cond-mat0208094/lb.tex
1: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
2: \documentstyle[aps]{revtex}
3: %----------------------------------------------------------------
4: \begin{document}
5: \draft
6: \title{
7: Kolmogorov Scaling for the $\epsilon$-entropy 
8: in a Forced Turbulence Simulation 
9: \cite{N}}
10: \author{
11: W. Sakikawa and O. Narikiyo}
12: \address{
13: Department of Physics, 
14: Kyushu University, 
15: Fukuoka 810-8560, 
16: Japan}
17: \date{
18: August, 2002}
19: \maketitle
20: %----------------------------------------------------------------
21: \begin{abstract}
22: We have shown that 
23: the $\epsilon$-entropy determined from the time-series 
24: of the velocity fluctuation in a forced turbulence 
25: simulated by the lattice Boltzmann method 
26: obeys the Kolmogorov scaling. 
27: \vskip 8pt
28: \noindent
29: {\it Keywords:} 
30: forced turbulence, 
31: Kolmogorov scaling, 
32: $\epsilon$-entropy, 
33: lattice Boltzmann method
34: \end{abstract}
35: %----------------------------------------------------------------
36: \vskip 18pt
37:   Turbulence is one of the typical phenomena 
38: with multiscale motions. 
39:   Each phenomena of a scale strongly couples with 
40: all the other scales of turbulent motion. 
41:   In order to analyze such a system 
42: a scale-dependent entropy, the so-called $\epsilon$-entropy, 
43: works well.\cite{Vul} 
44:   For example, the time series of the velocity fluctuation in turbulence 
45: leads to a non-trivial scaling relation of the $\epsilon$-entropy, 
46: $ h(\epsilon) \propto \epsilon^{-3} $, 
47: expected from the Kolmogorov scaling.\cite{Vul} 
48:   The existing experimental data are consistent with this scaling.\cite{Vul} 
49:   In this Short Note we try to show the consistency of the scaling 
50: with the simulation on the basis of the lattice Boltzmann method 
51: which is one of the easiest ways to simulate turbulence. 
52: 
53:   The distribution function $f_i({\vec r},t)$ 
54: of the particle with the velocity ${\vec c}_i$ 
55: at position ${\vec r}$ and time $t$ 
56: obeys the lattice Boltzmann equation\cite{S,WG} 
57: %----------------------------------------------------------------
58: \begin{equation}
59: f_i({\vec r}+{\vec c}_i,t+1) - f_i({\vec r},t)
60: = - \omega [ f_i({\vec r},t) - f_i^0({\vec r},t) ],
61: \end{equation}
62: %----------------------------------------------------------------
63: where we have adopted the Bhatnagar-Gross-Krook (BGK) model 
64: for the collision term 
65: and the time step has been chosen as unity. 
66:   The position ${\vec r}$ of the particle 
67: is restricted on the cubic lattice 
68: and the lattice spacing is chosen as unity. 
69:   The local equilibrium distribution $f_i^0({\vec r},t)$ is assumed 
70: to be given as 
71: %----------------------------------------------------------------
72: \begin{equation}
73: f_i^0({\vec r},t)
74: = w_i \rho [ 1 + 3 ({\vec c}_i\cdot{\vec u}) 
75:                + {9 \over 2} ({\vec c}_i\cdot{\vec u})^2 
76:                - {3 \over 2} ({\vec u}\cdot{\vec u}) ].
77: \end{equation}
78: %----------------------------------------------------------------
79:   The density $\rho$ and the velocity ${\vec u}$ of the fluid 
80: are given by the sum of the contributions of the particles as 
81: $\rho = \sum_i f_i({\vec r},t)$ and 
82: $\rho {\vec u} = \sum_i f_i({\vec r},t) {\vec c}_i$, respectively. 
83:   Using the multiscale analysis 
84: the Navier-Stokes equation, 
85: which is supposed to be able to describe turbulence, 
86: is derived from the present BGK lattice Boltzmann equation 
87: and the corresponding viscosity $\nu = (1/\omega - 1/2)/3$. 
88: 
89:   In the following 
90: we use a 15-velocity model and ${\vec c}_i$ is chosen as 
91: ${\vec c}_1 = (0,0,0)$, 
92: ${\vec c}_i = (\pm1,0,0),(0,\pm1,0),(0,0,\pm1)$ for $i=2,3,\cdot\cdot\cdot,7$ 
93: and 
94: ${\vec c}_i = (\pm1,\pm1,\pm1)$ for $i=8,9,\cdot\cdot\cdot,15$. 
95:   Then the weight factor $w_i$ for the local equilibrium distribution 
96: is determined as 
97: $w_1 = 2/9$, 
98: $w_i = 1/9$ for $i=2,3,\cdot\cdot\cdot,7$ and 
99: $w_i = 1/72$ for $i=8,9,\cdot\cdot\cdot,15$. 
100:   The relaxation frequency $\omega$ is chosen as $\omega=1.94$. 
101: 
102:   In order to simulate forced turbulence 
103: we add the forcing term ${\vec g}_i\cdot{\vec F}({\vec r},t)$ 
104: due to an applied body force ${\vec F}({\vec r},t)$ 
105: to the right hand side of eq.\ (1). 
106:   In order to reproduce the Navier-Stokes equation 
107: we should choose as ${\vec g}_i = {\vec c}_i/10$ 
108: for the present 15-velocity model.\cite{WG} 
109:   For simplicity we apply a solenoidal force in $y$-direction,\cite{S} 
110: ${\vec F}=(0,F_y(z,t),0)$, and $F_y(z,t)$ is a Gaussian white noise 
111: whose variance $\sigma$ is a function of the lattice coordinate 
112: in $z$-direction, 
113: $\sigma = |0.01 \times \sin(2\pi k_f z / L)|$, 
114: where $L$ is the linear dimension of the simulation region 
115: of cubic box and $L=100$ ($1 \leq x,y,z \leq 100$). 
116:   We have adopted the periodic boundary condition. 
117:   Although we have chosen as $k_f = 4$, 
118: the results of the simulation is insensitive to the choice. 
119: 
120:   In the initial state of the simulation 
121: the fluid density is uniform, $\rho = 1$, and 
122: the spatial distribution of the fluid velocity ${\vec u}$ is Gaussian 
123: with the variance $\sigma = \sqrt{0.0075}$. 
124: 
125:   The Reynolds number $R$ for our simulation\cite{S,WG} is 
126: estimated as $R = \sqrt{ \langle {\vec u}^2 \rangle } L / 2\pi\nu$ 
127: and time-dependent as shown in Fig.\ 1. 
128:   Here $\langle \cdot\cdot\cdot \rangle$ represents the spatial average. 
129:   In our simulation 
130: we can realize a turbulent state for relatively small Reynolds number, 
131: since we add random force. 
132: 
133:   We focus our attention to the time series of 
134: the $y$-component $u_y({\vec r},t)$ of 
135: the fluid velocity ${\vec u}({\vec r},t)$ 
136: as shown in Fig.\ 2, 
137: since our system is anisotropic due to the forcing. 
138:   This turbulent signal is characterized by the exit-time 
139: $\tau(\epsilon;{\vec r})$ at position ${\vec r}$. 
140:   At the exit-time $t=\tau(\epsilon;{\vec r})$ 
141: the first exit satisfying the condition, 
142: $|u_y({\vec r},t_0+t) - u_y({\vec r},t_0)| > \epsilon/2$, 
143: occurs. 
144:   The time $t$ is measured from $t_0$ and $t_0=1000$ in our simulation 
145: regarding the states at the first 1000 time-steps as transitional. 
146: 
147:   The mean of the exit-time, 
148: $\tau(\epsilon) \equiv \langle \tau(\epsilon;{\vec r}) \rangle$, 
149: is related to the $\epsilon$-entropy, $h(\epsilon)$, as\cite{Vul} 
150: %----------------------------------------------------------------
151: \begin{equation}
152: h(\epsilon) \propto {1 \over \tau(\epsilon)}. 
153: \end{equation}
154: %----------------------------------------------------------------
155:   In the region where Kolmogorov's 4/5 law holds\cite{Vul,F} 
156: the velocity difference between two points behaves as 
157: $\langle|{\vec u}({\vec r}+{\vec R},t_0)-{\vec u}({\vec r},t_0)|\rangle 
158:      \propto |{\vec R}|^{1/3}$. 
159:   Under the assumption of Taylor's hypothesis,\cite{Vul,F} 
160: $[{\vec u}({\vec r},t_0+t)-{\vec u}({\vec r},t_0)] 
161:      \sim [{\vec u}({\vec r}+{\vec R},t_0)-{\vec u}({\vec r},t_0)]$, 
162: with $|{\vec R}|=Ut$, 
163: where $U$ is the mean velocity for large scale sweeping, 
164: we obtain 
165: %----------------------------------------------------------------
166: \begin{equation}
167: \langle|{\vec u}({\vec r},t_0+t)-{\vec u}({\vec r},t_0)|\rangle 
168:      \propto t^{1/3},
169: \end{equation}
170: %----------------------------------------------------------------
171: as a mean-field description. 
172:   Thus the definition of the exit-time leads to 
173: $\tau(\epsilon)^{1/3} \propto \epsilon$. 
174:   From eq.\ (3) we can conclude 
175: that $h(\epsilon) \propto \epsilon^{-3}$ 
176: in the above mean-field sense.\cite{Vul} 
177: 
178:   In accordance with the above scaling analysis 
179: our simulation data for $1/\tau(\epsilon)$ shown in Fig.\ 3 
180: have the scaling region for $\epsilon > \epsilon_0$ 
181: where $\epsilon_0 \sim 0.04$. 
182:   Thus by simulation we have established 
183: that $h(\epsilon) \propto \epsilon^{-3}$. 
184:   For $\epsilon < \epsilon_0$, namely for short-time events, 
185: we can not expect the scaling, 
186: since the mean-field assumption for large scale sweeping 
187: is valid for the average of many events and 
188: not applicable to a few events at short time-interval. 
189:   These behaviors are consistent with the experimantal data.\cite{Vul} 
190:   For $\epsilon \ll \epsilon_0$ 
191: the exit in the simulation occurs within the time step. 
192: 
193:   In summary 
194: we have shown that the velocity fluctuation in a forced turbulence 
195: simulated by the lattice Boltzmann method 
196: leads to the $\epsilon$-entropy scaling, 
197: $ h(\epsilon) \propto \epsilon^{-3} $, 
198: expected from the Kolmogorov scaling. 
199:   Although the scaling relation is a robust result 
200: of the mean-field description, 
201: the intermittent aspects of turbulence 
202: are beyond the scope of the present study. 
203: 
204:   This work was supported in part 
205: by a Grand-in-Aid for Scientific Research 
206: from the Ministry of Education, Culture, Sports, Science 
207: and Technology of Japan. 
208: 
209: %----------------------------------------------------------------
210: \begin{references}
211: 
212: \bibitem{N}
213: This paper has been submitted to the Short Note section 
214: of J. Phys. Soc. Jpn.
215: 
216: \bibitem{Vul}
217: M.\ Abel, L.\ Biferale, M.\ Cencini, M.\ Falcioni, D.\ Vergni 
218: and A.\ Vulpiani: 
219: Physica D {\bf 147} (2000) 12. 
220: 
221: \bibitem{S}
222: S.\ Succi: 
223: {\it The Lattice Boltzmann Equation for Fluid Dynamics and Beyond} 
224: (Clarendon Press, Oxford, 2001). 
225: 
226: \bibitem{WG}
227: D.\ A.\ Wolf-Gladrow: 
228: {\it Lattice-Gas Cellular Automata and Lattice Boltzmann Models} 
229: (Springer-Verlag, Berlin, 2000). 
230: 
231: \bibitem{F}
232: U.\ Frisch: 
233: {\it Turbulence} 
234: (Cambridge University Press, Cambridge, 1995). 
235: 
236: \end{references}
237: 
238: \vskip 15pt
239: %----------------------------------------------------------------
240: \begin{figure}
241: \caption{
242: The Reynolds number as a function of time. }
243: \label{fig:1}
244: \end{figure}
245: 
246: \vskip -15pt
247: 
248: \begin{figure}
249: \caption{
250: The $y$-component of the fluid velocity $u_y$ 
251: as a function of time at ${\vec r}=(50,50,50)$. }
252: \label{fig:2}
253: \end{figure}
254: 
255: \vskip -15pt
256: 
257: \begin{figure}
258: \caption{
259: The inverse of the mean exit-time $\tau(\epsilon)$. 
260: Here the dots and straight line represent the simulation data 
261: and $\epsilon^{-3}$ law, respectively. }
262: \label{fig:3}
263: \end{figure}
264: %----------------------------------------------------------------
265: \end{document}
266: %----------------------------------------------------------------
267: \bye
268: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
269: