nlin0105054/govm.tex
1: \documentclass[12pt,a4paper]{article}
2: \usepackage[dvips]{graphics}
3: \usepackage{amssymb}
4: \begin{document}
5: 
6: \begin{center}
7:   {\large\bf Generalized Optimal Velocity Model for Traffic Flow}
8: \end{center}
9: 
10: \vspace{5mm}
11: \begin{center}
12:   Shiro Sawada\footnote{e-mail: sawada@dt.takuma-ct.ac.jp}
13: \end{center}
14: 
15: \begin{center}
16: {\it  Department of Telecommunications, \\Takuma National
17: College of Technology,\\ Mitoyo, Kagawa 769-1192, Japan}
18: \end{center}
19: 
20: \vspace{30mm}
21: 
22: \begin{center}
23: {\bf  Abstract}
24: \end{center}
25: A generalized optimal velocity model is analyzed, where
26: the optimal velocity function depends not 
27: only on the headway of each car but also the headway of 
28: the immediately preceding one. 
29: The stability condition of the model is derived by
30: considering a small perturbation around the homogeneous
31: flow solution.
32: The effect of the generalized optimal velocity function 
33: is also confirmed with
34: numerical simulation, by examining the hysteresis loop in
35: the headway-velocity phase space,
36: and the relation between flow and density of cars.
37: In the model with a specific parameter choice,
38: it is found that an intermediate state 
39: appears for the movement of cars,
40: where the car keeps a certain velocity though
41: the headway is short or long.
42: This phenomenon is different from the ordinary stop-and-go state.
43: 
44: \newpage
45: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
46: \section{Introduction}
47: 
48: Traffic flow problem has been extensively studied
49: from physical point of view.
50: Fluid-dynamical model\cite{fluid}, cellular automaton model
51: \cite{cellular},
52: and the car-following model\cite{car} have been proposed and 
53: analyzed in detail 
54: to understand the mechanism of the traffic congestion on a
55: freeway.
56: 
57: Toward a realistic model which explains the traffic flow dynamics, 
58: the optimal velocity (OV) model proposed by Bando, Hasebe,
59: Nakayama, Shibata, and Sugiyama\cite{ov, ov2}
60: has attracted considerable interest.
61: Based on the second-order differential equations,
62: the model reveals the density pattern formation of
63: the congested flow of traffic without introducing 
64: a time lag caused by the driver's response.
65: 
66: Although the OV model is shown to have a universal structure
67: in spatio-temporal patterns in the congestion, 
68: most of the analyses of the model have been done in the case
69: where the optimal velocity function depends only on 
70: the headway of each car. One of the approaches to generalize the OV 
71: model is that 
72: the backward reference function is introduced\cite{hn}.
73: Another approach to extend the OV model is to take into
74: account the next-nearest-neighbor interaction\cite{nagatani},
75: where the optimal velocity function depends not only on
76: the headway of each car but also on the headway of 
77: the immediately preceding one.
78: The generalized optimal velocity function is determined
79: by taking into account the driver's skill, experience, 
80: and psychological effect, so that it is expected to
81: describe more realistic traffic flow.
82: 
83: The purpose of this paper is to analyze the generalized 
84: optimal velocity model proposed by Nagatani\cite{nagatani,naga}.
85: This paper is organized as follows.
86: In Section 2 the generalized optimal velocity model is reviewed and
87: its stability condition around the homogeneous flow solution is derived
88: without long-wavelength approximation.
89: In Section 3 numerical simulations are carried out, in particular,
90: the hysteresis loop in the phase space and the flow-density
91: relation are examined.
92: In Section 4 we reconsider our model to compare the result
93: with the one in the original model.
94: In the model with a specific parameter choice,
95: it is found that an intermediate state appears for the
96: movement of cars,
97: which is different from the ordinary stop-and-go state.
98: 
99: \section{Generalized Optimal Velocity Model}
100: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
101: 
102: We first consider a dynamical model of the traffic flow
103: given by
104: \begin{equation}
105:   \label{eq:aa}
106:   \ddot x_n(t) = a(V(\Delta x_n(t), \Delta x_{n+1}(t)) - \dot x_n(t)),
107: \end{equation}
108: where $x_n(t)$ is the position of the $n$-th 
109: car at time $t$, $\Delta x_n(t)=x_{n+1}(t)-x_n(t)$
110: represents the headway of the car,
111: and $a$ is the sensitivity.
112: Thus $\Delta x_{n+1}(t)$ is the headway of the immediately
113: preceding car.
114: Here $n=1,2,\cdots,N$ is each car number
115: with $N$ being the total number of vehicles.
116: The driver's sensitivity $a$ is assumed to be independent of $n$.
117: Throughout this paper, we will consider the periodic boundary
118: condition with respect to the coordinate $x_n$ with period $L$.
119: 
120: At first, let us look for an appropriate
121: form of the optimal velocity function
122: to be suitable for our purpose.
123: The driver sometimes pays attention to not only the headway 
124: but also the headway of the immediately preceding one.
125: If the headway of the preceding car is short,
126: the driver assumes that the forward driver decelerates, thus
127: the driver decrease the optimal velocity even though
128: the headway of his car is long enough.
129: On the other hand, 
130: if the headway of the preceding car is long,
131: the driver assumes that the forward driver accelerates, thus
132: the driver increase the optimal velocity even though
133: the headway of his car is short.
134: 
135: Let us look at Figure 1, which describes the original optimal
136: velocity function in $\Delta x_n$, $\Delta x_{n+1}$, and $V$ 
137: space, where the numerical values
138: of axes are not important here.
139: The appropriate form of the function which satisfies the above
140: requirement will be seen in Figure 2.
141: 
142: \begin{figure}[htbp]
143:   \begin{center}
144:   \begin{minipage}[t]{.47\textwidth}
145:   \scalebox{0.55}{\includegraphics{ovfunction.eps}}
146:   \caption{The original optimal velocity function $V(\Delta x_n)$.}
147:   \label{fig:fig1}
148:   \end{minipage}
149:   \begin{minipage}[t]{.47\textwidth}
150:   \scalebox{0.55}{\includegraphics{govfunction.eps}}
151:   \caption{The generalized optimal velocity function
152:     $V(\Delta x_n,\Delta x_{n+1})$.}
153:   \label{fig:fig2}
154:   \end{minipage}
155:   \end{center}
156: \end{figure}
157: 
158: As is already introduced in Ref.\cite{nagatani,naga}, 
159: we can adopt the generalized optimal
160: velocity function of the form
161: \begin{equation}
162:   \label{eq:bb}
163:   V(\Delta x_n, \Delta x_{n+1})
164:   =(1-p) V(\Delta x_n) + p V(\Delta x_{n+1}),
165: \end{equation}
166: where $p$ is assumed to be independent of $n$
167: and satisfies $0\leqq p < \frac{1}{2}$ because the dominant part of
168: the optimal velocity function should be $\Delta x_n$ dependent term.
169: The advantage of the above form is to be able to find the effect
170: of the additional term with varying $p$.
171: In spite of introducing $p$ and $\Delta x_{n+1}$ dependent term,
172: the above model provides exactly the same homogeneous flow 
173: solution as the original one without $p$ dependence.
174: When $p=0$, the model reduces to the original one.
175: According to the original OV function proposed by Bando 
176: et al.\cite{ov}, we will take hyperbolic tangent function
177: of the form
178: \begin{equation}
179:   \label{eq:cc}
180:   V(\Delta x_n)=\tanh(\Delta x_n-2)+\tanh(2).
181: \end{equation}
182: As we can see from eq.(\ref{eq:bb}) with eq.(\ref{eq:cc}), 
183: the generalized optimal
184: velocity function has the required properties.
185: In general, $p$ might depend on $n$ and also depend on time $t$.
186: The numerical simulation can be still performed under these circumstances.
187: 
188: More generally, we can consider the form
189: \begin{equation}
190:   \label{eq:dd}
191:   V(\Delta x_n, \Delta x_{n+1})
192:   =U(\Delta x_n) + W(\Delta x_{n+1}),
193: \end{equation}
194: where $U$ and $W$ have the following properties:
195: (i) monotonically increasing functions,
196: (ii) they have upper bounds, and (iii)
197: they satisfy $|U|> |W|$ as a realistic model.
198: 
199: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
200: Now, let us analyze the generalized OV model given by 
201: eq.(\ref{eq:aa}) with eq.(\ref{eq:dd})
202: \begin{equation}
203:   \label{eq:ee}
204:   \ddot x_n(t) = 
205:   a\left(U(\Delta x_n(t))+ W(\Delta x_{n+1}(t)) - \dot x_n(t)\right).
206: \end{equation}
207: As is easily seen from eq.(\ref{eq:ee}), 
208: they have a homogeneous flow solution
209: \begin{equation}
210:   \label{eq:ff}
211:   x_n^{(0)}(t) = b n + c t,
212: \end{equation}
213: where $b=L/N$ and $c=U(b)+W(b)$.
214: We examine the stability against a small perturbation
215: $y_n(t)$ around the homogeneous flow solution (\ref{eq:ff}).
216: Substituting
217: \begin{equation}
218:   \label{eq:gg}
219:   x_n(t) = x_n^{(0)}(t)+ y_n(t)
220: \end{equation}
221: into eq.(\ref{eq:ee}), the linearized equation to $y_n(t)$ is obtained as
222: \begin{equation}
223:   \label{eq:hh}
224:   \ddot y_n(t) =
225:   a\left( g\Delta y_n(t) + h\Delta y_{n+1}(t) - \dot y_n(t)\right),
226: \end{equation}
227: where $\Delta y_n(t)= y_{n+1}(t) -y_n(t)$, and
228: $g$ and $h$ is the derivative of $U$ and $W$ at $b$, respectively.
229: The solutions to eq.(\ref{eq:hh}) is given by the Fourier series as
230: \begin{equation}
231:   \label{eq:ii}
232:   y_{n,k}(t) = \exp(i\alpha_k n + z t),
233: \end{equation}
234: where $\alpha_k=2\pi k/N$ with $k=1,2,\cdots, N$ and
235: $z$ satisfies
236: \begin{equation}
237:   \label{eq:jj}
238:   z^2 + a z 
239:   -a\left(g(e^{i\alpha_k}-1)+h(e^{2i\alpha_k}-e^{i\alpha_k})\right) =0.
240: \end{equation}
241: The stability condition is to find the condition
242: Re$z<0$ for all modes $\alpha_k$.
243: After a short calculation, it is equivalent to find the relation
244: \begin{equation}
245:   \label{eq:kk}
246:   F(Y)\equiv 
247: 32h^2Y^3+16h(g-3h)Y^2+2\left((g-3h)^2-2ah\right)Y-a(g-h)<0
248: \end{equation}
249: holds for all modes $\alpha_k$, where 
250: we put $Y=\cos^2\frac{\alpha_k}{2}$.
251: 
252: In order to proceed the analysis of stability condition,
253: we will take the optimal velocity function adopted in
254: eq.(\ref{eq:bb}). 
255: Eq.(\ref{eq:kk}) is rewritten by
256: \begin{equation}
257:   \label{eq:ll}
258:   F(Y)=
259:   32p^2f^2Y^3+16p(1-4p)f^2Y^2
260:   +2( (1-4p)^2f^2-2apf)Y -a(1-2p)f <0,
261: \end{equation}
262: where $f=V'(b)$.
263: We can finally find that if the condition
264: \begin{equation}
265:   \label{eq:mm}
266:   f<\frac{a}{2}(1+2p) \quad {\rm and} \quad p\leqq\frac{1}{2}
267: \end{equation}
268: is satisfied, eq.(\ref{eq:ll}) holds for all modes $\alpha_k$.
269: It should be emphasized that the stability condition (\ref{eq:mm})
270: is derived without long-wavelength approximation. 
271: In the long-wavelength approximation, only the former condition
272: in eq.(\ref{eq:mm}) is derived from the stability analysis.
273: The uniform solution is unstable if $f>\frac{a}{2}(1+2p)$ or 
274: $p>\frac{1}{2}$.
275: Comparing the result with the original OV model,
276: we can conclude that 
277: the model is stabilized in
278: the region $ \frac{a}{2}\leqq f< \frac{a}{2}(1+2p)$
279: by the effect of introducing the headway of the preceding car.
280: 
281: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
282: \section{Numerical Simulations}
283: 
284: To convince the analysis of stability condition for the
285: generalized OV model, we will
286: now solve
287: \begin{equation}
288:   \label{eq:nn}
289:   \ddot x_n(t) = a\left((1-p) V(\Delta x_n) 
290:     + p V(\Delta x_{n+1})- \dot x_n(t)\right)
291: \end{equation}
292: numerically.
293: In the simulation, $a=1$ is taken throughout this paper.
294: As the density of cars ($\rho=N/L$) varies and if
295: $f>\frac{a}{2}(1+2p)$ is satisfied,
296: a homogeneous flow becomes unstable and
297: makes a phase transition from free flow to congested one.
298: Looking at the spatio-temporal pattern, the congested patters 
299: propagate backward.
300: These characteristic features in the generalized model are 
301: the same as those appear in the original OV model.
302: 
303: One of the typical features of the model is that 
304: the movement of the car becomes the stop-and-go states in the 
305: congested region and the congested pattern is very stable.
306: It is well understood by examining the hysteresis loop in the
307: headway-velocity phase space.
308: In numerical simulations, we take
309: $N=100$ and $L=200$ as an example, where $f$ takes maximum value,
310: because the congested region is of our interest and the
311: result depends only on the density but not on the number of cars
312: and the circuit length.
313: The initial condition we considered here is the homogeneous
314: flow with small fluctuation, i.e.
315: \begin{equation}
316:   \label{eq:oo}
317:   x_n(0) = b n + y_n(0), \dot x_n(0) = c,
318: \end{equation}
319: where $y_n(0)$ is taken to be an uniform random 
320: distribution between $-0.5$ and $0.5$.
321: 
322: %%% figure 3.  hysteresis loop in phase space
323: \begin{figure}[htbp]
324:   \begin{center}
325:   \scalebox{0.7}{\includegraphics{headway-velocity.eps}}
326:   \caption{The orbit of a car in the 
327: headway-velocity phase space after the organization of
328: the congestion for $p=0$ and $p=0.2$. The dashed curved line 
329: denotes the optimal velocity function.}
330:   \label{fig:fig3}
331:   \end{center}
332: \end{figure}
333: 
334: Figure \ref{fig:fig3} shows the orbit of a particular car in the
335: headway-velocity phase space with 
336: the parameter $p=0,~ 0.2$ as an example. 
337: After about 1,000 time when the generation of
338: the congestion is finished, the shape of the hysteresis 
339: loop obtained in Figure 3 never changes.
340: Furthermore, the shape does not depend on the initial random 
341: distributions.
342: 
343: The numerical result shows that the effect of increasing $p$ 
344: seems to be equivalent to increasing the sensitivity $a$.
345: Of course, this is also guessed from the analysis
346: of the stability condition in eq.(\ref{eq:mm}).
347: However, as will be seen in Section 4, we will find 
348: that the change of 
349: the value of $p$ can not be compensated by rescaling the 
350: sensitivity $a$.
351: 
352: Examining the bottom end point $(\Delta x_{\rm c}, v_{\rm c})
353: =(\Delta x_n, \dot x_n)$
354: and the top end point $(\Delta x_{\rm f}, v_{\rm f})=
355: (\Delta x_n, \dot x_n)$
356: in the hysteresis loop, we can obtain
357: the backward velocity of the congestion given by
358: \begin{equation}
359:   \label{eq:pp}
360:   V_{\rm back}= \frac{v_{\rm f}\Delta x_{\rm c} - 
361: v_{\rm c} \Delta x_{\rm f}}{\Delta x_{\rm f}-\Delta x_{\rm c}}.
362: \end{equation}
363: Numerical simulation shows that
364: the backward velocity of the congestion increases as
365: $p$ increases. 
366: We have also confirmed by simulating spatio-temporal patterns.
367: The numerical results are listed in Table 1.
368: 
369: \begin{table}[htbp]
370:   \begin{center}
371: \begin{tabular}[htbp]{|c|c|c|c|c|c|}
372: \hline
373: $p$&$\Delta x_{\rm c}$&$v_{\rm c}$&$\Delta x_{\rm f}$
374: &$v_{\rm f}$&$V_{\rm back}$\\
375: \hline
376: 0.0 & 0.32274 & 0.03152 & 3.67726 & 1.89653 & 0.14791 \\
377: \hline
378: 0.1 & 0.62051 & 0.08319 & 3.37945 & 1.84485 & 0.31302 \\
379: \hline
380: 0.2 & 0.91196 & 0.16787 & 3.08804 & 1.76019 & 0.49945 \\
381: \hline
382: 0.3 & 1.18567 & 0.29206 & 2.81434 & 1.63600 & 0.68632 \\
383: \hline
384: 0.4 & 1.46814 & 0.47750 & 2.53275 & 1.45136 & 0.86548 \\
385: \hline
386: \end{tabular}
387:     \caption{Numerical data for $p=0,~0.1~,0.2~,0.3,~0.4.$}
388:     \label{tab:table1}
389:   \end{center}
390: \end{table}
391: 
392: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
393: 
394: Another important problem is to investigate the relation
395: between the flux and the density, which is called fundamental diagram.
396: The density $\rho$ of the cars is defined by $N/L$, where we choose
397: $L$=200 and vary $N$ from 10 and 300 in the simulation.
398: The flux $Q$ is defined by the number of cars passing by a position
399: per unit time.
400: The data was accumulated and averaged over during 20,000 time 
401: after first 1000 time. 
402: Numerical results are plotted in Figure \ref{fig:fig4}
403: for $p=0,~0.1,~0.2,~0.3$.
404: Here we omitted to plot the data which overwrites the other data.
405: 
406: %%% figure 4.  fundamental diagram, flow and density
407: \begin{figure}[htbp]
408:   \begin{center}
409:   \scalebox{0.9}{\includegraphics{flow-density.eps}}
410:   \caption{Flux-density fundamental diagram
411:     for $p=0,~0.1~,0.2~,0.3$.}
412:   \label{fig:fig4}
413:   \end{center}
414: \end{figure}
415: 
416: In the homogeneous flow, the relation between flux $Q$
417: and density $\rho$ is given by
418: \begin{equation}
419:   \label{eq:qq}
420:   Q=\rho V\left(\frac{1}{\rho}\right)
421:   = \rho\left(\tanh\left(\frac{1}{\rho}-2\right) + \tanh(2)\right).
422: \end{equation}
423: In Figure \ref{fig:fig4}, it is represented by the dashed curved line.
424: The numerical results agree with this line in the homogeneous flow
425: region.
426: 
427: As is discussed in Ref.\cite{ov2}, the flux-density relation in the
428: congested flow is evaluated from the data in Table \ref{tab:table1}.
429: In the congested flow region, the relation between the density $\rho$
430: and the flux $Q$ is given by
431: \begin{equation}
432:   \label{eq:rr}
433:   Q = \frac{v_{\rm f}-v_{\rm c}}
434: {\Delta x_{\rm f}-\Delta x_{\rm c}} -V_{\rm back} \rho.
435: \end{equation}
436: Substituting the values in Table \ref{tab:table1} 
437: into eq.(\ref{eq:rr}), we obtain the $Q$-$\rho$ relation as in
438: the Table \ref{tab:table2}.
439: \begin{table}[htbp]
440:   \begin{center}
441: \begin{tabular}[htbp]{|c|c|}
442: \hline
443:  $p$ &  $Q$-$\rho$ relation\\
444: \hline
445: 0.0 & $Q=0.55597-0.14792 ~\rho$ \\
446: \hline
447: 0.1 & $Q=0.63853-0.31302 ~\rho$ \\
448: \hline
449: 0.2 & $Q=0.73174-0.49945 ~\rho$ \\
450: \hline
451: 0.3 & $Q=0.82518-0.68632 ~\rho$ \\
452: \hline
453: 0.4 & $Q=0.91475-0.86548 ~\rho$ \\
454: \hline
455: \end{tabular}
456:     \caption{Flux-density relation in congested flow
457:       for $p=0,~0.1,~0.2,~0.3,~0.4$.}
458:     \label{tab:table2}
459:   \end{center}
460: \end{table}
461: These lines are plotted
462: in Figure \ref{fig:fig4} with the fundamental diagram. 
463: In Figure 4 the vertical lines represent the boundaries for
464: the stability condition given by eq.(\ref{eq:mm})
465: in the case of $p=0,~0.1,~0.2,~0.3$, where the same dashed
466: line is used as the one used to draw the predicted line
467: in eq.(\ref{eq:rr}).
468: The numerical results for various $p$ values are good
469: agreement with the predicted lines in Table \ref{tab:table2}.
470: 
471: We will summarize
472: the effect of the $\Delta x_{n+1}$ dependent term 
473: in fundamental diagram.
474: In the homogeneous flow region, there is no effect 
475: in the flux-density relation, 
476: because $p$ dependence in the generalized OV function disappears
477: in the case of homogeneous flow, as is seen in eq.(\ref{eq:bb}).
478: In the congested region, the flow 
479: increases as $p$ increases if $\rho<\frac{1}{2}$
480: and inversely the flow decrease as $p$ increase if $\rho>\frac{1}{2}$.
481: The reason is as follows.
482: When the density is low, the average of the 
483: headway is long.
484: If $\Delta x_{n+1}$ is long,
485: the larger value of the velocity
486: than that of the case without $\Delta x_{n+1}$ dependence
487: is allowed.
488: Hence the flow increases by taking into account the 
489: headway of the immediately preceding car.
490: Inversely, when the density is high, i.e. $\rho>\frac{1}{2}$,
491: the small value of the velocity is taken compared 
492: with the case of $p=0$,
493: because $\Delta x_{n+1}$ is short.
494: Hence the flow decreases by taking into account the 
495: headway of the immediately preceding car.
496: 
497: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
498: \section{Rescaled Model}
499: 
500: As far as the stability condition of traffic flow is concerned, 
501: our analytical result of eq.(\ref{eq:mm}) means that the effect of
502: the $\Delta x_{n+1}$ dependent term can be compensated
503: by rescaling the sensitivity $a$.
504: Thus we can rewrite the generalized OV model as
505: \begin{equation}
506:   \label{eq:tt}
507:   \ddot x_n(t) = a\left(\frac{1-p}{1+2p} V(\Delta x_n) 
508:     + \frac{p}{1+2p} V(\Delta x_{n+1})- \frac{1}{1+2p}\dot x_n(t)\right).
509: \end{equation}
510: Now the stability condition of the above model is 
511: given by $f<\frac{a}{2}$ which is independent of $p$.
512: By investigating the hysteresis loop in this model with
513: various values of $p$, we can clarify the effect of the
514: headway of the preceding car.
515: Numerical simulation can be performed in the same way as
516: in Section 3.
517: Obtained numerical data for the headway and velocity of a particular car 
518: is plotted in Figure \ref{fig:fig5}.
519: 
520: %%% figure Hysteresis loop
521: \begin{figure}[htbp]
522:   \begin{center}
523:   \scalebox{0.7}{\includegraphics{r-headway-velocity.eps}}
524:   \caption{Hysteresis loops for $p=0$ and $p=0.2$.}
525:   \label{fig:fig5}
526:   \end{center}
527: \end{figure}
528: 
529: We can find that the effect of the $\Delta x_{n+1}$ dependent term 
530: can not be compensated by rescaling the sensitivity $a$.
531: The effect of $p$ is now easily understood.
532: Compared with $p=0$, shorter value of the 
533: minimum of the headway is allowed,
534: and also longer value of the maximum of the headway 
535: is allowed.
536: Furthermore, when the car accelerates, 
537: larger value of the velocity can be taken even if
538: the headway is short.
539: Inversely, when the car decelerates,
540: smaller optimal velocity is taken even if
541: the headway is long enough.
542: 
543: The above characteristic feature in the phase space
544: holds for various $p$ values as long as $p<\frac{1}{2}$.
545: If we take $p>\frac{1}{2}$, the numerical simulation shows that
546: the cars take over the cars ahead, thus it is not realistic.
547: Of course, a room to change the form of OV function is left,
548: but we will not consider it here.
549: 
550: It is found that the model give by eq.(\ref{eq:tt})
551: has a different feature in
552: the unstable state when we take $p=\frac{1}{2}$.
553: Figure \ref{fig:p-hv} shows the hysteresis loop for $p=\frac{1}{2}$
554: after some relaxation time, e.g. about $10^8$ time in the
555: simulation.
556: More precisely, two independent hysteresis loops appear
557: for two successive cars.
558: In Figure \ref{fig:p-hv}, the line and the dotted line are 
559: the hysteresis loops
560: for the odd car number and even car number, respectively.
561: This hysteresis loop is well understood when the orbit of the
562: car is shown in the $(\Delta x_n, \Delta x_{n+1}, \dot x_n)$ phase space.
563: \begin{figure}[htbp]
564: \begin{center}
565:   \begin{minipage}[t]{.47\textwidth}
566:   \scalebox{0.45}{\includegraphics{p-hv.eps}}
567:   \caption{Hysteresis loops for $p=\frac{1}{2}$ 
568: in $(\Delta x_n, \dot x_n)$ space.}
569:   \label{fig:p-hv}
570:   \end{minipage}
571:   \begin{minipage}[t]{.47\textwidth}
572:   \scalebox{0.55}{\includegraphics{p-hhv.eps}}
573:   \caption{Hysteresis loops for $p=\frac{1}{2}$
574: in $(\Delta x_n, \Delta x_{n+1}, \dot x_n)$ space.}
575:   \label{fig:p-hhv}
576:   \end{minipage}
577: \end{center}
578: \end{figure}
579: Figure \ref{fig:p-hhv} shows the orbit of two successive cars in
580: the phase space.
581: The simulation
582: shows that all of the vehicles sweep the same orbit as the one
583: in Figures \ref{fig:p-hv} and \ref{fig:p-hhv}.
584: We have examined the model with various initial configurations
585: such as random distributions, different number of cars $N$ 
586: and circuit length $L$.
587: The above result does not depend on the initial distributions
588: and the parameters.
589: 
590: \begin{figure}[htbp]
591: \begin{center}
592:   \begin{minipage}[t]{.47\textwidth}
593:   \scalebox{0.5}{\includegraphics{p-nh.eps}}
594:   \caption{Headways for all vehicles at $10^8$ time with
595: $N=100$, $L=200$.}
596:   \label{fig:p-nh}
597:   \end{minipage}
598:   \begin{minipage}[t]{.47\textwidth}
599:   \scalebox{0.5}{\includegraphics{p-nv.eps}}
600:   \caption{Velocities for all vehicles at $10^8$ time with
601: $N=100$, $L=200$.}
602:   \label{fig:p-nv}
603:   \end{minipage}
604: \end{center}
605: \end{figure}
606: 
607: 
608: Figure \ref{fig:p-nh} and \ref{fig:p-nv} show the 
609: headways and the velocities for all vehicles,
610: respectively, in the case of $N=100$ and $L=200$ at 
611: $10^8$ time, starting
612: with arbitrary random distributions with zero velocities.
613: Numerical simulations show that the characteristic feature
614: does not depend on the initial distributions and the parameters.
615: %%From the numerical simulations, we can extract the common 
616: %%feature which appears in the model give by eq.(\ref{eq:tt})
617: %%with the specific parameter $p=\frac{1}{2}$.
618: We can find that an intermediate state appears
619: for the movement of cars,
620: which is different from the ordinary stop-and-go state.
621: Numerically, the state corresponds to the points
622: $(\Delta x_n, \Delta x_{n+1}, \dot x_n)=(3.677, ~0.323, ~0.964)$
623: and
624: $(\Delta x_n, \Delta x_{n+1}, \dot x_n)=(0.323, ~3.677, ~0.964)$
625: in the phase space as is given in Figure \ref{fig:p-hhv}.
626: Thus the intermediate state means that the car can keep a certain 
627: velocity (0.964) though the headway is short (0.323) or long (3.677)
628: with holding $\Delta x_n + \Delta x_{n+1} =4$.
629: The appearance of the intermediate state is universal in the
630: sense that it does not depend on
631: the initial distributions,
632: the number of cars $N$ and the length $L$.
633: It should be noted that this state never appears if $p<\frac{1}{2}$.
634: 
635: This characteristic feature is understood by considering the
636: equation for the headway which is given by
637: \begin{equation}
638:   \label{eq:vv}
639: \Delta \ddot x_n = \frac{a}{1+2p}\left(
640: (1-2p)V(\Delta x_{n+1}) + p V(\Delta x_{n+2}) -
641: (1-p)V(\Delta x_n) - \Delta \dot x_n\right).
642: \end{equation}
643: If $p=\frac{1}{2}$ is chosen, the first term in the
644: right hand side of eq.(\ref{eq:vv}) vanishes and hence
645: $\Delta x_{n+1}$ dependence disappears.
646: One might expect that the intermediate state is
647: understood by the nonlinear analysis near the critical
648: point $a=a_c$.
649: Following the analysis by Komatsu and Sasa\cite{ks},
650: the equation for the headway is derived near
651: the critical point and
652: the modified Korteweg-de Vries(MKdV) equation and its higher-oder
653: corrections is obtained by introducing
654: a small scaling parameter $\epsilon=\sqrt{(a_c-a)/a_c}$.
655: However, the equation near the critical point 
656: in the case of $p=\frac{1}{2}$ is 
657: exactly the same as the one in the case of $p=0$ up to
658: the order $\epsilon$.
659: Because the difference in eq.(\ref{eq:vv}) with
660: $p=0$ and $p=\frac{1}{2}$ near the critical point
661: is that the Fourier mode with $p=\frac{1}{2}$ 
662: is just twice that of the other one with $p=0$.
663: Since our analysis is performed far from the critical point,
664: the higher-oder corrections or non-perturbative effect
665: should be considered to understand the intermediate state.
666: 
667: 
668: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
669: \section{Summary and Discussion}
670: We have analyzed the stability of the generalized optimal 
671: velocity model where the optimal velocity function
672: depends not only on the headway of each car but also on the
673: headway of the immediately preceding one.
674: The effect of the newly introduced $\Delta x_{n+1}$ dependent term
675: was examined by numerical simulation.
676: In particular, the hysteresis loop in the phase space
677: and the flux-density relation were examined in detail
678: by taking the various values of the parameter.
679: 
680: We found that the effect of the $\Delta x_{n+1}$ dependent term 
681: can not be compensated by rescaling the sensitivity $a$.
682: In the model with the specific parameter choice $p=\frac{1}{2}$,
683: we found that the intermediate state appears for the
684: movement of cars,
685: which is different from the ordinary stop-and-go state.
686: Numerical simulation shows that the appearance of
687: the intermediate state is universal
688: because it does not depend on the initial conditions.
689: 
690: We would expect that the model is related with the exact solution
691: given by Jacobi's elliptic function\cite{exact}.
692: It is interesting to examine the
693: difference-differential equation
694: \begin{equation}
695:   \label{eq:ww}
696: \dot x_n(t+\tau) = V(\Delta x_n(t), \Delta x_{n+1}(t)).
697: \end{equation}
698: The details of the numerical simulation and the
699: analysis of the underlining mathematical structure of
700: the generalized optimal velocity model are under study.
701: 
702: \begin{center}
703:   {\bf Acknowledgment}
704: \end{center}
705: 
706: The author would like to thank S. Tadaki for informing of
707: the Refs.\cite{nagatani,naga}.
708: The part of the numerical computation in this work was carried out at
709: Yukawa Institute Computer Facility.
710: 
711: \begin{thebibliography}{99}
712: \bibitem{fluid}M.J.Lighthill and G.B. Witham, Proc. R. Soc. London
713: A {\bf 299} (1955) 317.
714: \bibitem{cellular}S. Wolfram, Rev. Mod. Phys. {\bf 55} (1983) 601;
715: S. Wolfram, {\it Theory and Applications of Cellular Automata}
716: (World Scientific, Singapore, 1986);
717: K. Nagel and M. Schreckenberg, J. Phys. I France {\bf 2} (1992) 2221;
718: K. Nagel and H.J. Herrmann, Physica A {\bf 199} (1993) 254;
719: S. Yukawa, M. Kikuchi and S. Tadaki, J. Phys. Soc. Japan {\bf 63} (1994) 3609;
720: K. Nagel, Phys. Rev. E {\bf 53} (1996) 4655.
721: \bibitem{car}L.A. Pipes, J. Appl. Phys. {\bf 24} (1953) 274;
722: G.F. Newell, J. Opns. Res. Soc. {\bf 9} (1961) 209;
723: D.C. Gazis, R. Herman and R.W. Rothery, J. Opns. Res. Soc. {\bf 9}
724: (1961) 545.
725: \bibitem{ov}
726: M. Bando, K. Hasebe, A. Nakayama, A. Shibata,
727: and Y. Sugiyama, Phys. Rev. E {\bf 51} (1995) 1035;
728: M. Bando, K. Hasebe, A. Nakayama, A. Shibata,
729: and Y. Sugiyama, Japan J. Indust. Appl. Math, {\bf 11} (1994) 203.
730: \bibitem{ov2} M. Bando, K. Hasebe, A. Nakayama, A. Shibata,
731: and Y. Sugiyama, J. Phys. I France {\bf 5} (1995) 1389.
732: \bibitem{hn} H. Hayakawa and K. Nakanishi,
733: Phys. Rev. E {\bf 57} (1998) 3839. 
734: \bibitem{nagatani} T. Nagatani, Phys. Rev. E {\bf 60} (1999) 6395.
735: \bibitem{naga} T. Nagatani and M. Muramatsu,
736: Proceedings of the 7th Symposium on Simulation of Traffic Flow
737: (in Japanese).
738: \bibitem{ks} T.S. Komatsu and S. Sasa, Phys. Rev. E {\bf 52} (1995) 5574.
739: \bibitem{exact}
740: K. Hasebe, A. Nakayama and Y. Sugiyama,
741: Phys. Lett. A {259} (1999) 135;
742: Y. Igarashi, K. Itoh and K. Nakanishi,
743: J. Phys. Soc. Japan, {\bf 68} (1999) 791;
744: Y. Igarashi, K. Itoh, K. Nakanishi, K. Ogura, K. Yokokawa,
745: patt-sol/9908002;
746: K. Nakanishi, patt-sol/9909005.
747: \end{thebibliography}
748: 
749: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
750: \end{document}
751: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
752: