Physics Letters A 373 (2009) 29172921

Contents lists available at ScienceDirect

Physics Letters A
www.elsevier.com/locate/pla

Freezing transition in bi-directional CA model for facing pedestrian trac

Takashi Nagatani
Department of Mechanical Engineering, Shizuoka University, Hamamatsu 432-8561, Japan

a r t i c l e

i n f o

Article history:
Received 25 March 2009
Received in revised form 17 May 2009
Accepted 8 June 2009
Available online 13 June 2009
Communicated by A.R. Bishop
PACS:
05.70.Fh
89.40.+k
05.90.+m

a b s t r a c t
We present a bi-directional cellular automaton (CA) model for facing trac of pedestrians on a wide
passage. The excluded-volume effect and bi-directionality of facing trac are taken into account. The
CA model is not stochastic but deterministic. We study the jamming and freezing transitions when
pedestrian density increases. We show that the dynamical phase transitions occur at three stages with
increasing density. There exist four trac states: the free trac, jammed trac 1, jammed trac 2, and
frozen state. At the frozen state, all pedestrians stop by preventing from going ahead each other. At three
transitions, the pedestrian ow changes from the free trac through the jammed trac 1 and jammed
trac 2, to the frozen state.
2009 Elsevier B.V. All rights reserved.

Keywords:
Pedestrian ow
Facing trac
Trac ow
Phase transition
Cellular automaton
Freezing

1. Introduction
Recently, pedestrian and vehicular trac ows have attracted
considerable attention [15]. Many observed dynamical phenomena in pedestrian and trac ows have been successfully reproduced with physical methods. The pedestrian ow dynamics is
closely connected with the self-driven many-particle system [6].
It has also encouraged physicists to study evacuation processes by
self-driven many-particle models [713]. The pedestrian and vehicular trac models have been applied to the trac ow of such
mechanical mobile objects as robots [14,15].
The typical pedestrian ows have been simulated by the use of
a few models in two-dimensional space: the lattice-gas model of
biased-random walkers [1116], the molecular dynamic model of
active walkers [6,10,17], and the cellular automaton model [7,8].
Their models are not deterministic but stochastic. Their models
are described in two-dimensional space. The molecular dynamic
model of active walkers is described by the behavioral (or generalized) force on two-dimensional off-lattice. The lattice-gas model
of biased-random walkers and the CA model are described by
stochastic rules on the square lattice. Helbing et al. have found that
the freezing by heating occurs in the facing pedestrian trac by

E-mail address: tmtnaga@ipc.shizuoka.ac.jp.

0375-9601/$ see front matter 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.physleta.2009.06.017

the use of the molecular dynamic model of active walkers [17].

By using the lattice gas model of biased-random walkers, Muramatsu et al. have found independently that the freezing transition
occurs from the free trac to the frozen (stopping) state when the
pedestrian density is higher than the critical value [16]. The freezing transition in the facing pedestrian trac has been studied by
some researchers [18,19].
In the jamming transition, pedestrian ow in the crowd
changes from the free trac to the jammed trac in which pedestrians are distributed heterogeneously and move slowly. In the
freezing transition, pedestrian ow change to the frozen state
in which all pedestrians cannot move by preventing from going
ahead each other. The analytical works are unknown for the facing
pedestrian ow. The pedestrian ow has been investigated by the
numerical simulation of the stochastic models on two-dimensional
space. It is not easy to analyze the two-dimensional stochastic
models because the dynamical behavior is complex. However, the
one-dimensional deterministic CA models have not been proposed
for facing trac of pedestrians until now.
In this Letter, we present the one-dimensional, deterministic,
and bi-directional CA model for the facing pedestrian trac. We
study the dynamical states and dynamical phase transitions in
the model of facing pedestrian trac. We show that there exist
four pedestrian states and the jamming and freezing transitions
occur when pedestrian density increases. We show that the bi-

2918

T. Nagatani / Physics Letters A 373 (2009) 29172921

directional CA model reproduces the Burgers CA model of unidirectional multi-lane trac in the limit of no facing pedestrians.

N E (i , t + 1) = N E (i , t ) + min N E (i 1, t ), M N E (i , t )

2. Bi-directional CA model

When M = 1 in Eq. (3), the Burgers CA model (3) reduces to the

rule-184 CA.

We consider the facing (bi-directional) trac of pedestrians on

a wide passage. There exist two kinds of walkers on the passage:
the one is the walkers moving to the east and the other the walkers moving to the west. The walker moving to the east (or west)
interacts highly with the other walkers in the front. When the density of walkers ahead is higher, the current decreases more because
the movement of walkers will be prevented by other walkers.
We consider the one-dimensional approximation for twodimensional facing pedestrian trac. We approximate the facing
trac (counter ow) on two-dimensional lattice as that on onedimensional lattice because walkers to east or to west move unidirectionally on the average. M walkers can exist on a cell (site) at
its maximum. This means that the passage consists of M lanes for
walkers. The width of passage is M.
We dene the number that walkers to east (to west) exist on
cell (site) i at time t as N E (i , t ) ( N W (i , t )). The states of walkers to east are updated in parallel at every odd discrete time step,
whereas those of walkers to west are updated in parallel at every even discrete time step. We apply the conservation law of
walker number N E (i , t ) ( N W (i , t )) to the facing trac. The number N E (i , 2t + 1) of walkers to east existing on site i at time 2t + 1
is described by the following:

N E (i , 2t + 1)

= N E (i , 2t 1)


+ min N E (i 1, 2t 1), M N E (i , 2t 1) N W (i , 2t )

min N E (i , 2t 1),

M N E (i + 1, 2t 1) N W (i + 1, 2t ) .
(1)
The number N W (i , 2t + 2) of walkers to west existing on site i
at time 2t + 2 is described by the following:

N W (i , 2t + 2)

= N W (i , 2t )


+ min N W (i + 1, 2t ), M N E (i , 2t + 1) N W (i , 2t )


min N W (i , 2t ), M N E (i 1, 2t + 1) N W (i 1, 2t ) ,
(2)
where min[ A , B ] is the minimal function: min[ A , B ] = A if A < B
and min[ A , B ] = B if A > B. The second term on the right hand in
Eq. (1) represents the inow of a walker to east from site i 1 to
site i between 2t and 2t + 1. The third term represents the outow
of a walker to east from site i to site i + 1 between 2t and 2t + 1.
Similarly, the second and third terms of Eq. (2) represents the inow and outow of a walker to west on site i between 2t + 1 and
2t + 2.
The excluded-volume effect is taken into account via
min[ N E (i 1, t ), M N E (i , t ) N W (i , t )]. M N E (i , t ) N W (i , t )
represents the unoccupied space at site i and time t. If N E (i
1, t ) > ( M N E (i , t ) N W (i , t )), ( M N E (i , t ) N W (i , t )) walkers
of N E (i 1, t ) walkers can move to the unoccupied space.
Eqs. (1) and (2) are a couple of nonlinear difference equations.
It is not easy to obtain the analytical solution but possible to obtain the numerical solution.
In the limit of no walkers to west, Eq. (1) reduces to the Burgers
CA model proposed by Nishinari and Takahashi [20]:



min N E (i , t ), M N E (i + 1, t ) .

(3)

3. Simulation and result

We carry out the numerical simulation for bi-directional CA
model described by Eqs. (1) and (2). The boundaries are periodic.
We consider the following initial condition. Walkers to east and to
west distribute uniformly on the passage, respectively. The initial
condition is described by

N E (i , 0) = N E ,0

and

N W (i , 0) = N W ,0 .

(4)

The densities of walkers to east and to west are dened, respectively, as E = N E ,0 / M and W = N W ,0 / M.
3.1. Current-density diagram
We study the dependence of current J E / M of walkers to east
on initial density N E ,0 / M of walkers to east under the initial condition of N W ,0 = const.
If passage width M is suciently large, the facing trac shows
a sharp transition and the freezing transition depends little on M.
Because the boundary is periodic, passage length L has little effect
on the trac when L is large. The width and length of the passage
are set as M = 200 and L = 100.
We add a small perturbation to initial condition (4):
N E (50, 0) = N E ,0 1 and N E (60, 0) = N E ,0 + 1. Fig. 1(a) shows the
plots of current J E / M against initial density N E ,0 / M under the
initial condition of N W ,0 = 25. Circles indicate the simulation result. The current increases linearly with density until point a, then
decrease from point a to pint b, decrease abruptly and discontinuously at point b, and become zero at point c. Thus, the slope
of current changes discontinuously at three transition points a, b,
and c. The current becomes zero at transition point c. All walkers
cannot move when the density is higher than c (the density at
transition point c). The slopes of segment a b and segment b c
are 1.
We study the effect of the initial perturbation on the facing
trac. Fig. 1(b) shows the plots of current J E / M against initial density N E ,0 / M for perturbation N E (50, 0) = N E ,0 1 and
N E (70, 0) = N E ,0 + 1, where passage width M = 200, passage
length L = 100, and N W ,0 = 25. The density of transition point
b is higher than that in Fig. 1(a). Also, the density of transition
point c is higher than that in Fig. 1(a). Transition points b and c
changes with the perturbation. Fig. 1(c) shows the plots of current
J E / M against initial density N E ,0 / M for no perturbations. The initial condition in the case of no perturbations is given by Eq. (4).
The current increases linearly with density, reaches its maximum
value at point a, and then decreases from point a with increasing density. The transition points b and c do not appear but the
current-density diagram shows only one transition point a. The
density of transition point a does not change with any perturbations.
When the density is less than that of transition point a, the
current is given by

J E = N E ,0

for N E ,0 / M < a ,

(5)

where a is the density at transition point a. When the density is

higher than a , the current is given by

J E = M N E ,0 N W ,0

for N E ,0 / M > a .

At the rst transition point, Eq. (5) equals to Eq. (6).

(6)

T. Nagatani / Physics Letters A 373 (2009) 29172921

2919

(a)

(b)

(c)

(d)

Fig. 1. (a) Plots of current J E / M against initial density N E ,0 / M under the initial condition of N W ,0 = 25 where passage width is M = 200 and passage length L = 100. Circles
indicate the simulation result. (b) Plots of current against initial density for the initial condition with a different perturbation from (a). (c) Plots of current against initial
density for no perturbations. (d) Plots of current against initial density for the uni-directional ow with no walkers to west.

N E ,0 = M N E ,0 N W ,0 .
Therefore, transition point a is given by

a = ( M N W ,0 )/2M .

(7)

For N W ,0 = 25 and M = 200, the rst transition point a = 0.4375

is derived analytically. This analytical value is consistent with that
of the simulation result in Fig. 1(a)(c). Also, the current-density
curve on segment a b agrees with Eq. (6).
For comparison, we show the current-density diagram in
Fig. 1(d) for the unidirectional ow when there are no walkers
to west: N W ,0 = 0. The current increases with density, reaches
its maximum value at point a, and then decreases with increasing density. It is known that the jamming transition occurs from
the free trac to the jammed trac at point a. In the free trac,
all walkers move with their maximum speed, while the stop- and
go-waves occur in the jammed trac.
3.2. Trac states
We study the trac states in the facing pedestrian trac. The
pedestrian trac is classied into four trac states 14. Trac
state 1 for 0 < E  a is into the free trac because pedestrians walk freely at their maximum speed. Trac state 2 for
a < E  b is into the jammed trac with stop- and go-waves.
Trac states 1 and 2 are found in the unidirectional ow. We
study the number proles for trac state 3 in b < E  c
and trac state 4 in E > c . Fig. 2(a) shows the plots of numbers N E (i , t ) and N W (i , t ) against position i at E = 120/200 in
trac state 3 where M = 200, N E ,0 = 120, and N W ,0 = 25. The
number of walkers to east is indicated by large circles, while

the number of walkers to west is represented by small circles.

Walkers to east collide with those to west at point a. The kink
antikink (pulse) proles are formed. The density of walkers to
east is low in the region of a < i < c and high in the region
of i  a and i  c. The density of walkers to west is high in
the region of a < i < b and low in the region of i  a and
i  b. Walkers take either high-density value or low-density value.
The low density is not zero but a positive value. Both walkers
move slowly. The second transition point b is that of the jamming transition between the jammed states 1 and 2. The kink
antikink proles are stationary and do not change with time. The
colliding point a is also stationary. This stationary kinkantikink
solution in trac state 3 does not occur in the unidirectional
ow.
Fig. 2(b) shows the plots of numbers N E (i , t ) and N W (i , t )
against position i at E = 140/200 in trac state 4 where M =
200, N E ,0 = 140, and N W ,0 = 25. Walkers to east collide with those
to west at point a. The kinkantikink (pulse) proles are formed.
The density of walkers to east is zero in the region of a < i < c and
maximum in the region of i  a and i  c. The density of walkers
to west is maximum in the region of a < i < b and zero in the
region of i  a and i  b. Walkers have either zero density or maximum density. The lowest density is zero and the highest density is
one. Both walkers do not move. The kinkantikink proles are stationary and do not change with time. The colliding point a is also
stationary. This stationary kinkantikink solution does not occur in
the unidirectional ow. The kinkantikink solution with zero and
one densities means the frozen state. Therefore, the freezing transition between the jammed state 2 and frozen state occurs at third
transition point c. The frozen state in Fig. 2(b) is denitely different from the jammed state 2 in Fig. 2(a).

2920

T. Nagatani / Physics Letters A 373 (2009) 29172921

(a)
(a)

(b)

(b)
Fig. 2. (a) Plots of numbers N E (i , t ) and N W (i , t ) against position i at E = 120/200
in trac state 3 where M = 200, N E ,0 = 120, and N W ,0 = 25. The number of walkers to east is indicated by large circles, while the number of walkers to west is
represented by small circles. (b) Plots of numbers N E (i , t ) and N W (i , t ) against position i at E = 140/200 in trac state 4.

3.3. Case of N E ,0 = N W ,0
We study the case of N E ,0 = N W ,0 . Fig. 3(a) shows the plots
of current J E / M against initial density N E ,0 / M where passage
width M = 200 and passage length L = 100. Circles indicate the
simulation result. The current increases linearly with density until
point a, then decrease from point a to pint b, and become zero at
point c. Thus, the slope of current changes discontinuously at three
transition points a, b, and c. The current becomes zero at transition
point c. All walkers cannot move when the density is higher than
c (the density at transition point c). The slope of segment b c
are 2. Fig. 3(b) shows the plots of current J E / M against initial
density N E ,0 / M for no perturbations. The current increases linearly
with density, reaches its maximum value at point a, and then decreases from point a with increasing density. The transition points
b and c do not appear but the current-density diagram shows only
one transition point a. The density of transition point a does not
change with any perturbations. In the case of N E ,0 = N W ,0 , the dynamical phase transitions occur at three stages, similarly to those
in Fig. 1.
3.4. Discussion
We consider the necessary conditions for the occurrence of
the observed four states. For the occurrence of four distinct trafc states, it is necessary that (1) pedestrians move with facing,
(2) the pedestrian density varies from zero to one, (3) passage

Fig. 3. (a) Plots of current J E / M against initial density N E ,0 / M for the case of
N E ,0 = N W ,0 where passage width M = 200 and passage length L = 100. Circles
indicate the simulation result. (b) Plots of current against initial density for no perturbations.

width M is suciently large, and (4) the initial density distribution is not homogeneous. Especially, condition (4) is necessary for
the occurrence of the frozen state. If the initial density distribution
is homogeneous, the freezing transition does not occur and also
the frozen state does not appear.
Until now, it has been known that the freezing transition occurs
in two-dimensional and stochastic models [16,17]. The present
model is the rst example exhibiting the freezing transition in
one-dimensional and deterministic CA models. The freezing transition is similar to that shown in the two-dimensional facing trac
model of pedestrians [16,17]. The freezing transition point E =
0.3 is nearly consistent with that obtained by the two-dimensional
stochastic CA model in the case of N E ,0 = N W ,0 [16]. Thus, the
present model is a good approximation of the two-dimensional CA
model.
In real pedestrian trac, it is important to avoid the freezing
transition in the crowd because the frozen state of pedestrians induces frequently the falling down one upon another.
4. Summary
We have presented the bi-directional cellular automaton (CA)
model for facing pedestrian trac. The bi-directional CA model is
one-dimensional and deterministic. We have studied the dynamical
phase transition and trac states. We have found that two jamming transitions and a freezing transition occur. We have shown
that there exist four trac states: the free trac, jammed trac 1
and jammed trac 2, and frozen state.
The present model is the rst one of the deterministic CA models for the facing pedestrian trac. It will be useful to analyze the
complex behavior of pedestrian trac.

T. Nagatani / Physics Letters A 373 (2009) 29172921

References
[1]
[2]
[3]
[4]
[5]

T. Nagatani, Rep. Prog. Phys. 65 (2002) 1331.

D. Helbing, Rev. Mod. Phys. 73 (2001) 1067.
D. Chowdhury, L. Santen, A. Schadschneider, Phys. Rep. 329 (2000) 199.
B.S. Kerner, The Physics of Trac, Springer, Heidelberg, 2004.
D.E. Wolf, M. Schrekenberg, A. Bachem (Eds.), Trac and Granular Flow, World
Scientic, Singapore, 1996.
[6] D. Helbing, P. Mulnar, Phys. Rev. E 51 (1995) 4282.
[7] C. Burstedde, K. Klauck, A. Schadschneider, J. Zittartz, Physica A 295 (2001)
507.
[8] A. Kirchner, A. Schadschneider, Physica A 312 (2002) 260.

2921

[9] H. Klupfel, M. Meyer-Konig, J. Wahle, M. Schreckenberg, in: S. Babdini, T.

Worsch (Eds.), Proc. 4th Int. Conf. CA, Springer, London, 2000, p. 63.
[10] D. Helbing, I. Farkas, T. Vicsek, Nature 407 (2000) 487.
[11] Y. Tajima, T. Nagatani, Physica A 292 (2001) 545.
[12] M. Isobe, T. Adachi, T. Nagatani, Physica A 336 (2004) 638.
[13] M. Isobe, D. Helbing, T. Nagatani, Phys. Rev. E 69 (2004) 066132.
[14] R. Nagai, T. Nagatani, Physica A 366 (2006) 503.
[15] R. Nagai, M. Fukamachi, T. Nagatani, Physica A 358 (2006) 516.
[16] M. Muramatsu, T. Irie, T. Nagatani, Physica A 267 (1999) 487.
[17] D. Helbing, I. Farkas, T. Vicsek, Phys. Rev. Lett. 84 (2000) 1240.
[18] R. Jiang, Q.S. Wu, Physica A 373 (2007) 683.
[19] W.G. Weng, S.F. Shen, H.Y. Tuan, W.C. Fan, Physica A 375 (2007) 668.
[20] K. Nishinari, D. Takahashi, J. Phys. A 31 (1998) 5439.