645 |
|
646 |
\caption{Calculation results for thermal conductivity of crystal gold |
647 |
using different force fields at different temperatures and various |
648 |
thermal exchange rates. Errors of calculations in parentheses.} |
649 |
|
650 |
\begin{tabular}{ccccc} |
651 |
\hline |
652 |
Force Field Used & $\rho$(g/cm$^3$) & ${\langle T\rangle}$/K & |
653 |
$\langle dT/dz\rangle$(K/\AA) & $\lambda$(W/m/K)\\ |
654 |
\hline |
655 |
QSC & 19.188 & 300 & 1.44 & 1.10(0.06)\\ |
656 |
& & & 2.86 & 1.08(0.05)\\ |
657 |
& & & 5.14 & 1.15(0.07)\\ |
658 |
\\ |
659 |
& 19.263 & 300 & 2.31 & 1.25(0.06)\\ |
660 |
& & 575 & 3.02 & 1.02(0.07)\\ |
661 |
\\ |
662 |
\hline |
663 |
EAM & 19.045 & 300 & 1.24 & 1.24(0.16)\\ |
664 |
& & & 2.06 & 1.37(0.04)\\ |
665 |
& & & 2.55 & 1.41(0.07)\\ |
666 |
\\ |
667 |
& 19.263 & 300 & 1.06 & 1.45(0.13)\\ |
668 |
& & & 2.04 & 1.41(0.07)\\ |
669 |
& & & 2.41 & 1.53(0.10)\\ |
670 |
& & 575 & 2.82 & 1.08(0.03)\\ |
671 |
\hline |
672 |
\end{tabular} |
673 |
\label{AuThermal} |
674 |
\end{center} |
675 |
\end{minipage} |
676 |
\end{table*} |
677 |
|
678 |
|
679 |
\subsection{Interfaciel Thermal Conductivity} |
680 |
After simulations of homogeneous water and gold systems using |
681 |
NIVS-RNEMD method were proved valid, calculation of gold/water |
682 |
interfacial thermal conductivity was followed. It is found out that |
683 |
the low interfacial conductance is probably due to the hydrophobic |
684 |
surface in our system. Figure \ref{interface} (a) demonstrates mass |
685 |
density change along $z$-axis, which is perpendicular to the |
686 |
gold/water interface. It is observed that water density significantly |
687 |
decreases when approaching the surface. Under this low thermal |
688 |
conductance, both gold and water phase have sufficient time to |
689 |
eliminate temperature difference inside respectively (Figure |
690 |
\ref{interface} b). With indistinguishable temperature difference |
691 |
within respective phase, it is valid to assume that the temperature |
692 |
difference between gold and water on surface would be approximately |
693 |
the same as the difference between the gold and water phase. This |
694 |
assumption enables convenient calculation of $G$ using |
695 |
Eq. \ref{interfaceCalc} instead of measuring temperatures of thin |
696 |
layer of water and gold close enough to surface, which would have |
697 |
greater fluctuation and lower accuracy. Reported results (Table |
698 |
\ref{interfaceRes}) are of two orders of magnitude smaller than our |
699 |
calculations on homogeneous systems, and thus have larger relative |
700 |
errors than our calculation results on homogeneous systems. |
701 |
|
702 |
\begin{figure} |
703 |
\includegraphics[width=\linewidth]{interface} |
704 |
\caption{Simulation results for Gold/Water interfacial thermal |
705 |
conductivity: (a) Significant water density decrease is observed on |
706 |
crystalline gold surface, which indicates low surface contact and |
707 |
leads to low thermal conductance. (b) Temperature profiles for a |
708 |
series of simulations. Temperatures of different slabs in the same |
709 |
phase show no significant differences.} |
710 |
\label{interface} |
711 |
\end{figure} |
712 |
|
713 |
\begin{table*} |
714 |
\begin{minipage}{\linewidth} |
715 |
\begin{center} |
716 |
|
717 |
\caption{Calculation results for interfacial thermal conductivity |
718 |
at ${\langle T\rangle \sim}$ 300K at various thermal exchange |
719 |
rates. Errors of calculations in parentheses. } |
720 |
|
721 |
\begin{tabular}{cccc} |
722 |
\hline |
723 |
$J_z$(MW/m$^2$) & $T_{gold}$ & $T_{water}$ & $G$(MW/m$^2$/K)\\ |
724 |
\hline |
725 |
98.0 & 355.2 & 295.8 & 1.65(0.21) \\ |
726 |
78.8 & 343.8 & 298.0 & 1.72(0.32) \\ |
727 |
73.6 & 344.3 & 298.0 & 1.59(0.24) \\ |
728 |
49.2 & 330.1 & 300.4 & 1.65(0.35) \\ |
729 |
\hline |
730 |
\end{tabular} |
731 |
\label{interfaceRes} |
732 |
\end{center} |
733 |
\end{minipage} |
734 |
\end{table*} |
735 |
|
736 |
\subsection{Shear Viscosity} |
737 |
Our calculations (Table \ref{shearRate}) shows that scale RNEMD method |
738 |
produced comparable shear viscosity to swap RNEMD method. In Table |
739 |
\ref{shearRate}, the names of the calculated samples are devided into |
740 |
two parts. The first number refers to total slabs in one simulation |
741 |
box. The second number refers to the swapping interval in swap method, or |
742 |
in scale method the equilvalent swapping interval that the same |
743 |
momentum flux would theoretically result in swap method. All the scale |
744 |
method results were from simulations that had a scaling interval of 10 |
745 |
time steps. The average molecular momentum gradients of these samples |
746 |
are shown in Figure \ref{shear} (a) and (b). |
747 |
|
748 |
\begin{table*} |
749 |
\begin{minipage}{\linewidth} |
750 |
\begin{center} |
751 |
|
752 |
\caption{Calculation results for shear viscosity of Lennard-Jones |
753 |
fluid at ${T^* = 0.72}$ and ${\rho^* = 0.85}$, with swap and scale |
754 |
methods at various momentum exchange rates. Results in reduced |
755 |
unit. Errors of calculations in parentheses. } |
756 |
|
757 |
\begin{tabular}{ccccc} |
758 |
Swapping method & & & NIVS-RNEMD & \\ |
759 |
\hline |
760 |
Swap Interval (fs) & $\eta^*_{swap}$ & & Equilvalent $j_p^*(v_x)$ & |
761 |
$\eta^*_{scale}$\\ |
762 |
\hline |
763 |
500 & 3.64(0.05) & & 0.09 & 3.76(0.09)\\ |
764 |
1000 & 3.52(0.16) & & 0.046 & 3.66(0.06)\\ |
765 |
2000 & 3.72(0.05) & & 0.024 & 3.32(0.18)\\ |
766 |
2500 & 3.42(0.06) & & 0.019 & 3.43(0.08)\\ |
767 |
\hline |
768 |
\end{tabular} |
769 |
\label{shearRate} |
770 |
\end{center} |
771 |
\end{minipage} |
772 |
\end{table*} |
773 |
|
774 |
\begin{figure} |
775 |
\includegraphics[width=\linewidth]{shear} |
776 |
\caption{Average momentum gradients in shear viscosity simulations, |
777 |
using (a) ``swapping'' method and (b) NIVS-RNEMD method |
778 |
respectively. (c) Temperature difference among x and y, z dimensions |
779 |
observed when using NIVS-RNEMD with equivalent exchange interval of |
780 |
500 fs.} |
781 |
\label{shear} |
782 |
\end{figure} |
783 |
|
784 |
However, observations of temperatures along three dimensions show that |
785 |
inhomogeneity occurs in scaling RNEMD simulations, particularly in the |
786 |
two slabs which were scaled. Figure \ref{shear} (c) indicate that with |
787 |
relatively large imposed momentum flux, the temperature difference among $x$ |
788 |
and the other two dimensions was significant. This would result from the |
789 |
algorithm of scaling method. From Eq. \ref{eq:fluxPlane}, after |
790 |
momentum gradient is set up, $P_c^x$ would be roughly stable |
791 |
($<0$). Consequently, scaling factor $x$ would most probably larger |
792 |
than 1. Therefore, the kinetic energy in $x$-dimension $K_c^x$ would |
793 |
keep increase after most scaling steps. And if there is not enough time |
794 |
for the kinetic energy to exchange among different dimensions and |
795 |
different slabs, the system would finally build up temperature |
796 |
(kinetic energy) difference among the three dimensions. Also, between |
797 |
$y$ and $z$ dimensions in the scaled slabs, temperatures of $z$-axis |
798 |
are closer to neighbor slabs. This is due to momentum transfer along |
799 |
$z$ dimension between slabs. |
800 |
|
801 |
Although results between scaling and swapping methods are comparable, |
802 |
the inherent temperature inhomogeneity even in relatively low imposed |
803 |
exchange momentum flux simulations makes scaling RNEMD method less |
804 |
attractive than swapping RNEMD in shear viscosity calculation. |
805 |
|
806 |
\section{Conclusions} |
807 |
NIVS-RNEMD simulation method is developed and tested on various |
808 |
systems. Simulation results demonstrate its validity in thermal |
809 |
conductivity calculations, from Lennard-Jones fluid to multi-atom |
810 |
molecule like water and metal crystals. NIVS-RNEMD improves |
811 |
non-Boltzmann-Maxwell distributions, which exist in previous RNEMD |
812 |
methods. Furthermore, it develops a valid means for unphysical thermal |
813 |
transfer between different species of molecules, and thus extends its |
814 |
applicability to interfacial systems. Our calculation of gold/water |
815 |
interfacial thermal conductivity demonstrates this advantage over |
816 |
previous RNEMD methods. NIVS-RNEMD has also limited application on |
817 |
shear viscosity calculations, but could cause temperature difference |
818 |
among different dimensions under high momentum flux. Modification is |
819 |
necessary to extend the applicability of NIVS-RNEMD in shear viscosity |
820 |
calculations. |
821 |
|
822 |
\section{Acknowledgments} |
823 |
Support for this project was provided by the National Science |
824 |
Foundation under grant CHE-0848243. Computational time was provided by |
825 |
the Center for Research Computing (CRC) at the University of Notre |
826 |
Dame. \newpage |
827 |
|
828 |
\bibliographystyle{aip} |
829 |
\bibliography{nivsRnemd} |
830 |
|
831 |
\end{doublespace} |
832 |
\end{document} |
833 |
|