ViewVC Help
View File | Revision Log | Show Annotations | View Changeset | Root Listing
root/group/trunk/nivsRnemd/nivsRnemd.tex
Revision: 3589
Committed: Thu Apr 15 19:42:05 2010 UTC (14 years, 2 months ago) by skuang
Content type: application/x-tex
File size: 37159 byte(s)
Log Message:
combine two distribution histograms into one. and some minor edit.

File Contents

# Content
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 methods
104 \cite{ISI:A1988Q205300014} because it imposes what is typically
105 difficult to measure (a flux or stress) and it is typically much
106 easier to compute momentum gradients or strains (the response). For
107 similar reasons, RNEMD is also preferable to slowly-converging
108 equilibrium methods for measuring thermal conductivity and shear
109 viscosity (using Green-Kubo relations [CITATIONS NEEDED]
110 or the Helfand moment approach of Viscardy {\it et
111 al.}\cite{Viscardy:2007bh,Viscardy:2007lq}) because these rely on
112 computing difficult to measure quantities.
113
114 Another attractive feature of RNEMD is that it conserves both total
115 linear momentum and total energy during the swaps (as long as the two
116 molecules have the same identity), so the swapped configurations are
117 typically samples from the same manifold of states in the
118 microcanonical ensemble.
119
120 Recently, Tenney and Maginn\cite{Maginn:2010} have discovered
121 some problems with the original RNEMD swap technique. Notably, large
122 momentum fluxes (equivalent to frequent momentum swaps between the
123 slabs) can result in ``notched'', ``peaked'' and generally non-thermal
124 momentum distributions in the two slabs, as well as non-linear thermal
125 and velocity distributions along the direction of the imposed flux
126 ($z$). Tenney and Maginn obtained reasonable limits on imposed flux
127 and self-adjusting metrics for retaining the usability of the method.
128
129 In this paper, we develop and test a method for non-isotropic velocity
130 scaling (NIVS-RNEMD) which retains the desirable features of RNEMD
131 (conservation of linear momentum and total energy, compatibility with
132 periodic boundary conditions) while establishing true thermal
133 distributions in each of the two slabs. In the next section, we
134 present the method for determining the scaling constraints. We then
135 test the method on both single component, multi-component, and
136 non-isotropic mixtures and show that it is capable of providing
137 reasonable estimates of the thermal conductivity and shear viscosity
138 in these cases.
139
140 \section{Methodology}
141 We retain the basic idea of M\"{u}ller-Plathe's RNEMD method; the
142 periodic system is partitioned into a series of thin slabs along one
143 axis ($z$). One of the slabs at the end of the periodic box is
144 designated the ``hot'' slab, while the slab in the center of the box
145 is designated the ``cold'' slab. The artificial momentum flux will be
146 established by transferring momentum from the cold slab and into the
147 hot slab.
148
149 Rather than using momentum swaps, we use a series of velocity scaling
150 moves. For molecules $\{i\}$ located within the cold slab,
151 \begin{equation}
152 \vec{v}_i \leftarrow \left( \begin{array}{ccc}
153 x & 0 & 0 \\
154 0 & y & 0 \\
155 0 & 0 & z \\
156 \end{array} \right) \cdot \vec{v}_i
157 \end{equation}
158 where ${x, y, z}$ are a set of 3 scaling variables for each of the
159 three directions in the system. Likewise, the molecules $\{j\}$
160 located in the hot slab will see a concomitant scaling of velocities,
161 \begin{equation}
162 \vec{v}_j \leftarrow \left( \begin{array}{ccc}
163 x^\prime & 0 & 0 \\
164 0 & y^\prime & 0 \\
165 0 & 0 & z^\prime \\
166 \end{array} \right) \cdot \vec{v}_j
167 \end{equation}
168
169 Conservation of linear momentum in each of the three directions
170 ($\alpha = x,y,z$) ties the values of the hot and cold scaling
171 parameters together:
172 \begin{equation}
173 P_h^\alpha + P_c^\alpha = \alpha^\prime P_h^\alpha + \alpha P_c^\alpha
174 \end{equation}
175 where
176 \begin{eqnarray}
177 P_c^\alpha & = & \sum_{i = 1}^{N_c} m_i \left[\vec{v}_i\right]_\alpha \\
178 P_h^\alpha & = & \sum_{j = 1}^{N_h} m_j \left[\vec{v}_j\right]_\alpha
179 \label{eq:momentumdef}
180 \end{eqnarray}
181 Therefore, for each of the three directions, the hot scaling
182 parameters are a simple function of the cold scaling parameters and
183 the instantaneous linear momentum in each of the two slabs.
184 \begin{equation}
185 \alpha^\prime = 1 + (1 - \alpha) p_\alpha
186 \label{eq:hotcoldscaling}
187 \end{equation}
188 where
189 \begin{equation}
190 p_\alpha = \frac{P_c^\alpha}{P_h^\alpha}
191 \end{equation}
192 for convenience.
193
194 Conservation of total energy also places constraints on the scaling:
195 \begin{equation}
196 \sum_{\alpha = x,y,z} K_h^\alpha + K_c^\alpha = \sum_{\alpha = x,y,z}
197 \left(\alpha^\prime\right)^2 K_h^\alpha + \alpha^2 K_c^\alpha
198 \end{equation}
199 where the translational kinetic energies, $K_h^\alpha$ and
200 $K_c^\alpha$, are computed for each of the three directions in a
201 similar manner to the linear momenta (Eq. \ref{eq:momentumdef}).
202 Substituting in the expressions for the hot scaling parameters
203 ($\alpha^\prime$) from Eq. (\ref{eq:hotcoldscaling}), we obtain the
204 {\it constraint ellipsoid}:
205 \begin{equation}
206 \sum_{\alpha = x,y,z} a_\alpha \alpha^2 + b_\alpha \alpha + c_\alpha = 0
207 \label{eq:constraintEllipsoid}
208 \end{equation}
209 where the constants are obtained from the instantaneous values of the
210 linear momenta and kinetic energies for the hot and cold slabs,
211 \begin{eqnarray}
212 a_\alpha & = &\left(K_c^\alpha + K_h^\alpha
213 \left(p_\alpha\right)^2\right) \\
214 b_\alpha & = & -2 K_h^\alpha p_\alpha \left( 1 + p_\alpha\right) \\
215 c_\alpha & = & K_h^\alpha p_\alpha^2 + 2 K_h^\alpha p_\alpha - K_c^\alpha
216 \label{eq:constraintEllipsoidConsts}
217 \end{eqnarray}
218 This ellipsoid (Eq. \ref{eq:constraintEllipsoid}) defines the set of
219 cold slab scaling parameters which can be applied while preserving
220 both linear momentum in all three directions as well as total kinetic
221 energy.
222
223 The goal of using velocity scaling variables is to transfer linear
224 momentum or kinetic energy from the cold slab to the hot slab. If the
225 hot and cold slabs are separated along the z-axis, the energy flux is
226 given simply by the decrease in kinetic energy of the cold bin:
227 \begin{equation}
228 (1-x^2) K_c^x + (1-y^2) K_c^y + (1-z^2) K_c^z = J_z\Delta t
229 \end{equation}
230 The expression for the energy flux can be re-written as another
231 ellipsoid centered on $(x,y,z) = 0$:
232 \begin{equation}
233 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
234 \label{eq:fluxEllipsoid}
235 \end{equation}
236 The spatial extent of the {\it thermal flux ellipsoid} is governed
237 both by a targetted value, $J_z$ as well as the instantaneous values
238 of the kinetic energy components in the cold bin.
239
240 To satisfy an energetic flux as well as the conservation constraints,
241 we must determine the points ${x,y,z}$ which lie on both the
242 constraint ellipsoid (Eq. \ref{eq:constraintEllipsoid}) and the flux
243 ellipsoid (Eq. \ref{eq:fluxEllipsoid}), i.e. the intersection of the
244 two ellipsoids in 3-dimensional space.
245
246 \begin{figure}
247 \includegraphics[width=\linewidth]{ellipsoids}
248 \caption{Scaling points which maintain both constant energy and
249 constant linear momentum of the system lie on the surface of the
250 {\it constraint ellipsoid} while points which generate the target
251 momentum flux lie on the surface of the {\it flux ellipsoid}. The
252 velocity distributions in the cold bin are scaled by only those
253 points which lie on both ellipsoids.}
254 \label{ellipsoids}
255 \end{figure}
256
257 One may also define {\it momentum} flux (say along the $x$-direction) as:
258 \begin{equation}
259 (1-x) P_c^x = j_z(p_x)\Delta t
260 \label{eq:fluxPlane}
261 \end{equation}
262 The above {\it momentum flux plane} is perpendicular to the $x$-axis,
263 with its position governed both by a target value, $j_z(p_x)$ as well
264 as the instantaneous value of the momentum along the $x$-direction.
265
266 In order to satisfy a momentum flux as well as the conservation
267 constraints, we must determine the points ${x,y,z}$ which lie on both
268 the constraint ellipsoid (Eq. \ref{eq:constraintEllipsoid}) and the
269 flux plane (Eq. \ref{eq:fluxPlane}), i.e. the intersection of an
270 ellipsoid and a plane in 3-dimensional space.
271
272 In both the momentum and energy flux scenarios, valid scaling
273 parameters are arrived at by solving geometric intersection problems
274 in $x, y, z$ space in order to obtain cold slab scaling parameters.
275 Once the scaling variables for the cold slab are known, the hot slab
276 scaling has also been determined.
277
278
279 The following problem will be choosing an optimal set of scaling
280 variables among the possible sets. Although this method is inherently
281 non-isotropic, the goal is still to maintain the system as isotropic
282 as possible. Under this consideration, one would like the kinetic
283 energies in different directions could become as close as each other
284 after each scaling. Simultaneously, one would also like each scaling
285 as gentle as possible, i.e. ${x,y,z \rightarrow 1}$, in order to avoid
286 large perturbation to the system. Therefore, one approach to obtain
287 the scaling variables would be constructing an criteria function, with
288 constraints as above equation sets, and solving the function's minimum
289 by method like Lagrange multipliers.
290
291 In order to save computation time, we have a different approach to a
292 relatively good set of scaling variables with much less calculation
293 than above. Here is the detail of our simplification of the problem.
294
295 In the case of kinetic energy transfer, we impose another constraint
296 ${x = y}$, into the equation sets. Consequently, there are two
297 variables left. And now one only needs to solve a set of two {\it
298 ellipses equations}. This problem would be transformed into solving
299 one quartic equation for one of the two variables. There are known
300 generic methods that solve real roots of quartic equations. Then one
301 can determine the other variable and obtain sets of scaling
302 variables. Among these sets, one can apply the above criteria to
303 choose the best set, while much faster with only a few sets to choose.
304
305 In the case of momentum flux transfer, we impose another constraint to
306 set the kinetic energy transfer as zero. In another word, we apply
307 Eq. \ref{eq:fluxEllipsoid} and let ${J_z = 0}$. After that, with one
308 variable fixed by Eq. \ref{eq:fluxPlane}, one now have a similar set
309 of equations on the above kinetic energy transfer problem. Therefore,
310 an approach similar to the above would be sufficient for this as well.
311
312 \section{Computational Details}
313 \subsection{Lennard-Jones Fluid}
314 Our simulation consists of a series of systems. All of these
315 simulations were run with the OpenMD simulation software
316 package\cite{Meineke:2005gd} integrated with RNEMD codes.
317
318 A Lennard-Jones fluid system was built and tested first. In order to
319 compare our method with swapping RNEMD, a series of simulations were
320 performed to calculate the shear viscosity and thermal conductivity of
321 argon. 2592 atoms were in a orthorhombic cell, which was ${10.06\sigma
322 \times 10.06\sigma \times 30.18\sigma}$ by size. The reduced density
323 ${\rho^* = \rho\sigma^3}$ was thus 0.85, which enabled direct
324 comparison between our results and others. These simulations used
325 velocity Verlet algorithm with reduced timestep ${\tau^* =
326 4.6\times10^{-4}}$.
327
328 For shear viscosity calculation, the reduced temperature was ${T^* =
329 k_B T/\varepsilon = 0.72}$. Simulations were first equilibrated in canonical
330 ensemble (NVT), then equilibrated in microcanonical ensemble
331 (NVE). Establishing and stablizing momentum gradient were followed
332 also in NVE ensemble. For the swapping part, Muller-Plathe's algorithm was
333 adopted.\cite{ISI:000080382700030} The simulation box was under
334 periodic boundary condition, and devided into ${N = 20}$ slabs. In each swap,
335 the top slab ${(n = 1)}$ exchange the most negative $x$ momentum with the
336 most positive $x$ momentum in the center slab ${(n = N/2 + 1)}$. Referred
337 to Tenney {\it et al.}\cite{ISI:000273472300004}, a series of swapping
338 frequency were chosen. According to each result from swapping
339 RNEMD, scaling RNEMD simulations were run with the target momentum
340 flux set to produce a similar momentum flux, and consequently shear
341 rate. Furthermore, various scaling frequencies can be tested for one
342 single swapping rate. To test the temperature homogeneity in our
343 system of swapping and scaling methods, temperatures of different
344 dimensions in all the slabs were observed. Most of the simulations
345 include $10^5$ steps of equilibration without imposing momentum flux,
346 $10^5$ steps of stablization with imposing unphysical momentum
347 transfer, and $10^6$ steps of data collection under RNEMD. For
348 relatively high momentum flux simulations, ${5\times10^5}$ step data
349 collection is sufficient. For some low momentum flux simulations,
350 ${2\times10^6}$ steps were necessary.
351
352 After each simulation, the shear viscosity was calculated in reduced
353 unit. The momentum flux was calculated with total unphysical
354 transferred momentum ${P_x}$ and data collection time $t$:
355 \begin{equation}
356 j_z(p_x) = \frac{P_x}{2 t L_x L_y}
357 \end{equation}
358 where $L_x$ and $L_y$ denotes $x$ and $y$ lengths of the simulation
359 box, and physical momentum transfer occurs in two ways due to our
360 periodic boundary condition settings. And the velocity gradient
361 ${\langle \partial v_x /\partial z \rangle}$ can be obtained by a
362 linear regression of the velocity profile. From the shear viscosity
363 $\eta$ calculated with the above parameters, one can further convert
364 it into reduced unit ${\eta^* = \eta \sigma^2 (\varepsilon m)^{-1/2}}$.
365
366 For thermal conductivity calculations, simulations were first run under
367 reduced temperature ${\langle T^*\rangle = 0.72}$ in NVE
368 ensemble. Muller-Plathe's algorithm was adopted in the swapping
369 method. Under identical simulation box parameters with our shear
370 viscosity calculations, in each swap, the top slab exchanges all three
371 translational momentum components of the molecule with least kinetic
372 energy with the same components of the molecule in the center slab
373 with most kinetic energy, unless this ``coldest'' molecule in the
374 ``hot'' slab is still ``hotter'' than the ``hottest'' molecule in the
375 ``cold'' slab. According to swapping RNEMD results, target energy flux
376 for scaling RNEMD simulations can be set. Also, various scaling
377 frequencies can be tested for one target energy flux. To compare the
378 performance between swapping and scaling method, distributions of
379 velocity and speed in different slabs were observed.
380
381 For each swapping rate, thermal conductivity was calculated in reduced
382 unit. The energy flux was calculated similarly to the momentum flux,
383 with total unphysical transferred energy ${E_{total}}$ and data collection
384 time $t$:
385 \begin{equation}
386 J_z = \frac{E_{total}}{2 t L_x L_y}
387 \end{equation}
388 And the temperature gradient ${\langle\partial T/\partial z\rangle}$
389 can be obtained by a linear regression of the temperature
390 profile. From the thermal conductivity $\lambda$ calculated, one can
391 further convert it into reduced unit ${\lambda^*=\lambda \sigma^2
392 m^{1/2} k_B^{-1}\varepsilon^{-1/2}}$.
393
394 \subsection{ Water / Metal Thermal Conductivity}
395 Another series of our simulation is the calculation of interfacial
396 thermal conductivity of a Au/H$_2$O system. Respective calculations of
397 liquid water (Extended Simple Point Charge model) and crystal gold
398 thermal conductivity were performed and compared with current results
399 to ensure the validity of NIVS-RNEMD. After that, a mixture system was
400 simulated.
401
402 For thermal conductivity calculation of bulk water, a simulation box
403 consisting of 1000 molecules were first equilibrated under ambient
404 pressure and temperature conditions using NPT ensemble, followed by
405 equilibration in fixed volume (NVT). The system was then equilibrated in
406 microcanonical ensemble (NVE). Also in NVE ensemble, establishing a
407 stable thermal gradient was followed. The simulation box was under
408 periodic boundary condition and devided into 10 slabs. Data collection
409 process was similar to Lennard-Jones fluid system.
410
411 Thermal conductivity calculation of bulk crystal gold used a similar
412 protocol. Two types of force field parameters, Embedded Atom Method
413 (EAM) and Quantum Sutten-Chen (QSC) force field were used
414 respectively. The face-centered cubic crystal simulation box consists of
415 2880 Au atoms. The lattice was first allowed volume change to relax
416 under ambient temperature and pressure. Equilibrations in canonical and
417 microcanonical ensemble were followed in order. With the simulation
418 lattice devided evenly into 10 slabs, different thermal gradients were
419 established by applying a set of target thermal transfer flux. Data of
420 the series of thermal gradients was collected for calculation.
421
422 After simulations of bulk water and crystal gold, a mixture system was
423 constructed, consisting of 1188 Au atoms and 1862 H$_2$O
424 molecules. Spohr potential was adopted in depicting the interaction
425 between metal atom and water molecule.\cite{ISI:000167766600035} A
426 similar protocol of equilibration was followed. Several thermal
427 gradients was built under different target thermal flux. It was found
428 out that compared to our previous simulation systems, the two phases
429 could have large temperature difference even under a relatively low
430 thermal flux. Therefore, under our low flux conditions, it is assumed
431 that the metal and water phases have respectively homogeneous
432 temperature, excluding the surface regions. In calculating the
433 interfacial thermal conductivity $G$, this assumptioin was applied and
434 thus our formula becomes:
435
436 \begin{equation}
437 G = \frac{E_{total}}{2 t L_x L_y \left( \langle T_{gold}\rangle -
438 \langle T_{water}\rangle \right)}
439 \label{interfaceCalc}
440 \end{equation}
441 where ${E_{total}}$ is the imposed unphysical kinetic energy transfer
442 and ${\langle T_{gold}\rangle}$ and ${\langle T_{water}\rangle}$ are the
443 average observed temperature of gold and water phases respectively.
444
445 \section{Results And Discussions}
446 \subsection{Thermal Conductivity}
447 \subsubsection{Lennard-Jones Fluid}
448 Our thermal conductivity calculations show that scaling method results
449 agree with swapping method. Four different exchange intervals were
450 tested (Table \ref{thermalLJRes}) using swapping method. With a fixed
451 10fs exchange interval, target exchange kinetic energy was set to
452 produce equivalent kinetic energy flux as in swapping method. And
453 similar thermal gradients were observed with similar thermal flux in
454 two simulation methods (Figure \ref{thermalGrad}).
455
456 \begin{table*}
457 \begin{minipage}{\linewidth}
458 \begin{center}
459
460 \caption{Calculation results for thermal conductivity of Lennard-Jones
461 fluid at ${\langle T^* \rangle = 0.72}$ and ${\rho^* = 0.85}$, with
462 swap and scale methods at various kinetic energy exchange rates. Results
463 in reduced unit. Errors of calculations in parentheses.}
464
465 \begin{tabular}{ccc}
466 \hline
467 (Equilvalent) Exchange Interval (fs) & $\lambda^*_{swap}$ &
468 $\lambda^*_{scale}$\\
469 \hline
470 250 & 7.03(0.34) & 7.30(0.10)\\
471 500 & 7.03(0.14) & 6.95(0.09)\\
472 1000 & 6.91(0.42) & 7.19(0.07)\\
473 2000 & 7.52(0.15) & 7.19(0.28)\\
474 \hline
475 \end{tabular}
476 \label{thermalLJRes}
477 \end{center}
478 \end{minipage}
479 \end{table*}
480
481 \begin{figure}
482 \includegraphics[width=\linewidth]{thermalGrad}
483 \caption{NIVS-RNEMD method introduced similar temperature gradients
484 compared to ``swapping'' method under various kinetic energy flux in
485 thermal conductivity simulations.}
486 \label{thermalGrad}
487 \end{figure}
488
489 During these simulations, molecule velocities were recorded in 1000 of
490 all the snapshots of one single data collection process. These
491 velocity data were used to produce histograms of velocity and speed
492 distribution in different slabs. From these histograms, it is observed
493 that under relatively high unphysical kinetic energy flux, speed and
494 velocity distribution of molecules in slabs where swapping occured
495 could deviate from Maxwell-Boltzmann distribution. Figure
496 \ref{thermalHist} a) illustrates how these distributions deviate from an
497 ideal distribution. In high temperature slab, probability density in
498 low speed is confidently smaller than ideal curve fit; in low
499 temperature slab, probability density in high speed is smaller than
500 ideal, while larger than ideal in low speed. This phenomenon is more
501 obvious in our high swapping rate simulations. And this deviation
502 could also leads to deviation of distribution of velocity in various
503 dimensions. One feature of these deviated distribution is that in high
504 temperature slab, the ideal Gaussian peak was changed into a
505 relatively flat plateau; while in low temperature slab, that peak
506 appears sharper. This problem is rooted in the mechanism of the
507 swapping method. Continually depleting low (high) speed particles in
508 the high (low) temperature slab could not be complemented by
509 diffusions of low (high) speed particles from neighbor slabs, unless
510 in suffciently low swapping rate. Simutaneously, surplus low speed
511 particles in the low temperature slab do not have sufficient time to
512 diffuse to neighbor slabs. However, thermal exchange rate should reach
513 a minimum level to produce an observable thermal gradient under noise
514 interference. Consequently, swapping RNEMD has a relatively narrow
515 choice of swapping rate to satisfy these above restrictions.
516
517 Comparatively, NIVS-RNEMD has a speed distribution closer to the ideal
518 curve fit (Figure \ref{thermalHist} b). Essentially, after scaling, a
519 Gaussian distribution function would remain Gaussian. Although a
520 single scaling is non-isotropic in all three dimensions, our scaling
521 coefficient criteria could help maintian the scaling region as
522 isotropic as possible. On the other hand, scaling coefficients are
523 preferred to be as close to 1 as possible, which also helps minimize
524 the difference among different dimensions. This is possible if scaling
525 interval and one-time thermal transfer energy are well
526 chosen. Consequently, NIVS-RNEMD is able to impose an unphysical
527 thermal flux as the previous RNEMD method without large perturbation
528 to the distribution of velocity and speed in the exchange regions.
529
530 \begin{figure}
531 \includegraphics[width=\linewidth]{thermalHist}
532 \caption{Speed distribution for thermal conductivity using a)
533 ``swapping'' and b) NIVS- RNEMD methods. Shown is from the
534 simulations with an exchange or equilvalent exchange interval of 250
535 fs.}
536 \label{thermalHist}
537 \end{figure}
538
539 \subsubsection{SPC/E Water}
540 Our results of SPC/E water thermal conductivity are comparable to
541 Bedrov {\it et al.}\cite{ISI:000090151400044}, which employed the
542 previous swapping RNEMD method for their calculation. Bedrov {\it et
543 al.}\cite{ISI:000090151400044} argued that exchange of the molecule
544 center-of-mass velocities instead of single atom velocities in a
545 molecule conserves the total kinetic energy and linear momentum. This
546 principle is adopted in our simulations. Scaling is applied to the
547 velocities of the rigid bodies of SPC/E model water molecules, instead
548 of each hydrogen and oxygen atoms in relevant water molecules. As
549 shown in Figure \ref{spceGrad}, temperature gradients were established
550 similar to their system. However, the average temperature of our
551 system is 300K, while theirs is 318K, which would be attributed for
552 part of the difference between the final calculation results (Table
553 \ref{spceThermal}). Both methods yields values in agreement with
554 experiment. And this shows the applicability of our method to
555 multi-atom molecular system.
556
557 \begin{figure}
558 \includegraphics[width=\linewidth]{spceGrad}
559 \caption{Temperature gradients for SPC/E water thermal conductivity.}
560 \label{spceGrad}
561 \end{figure}
562
563 \begin{table*}
564 \begin{minipage}{\linewidth}
565 \begin{center}
566
567 \caption{Calculation results for thermal conductivity of SPC/E water
568 at ${\langle T\rangle}$ = 300K at various thermal exchange rates. Errors of
569 calculations in parentheses. }
570
571 \begin{tabular}{cccc}
572 \hline
573 $\langle dT/dz\rangle$(K/\AA) & & $\lambda$(W/m/K) & \\
574 & This work & Previous simulations\cite{ISI:000090151400044} &
575 Experiment$^a$\\
576 \hline
577 0.38 & 0.816(0.044) & & 0.64\\
578 0.81 & 0.770(0.008) & 0.784\\
579 1.54 & 0.813(0.007) & 0.730\\
580 \hline
581 \end{tabular}
582 \label{spceThermal}
583 \end{center}
584 \end{minipage}
585 \end{table*}
586
587 \subsubsection{Crystal Gold}
588 Our results of gold thermal conductivity using two force fields are
589 shown separately in Table \ref{qscThermal} and \ref{eamThermal}. In
590 these calculations,the end and middle slabs were excluded in thermal
591 gradient regession and only used as heat source and drain in the
592 systems. Our yielded values using EAM force field are slightly larger
593 than those using QSC force field. However, both series are
594 significantly smaller than experimental value by an order of more than
595 100. It has been verified that this difference is mainly attributed to
596 the lack of electron interaction representation in these force field
597 parameters. Richardson {\it et al.}\cite{Clancy:1992} used EAM
598 force field parameters in their metal thermal conductivity
599 calculations. The Non-Equilibrium MD method they employed in their
600 simulations produced comparable results to ours. As Zhang {\it et
601 al.}\cite{ISI:000231042800044} stated, thermal conductivity values
602 are influenced mainly by force field. Therefore, it is confident to
603 conclude that NIVS-RNEMD is applicable to metal force field system.
604
605 \begin{figure}
606 \includegraphics[width=\linewidth]{AuGrad}
607 \caption{Temperature gradients for thermal conductivity calculation of
608 crystal gold using QSC force field.}
609 \label{AuGrad}
610 \end{figure}
611
612 \begin{table*}
613 \begin{minipage}{\linewidth}
614 \begin{center}
615
616 \caption{Calculation results for thermal conductivity of crystal gold
617 using QSC force field at ${\langle T\rangle}$ = 300K at various
618 thermal exchange rates. Errors of calculations in parentheses. }
619
620 \begin{tabular}{cc}
621 \hline
622 $\langle dT/dz\rangle$(K/\AA) & $\lambda$(W/m/K)\\
623 \hline
624 1.44 & 1.10(0.01)\\
625 2.86 & 1.08(0.02)\\
626 5.14 & 1.15(0.01)\\
627 \hline
628 \end{tabular}
629 \label{qscThermal}
630 \end{center}
631 \end{minipage}
632 \end{table*}
633
634 \begin{figure}
635 \includegraphics[width=\linewidth]{eamGrad}
636 \caption{Temperature gradients for thermal conductivity calculation of
637 crystal gold using EAM force field.}
638 \label{eamGrad}
639 \end{figure}
640
641 \begin{table*}
642 \begin{minipage}{\linewidth}
643 \begin{center}
644
645 \caption{Calculation results for thermal conductivity of crystal gold
646 using EAM force field at ${\langle T\rangle}$ = 300K at various
647 thermal exchange rates. Errors of calculations in parentheses. }
648
649 \begin{tabular}{cc}
650 \hline
651 $\langle dT/dz\rangle$(K/\AA) & $\lambda$(W/m/K)\\
652 \hline
653 1.24 & 1.24(0.06)\\
654 2.06 & 1.37(0.04)\\
655 2.55 & 1.41(0.03)\\
656 \hline
657 \end{tabular}
658 \label{eamThermal}
659 \end{center}
660 \end{minipage}
661 \end{table*}
662
663
664 \subsection{Interfaciel Thermal Conductivity}
665 After simulations of homogeneous water and gold systems using
666 NIVS-RNEMD method were proved valid, calculation of gold/water
667 interfacial thermal conductivity was followed. It is found out that
668 the low interfacial conductance is probably due to the hydrophobic
669 surface in our system. Figure \ref{interfaceDensity} demonstrates mass
670 density change along $z$-axis, which is perpendicular to the
671 gold/water interface. It is observed that water density significantly
672 decreases when approaching the surface. Under this low thermal
673 conductance, both gold and water phase have sufficient time to
674 eliminate temperature difference inside respectively (Figure
675 \ref{interfaceGrad}). With indistinguishable temperature difference
676 within respective phase, it is valid to assume that the temperature
677 difference between gold and water on surface would be approximately
678 the same as the difference between the gold and water phase. This
679 assumption enables convenient calculation of $G$ using
680 Eq. \ref{interfaceCalc} instead of measuring temperatures of thin
681 layer of water and gold close enough to surface, which would have
682 greater fluctuation and lower accuracy. Reported results (Table
683 \ref{interfaceRes}) are of two orders of magnitude smaller than our
684 calculations on homogeneous systems, and thus have larger relative
685 errors than our calculation results on homogeneous systems.
686
687 \begin{figure}
688 \includegraphics[width=\linewidth]{interfaceDensity}
689 \caption{Density profile for interfacial thermal conductivity
690 simulation box. Significant water density decrease is observed on
691 gold surface.}
692 \label{interfaceDensity}
693 \end{figure}
694
695 \begin{figure}
696 \includegraphics[width=\linewidth]{interfaceGrad}
697 \caption{Temperature profiles for interfacial thermal conductivity
698 simulation box. Temperatures of different slabs in the same phase
699 show no significant difference.}
700 \label{interfaceGrad}
701 \end{figure}
702
703 \begin{table*}
704 \begin{minipage}{\linewidth}
705 \begin{center}
706
707 \caption{Calculation results for interfacial thermal conductivity
708 at ${\langle T\rangle \sim}$ 300K at various thermal exchange
709 rates. Errors of calculations in parentheses. }
710
711 \begin{tabular}{cccc}
712 \hline
713 $J_z$(MW/m$^2$) & $T_{gold}$ & $T_{water}$ & $G$(MW/m$^2$/K)\\
714 \hline
715 98.0 & 355.2 & 295.8 & 1.65(0.21) \\
716 78.8 & 343.8 & 298.0 & 1.72(0.32) \\
717 73.6 & 344.3 & 298.0 & 1.59(0.24) \\
718 49.2 & 330.1 & 300.4 & 1.65(0.35) \\
719 \hline
720 \end{tabular}
721 \label{interfaceRes}
722 \end{center}
723 \end{minipage}
724 \end{table*}
725
726 \subsection{Shear Viscosity}
727 Our calculations (Table \ref{shearRate}) shows that scale RNEMD method
728 produced comparable shear viscosity to swap RNEMD method. In Table
729 \ref{shearRate}, the names of the calculated samples are devided into
730 two parts. The first number refers to total slabs in one simulation
731 box. The second number refers to the swapping interval in swap method, or
732 in scale method the equilvalent swapping interval that the same
733 momentum flux would theoretically result in swap method. All the scale
734 method results were from simulations that had a scaling interval of 10
735 time steps. The average molecular momentum gradients of these samples
736 are shown in Figure \ref{shearGrad}.
737
738 \begin{table*}
739 \begin{minipage}{\linewidth}
740 \begin{center}
741
742 \caption{Calculation results for shear viscosity of Lennard-Jones
743 fluid at ${T^* = 0.72}$ and ${\rho^* = 0.85}$, with swap and scale
744 methods at various momentum exchange rates. Results in reduced
745 unit. Errors of calculations in parentheses. }
746
747 \begin{tabular}{ccc}
748 \hline
749 Series & $\eta^*_{swap}$ & $\eta^*_{scale}$\\
750 \hline
751 20-500 & 3.64(0.05) & 3.76(0.09)\\
752 20-1000 & 3.52(0.16) & 3.66(0.06)\\
753 20-2000 & 3.72(0.05) & 3.32(0.18)\\
754 20-2500 & 3.42(0.06) & 3.43(0.08)\\
755 \hline
756 \end{tabular}
757 \label{shearRate}
758 \end{center}
759 \end{minipage}
760 \end{table*}
761
762 \begin{figure}
763 \includegraphics[width=\linewidth]{shearGrad}
764 \caption{Average momentum gradients of shear viscosity simulations}
765 \label{shearGrad}
766 \end{figure}
767
768 \begin{figure}
769 \includegraphics[width=\linewidth]{shearTempScale}
770 \caption{Temperature profile for scaling RNEMD simulation.}
771 \label{shearTempScale}
772 \end{figure}
773 However, observations of temperatures along three dimensions show that
774 inhomogeneity occurs in scaling RNEMD simulations, particularly in the
775 two slabs which were scaled. Figure \ref{shearTempScale} indicate that with
776 relatively large imposed momentum flux, the temperature difference among $x$
777 and the other two dimensions was significant. This would result from the
778 algorithm of scaling method. From Eq. \ref{eq:fluxPlane}, after
779 momentum gradient is set up, $P_c^x$ would be roughly stable
780 ($<0$). Consequently, scaling factor $x$ would most probably larger
781 than 1. Therefore, the kinetic energy in $x$-dimension $K_c^x$ would
782 keep increase after most scaling steps. And if there is not enough time
783 for the kinetic energy to exchange among different dimensions and
784 different slabs, the system would finally build up temperature
785 (kinetic energy) difference among the three dimensions. Also, between
786 $y$ and $z$ dimensions in the scaled slabs, temperatures of $z$-axis
787 are closer to neighbor slabs. This is due to momentum transfer along
788 $z$ dimension between slabs.
789
790 Although results between scaling and swapping methods are comparable,
791 the inherent temperature inhomogeneity even in relatively low imposed
792 exchange momentum flux simulations makes scaling RNEMD method less
793 attractive than swapping RNEMD in shear viscosity calculation.
794
795 \section{Conclusions}
796 NIVS-RNEMD simulation method is developed and tested on various
797 systems. Simulation results demonstrate its validity in thermal
798 conductivity calculations, from Lennard-Jones fluid to multi-atom
799 molecule like water and metal crystals. NIVS-RNEMD improves
800 non-Boltzmann-Maxwell distributions, which exist in previous RNEMD
801 methods. Furthermore, it develops a valid means for unphysical thermal
802 transfer between different species of molecules, and thus extends its
803 applicability to interfacial systems. Our calculation of gold/water
804 interfacial thermal conductivity demonstrates this advantage over
805 previous RNEMD methods. NIVS-RNEMD has also limited application on
806 shear viscosity calculations, but could cause temperature difference
807 among different dimensions under high momentum flux. Modification is
808 necessary to extend the applicability of NIVS-RNEMD in shear viscosity
809 calculations.
810
811 \section{Acknowledgments}
812 Support for this project was provided by the National Science
813 Foundation under grant CHE-0848243. Computational time was provided by
814 the Center for Research Computing (CRC) at the University of Notre
815 Dame. \newpage
816
817 \bibliographystyle{aip}
818 \bibliography{nivsRnemd}
819
820 \end{doublespace}
821 \end{document}
822