Documente Academic
Documente Profesional
Documente Cultură
Setyawan Widyarto
FSKSM-UTM Skudai, Building N28, Room 337-01
Johor Bahru 81310,
swidyarto@siswa.utm.my
Telp. (+60-7)-55-36503, (+60)-127617388
Dr. Muhammad.Shafie Abd Latiff,
Abstracts
This paper discusses Burger’s equation to be used in modelling a crowd with traffic approach using conservation laws
and focuses on a moving mesh method that is applied to the solution of hyperbolic equations. A mesh consists of
nodes (points) at specific locations in space that are connected to form elements. The moving mesh method is
developed utilizing monitor functions to drive the mesh motion and is based on equidistribution ideas. The method is
derived in multi dimensions from a conservation of monitor function principle and from this initial principle a
conservation law for the monitor function is derived. Numerical results for the moving mesh methods to the Burger’s
equation with several values of ε , including when ε =0 or inviscid Burger’ equation is presented.
Key words: moving mesh, Burger’s equation, crowd model, conservation laws
t2
∫t (ρ (x1 , t )v(x1 , t ) − ρ (x2 , t )v(x2 , t )) dxdt =
2. Inviscid Burger Model 1
t2 x2
− ∫t ∫ x ∂ x (ρ ( x , t ) v ( x , t )) dxdt (2)
Consider the crowd flows as a traffic with only one 1 1
lane (i.e., overtaking is impossible, e.g. wheel chair
lane on the Sa’ey hajj ritual). Instead of modelling Since x1, x2 Є R, t1, t2 > 0 are arbitrary, it is concluded
pilgrims individually, the density ρ(x; t) (pilgrims per
meter) of pilgrims is applied in x Є R at time t ≥0. The ρ t + (ρ v )x = 0 x ∈ R, t > 0 (3)
number of agents which are in the interval (x1; x2) at
time t is
x2 This is a partial differential equation. It has to be
∫x ρ ( x, t )dx supplemented by the initial condition
1
and u solves (8)-(9) point wise. Equation (6) can be The moving mesh methods are required to generate a
simplified by bringing it in a dimensionless form. Let mapping from a regular domain in a parameter space
L and τ be a typical length and time, respectively, such Ωc to an irregularly shaped domain in physical
that L = τ = vmax . space Ω . By connecting points in the physical space
Introducing corresponding to discrete points in the parameter
x t 2ρ space, the physical domain can be covered with a
xs = ts = , u =1− , computational mesh suitable for the solution of finite
L, τ ρ max element equations. The moving mesh methods include:
• Mesh equations. The mesh equations determine a
it is obtained one-to-one mapping from a regular domain in a
1 ⎡ρ ⎤ ρ parameter space to an irregularly shaped domain
∂ t ρ = ∂ t s ⎢ max (1 − u )⎥ = − max ∂ t s u, in physical space. By connecting points in the
τ ⎣ 2 ⎦ 2τ
Proceedings of the Postgraduate Annual Research Seminar 2006 266
physical space corresponding to discrete points in x L = X 0 < Λ < X i (t ) < X i +1 (t ) < Λ < X N +1 = x R
the parameter space, the physical domain can be
covered with a computational mesh suitable for are chosen to satisfy a so-called ‘equidistribution’
the solution of element equations. Choosing (uniform) principle. Given a so-called monitor
suitable mesh equations and solving them
efficiently are very crucial for an effective moving
function m (u, x ) , which serves as an indicator for
mesh method; the error between computed solution and exact
• Monitor function. A monitor function is used to solution of the PDE
∂
guide the mesh redistribution. It may depend on u ( x , t ) = f (u ( x , t )), xL < x < xR ,t > 0
the solution arc length (in 1D), curvature, and a ∂t
posteriori errors. In practice, local (spatial) u (x L , t ) = u (x R , t ) = 0, t > 0,
smoothing of the monitor function is necessary, u ( x , 0 ) = u 0 ( x ), (17 a , b , c )
see, e.g., [Cao, et.al, 1999, Huang et.al, 1997]
generating grid points such that
• Interpolations. If the mesh equations are time-
dependent and are solved simultaneously with the x i +1 (t )
given differential equations, then interpolation of
∫x (t ) m(u (~
x , t ), ~
x )d~
x=
dependent variables from the old mesh to the new i (18)
mesh is unnecessary. Otherwise, some kind of 1 xR
interpolation is required to pass the solution
∫x m(u (~
x , t ), ~
x )d~
x , i = 0, ..., N
information on the old mesh to the newly N +1 L
generated mesh.
or, equivalently
x i +1 (t )
∫x (t ) m(u (~
x , t ), ~
x )d~
3.1. The Monitor Function
x=
( )
i
Let PDE within a spatial operator is u t = L u . xi + 2 (t )
∫x m(u (~
x , t ), ~
x )d~
x , i = 0, ... , N − 1 (19)
Discretions require meshes in space, which should i +1 (t )
adapt to the motion of the solution as time progresses. Usually the monitor function
So that an appropriate mesh to represent the solution
must be sought. The model problem in 1D is 1
represented by ⎛ n ⎞ 2
⎜ ∑
m(u, x ) = ⎜ α 0 + α i (ui )2x ⎟ ,
⎟
(20)
ut = L (u ), in (0,1)× (0, T ) (13) ⎝ i =1 ⎠
α i > 0, i = 0, ..., n, fixed ,
A mesh transformation x = x(ξ , t ) , such that is chosen, or in the case n =1
uniform points in the ξ coordinate. Let x and ξ
respectively denote the physical and computational m(u, x ) = α + (u x )2 ( )1
2
, α > 0. (21)
pilgrim coordinates in crowd over the unit interval [0,
1] and a one-to-one coordinate transformation between in other form it is written
those domains is designated by
(25) x (ξ , t ) 1
∫0 M (t , y, u ) dy = ξ ∫0 M (t, y, u ) dy (31)
where
1
θ (t ) = ∫0 M (~x , t )∂ ~x (26) A moving mesh PDE is therefore, taking two
derivatives of (31) with respect to ξ
Differentiating (25) with respect to ξ , it will be
∂ ⎛ ∂ ⎞
⎜ M (t , x(ξ , t ), u ) x(ξ .t )⎟ = 0 (32)
∂ ∂ξ ⎝ ∂ξ ⎠
M ( x(ξ , t ), t ) x(ξ , t ) = θ (t ) (27)
∂ξ In other version (32) is written
This EP, which does not contain the node
( )
speed x&ξ , t , is called quasi-static EPs (QSEPs). ∂⎛ ∂ ⎞
Related to this QSEP, an MMPDE is derived (Huang
⎜⎜M(t, x(ξ,t),u) x&(ξ,t)⎟⎟ =
∂ξ ⎝ ∂ξ ⎠ (33)
et.al, 1994)
1∂⎛ ∂ ⎞
2 − ⎜⎜M(t, x(ξ,t),u) x(ξ,t)⎟⎟
∂ x& 1 ∂ ⎛ ∂ x ⎞ (28) r ∂ξ ⎝ ∂ξ ⎠
=− ⎜M ⎟
∂ξ 2 τ ∂ξ ⎝ ∂ξ ⎠
This MMPDE does not only force the mesh ( x (ξ , t )) 3.3. Spatial smoothing
toward equidistribution but also prevent the mesh Some sort of smoothing of the mesh are often useful in
order to reach a reasonable accuracy in the computed
crossing. Especially, the term − 1 ∂ ⎛⎜ M ∂ x ⎞⎟ plays the solution. The values of the smoothed monitor function
τ ∂ξ ⎝ ∂ξ ⎠ ~
M at nodes are defined by (Huang et al, 1994)
fundamental role of a correction term as a source for
the mesh movement to stabilize the mesh trajectories k −i
and to make the mesh equidistribute the monitor 2⎛ γ ⎞
∑ k = i − p (M k )
i+ p
function. The parameter τ represents a timescale for
⎜ ⎟ (34)
~
Mi = ⎝1 + γ ⎠ ,
forcing the mesh toward equidistribution. Equation k −i
⎛ γ ⎞
∑
(28) is being discrete using the line method in space i+ p
⎜ ⎟
with control finite difference on the uniform mesh is k =i − p 1 + γ
⎝ ⎠
approximated by
1
[x&i +1 − 2 x&i + x&i −1 ] = − Ei , (29) where γ is a positive constant called the smoothing
⎛1⎞
2 τ parameter and p is a non-negative integer which is
⎜ ⎟ referred to smoothing index. The summations in the
⎝n⎠ above equation are understood to contain only
elements with indices in the range between 0 and n.
The final discrete moving mesh equation is obtained
Ei is the discrete approximation of ∂ ⎛⎜ M ∂x ⎞⎟ at ~
∂ξ ⎝ ∂ξ ⎠ by replacing Mi by Mi in its MMPDE.
ξ = ξi i.e.
4. Numerical Experiments result
of ε and initial solutions, including when ε =0 or Initial solution: u ( x, 0 ) = sin (2π x ) + 1 sin (π x ), 0 ≤ x ≤ 1 .
inviscid Burger’s equation (11), are presented. 2
Experiment 2 would be the way out to overcome the
u⎛ 2⎞
ut = ε u xx − ⎜⎜ ⎟⎟ , 0 < x < 1, t >0 jam of pilgrims flow. The crowd region line is the
most distributed over t and x.
⎝ 2 ⎠x 1
Burgers equation. Grid trajectories
0.9
∂u
f (u ) = u&− x&
0.7
∂x 0.6
0.5
Discreting by centred finite differences will give
t
(u − ui −1 ) x&, i = 1, 2, 3, ..., n − 1
f i = u&i − i +1
0.4
(xi +1 − xi −1 ) i
0.3
0.2
4.1. Experiment 1.
Initial solution: u ( x , 0 ) = 0.5 + sin (2π x ),
solution u
0 ≤ x ≤1 0.5
0.8 X: 0.75
Y: -0.5
0.7 -0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
0.6
0.5
Figure 2. Mesh trajectories and solution, ε =
t
0.1
4.3. Experiment 3.
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Initial solution: u ( x , 0 ) = 0.5 + sin (2π x ), 0 ≤ x ≤ 1 .
Experiment 3 used ε =0 that means the equation
x
X: 0.5925
0.16 Y: 0.1779
0.5
0.14
0 0.12
0.1
t
X: 0.6887
-0.5 Y: -0.6771 0.08
0.06
-1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.04
x
0.02
4.2. Experiment 2.
Proceedings of the Postgraduate Annual Research Seminar 2006 269
1.5
Burgers equation solution. K. Lipnikov, M.Shashkov (2006), Error-Minimization-
t=0
t = 0.2
Based Strategy for Moving Mesh Method,
Communications in Computational Physics 9(1): 53-
1 81
solution u
0.5
X: 0.7492
Y: -0.5
-0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x
5. Conclusion
The Burger’s equation has ε that dominates the mesh
trajectories. This feature would become the parameter
for the crowd density distribution. The moving mesh
methods involve three parameters, i.e. τ, N, and p. It is
recommended to make more experiments in future to
get the most important parameter in crowd control. The
initial solution can be also approached by real situation
by plotting the real data video to get an initial solution
for simulation.
References
W.M. Cao, W.-Z. Huang, and R.D. Russell, A study of
monitor functions for two dimensional adaptive mesh
generation, SIAM J. Sci. Comput. 20 (1999), 1978–
1994,