Preservation of the qualitative properties in the vertical advection-diffusion DEM sub-model of the air pollution process
The analysis of the transport of air pollutants is one of the most widely investigated problems and it has great practical importance. The operator splitting method is a widely used tool for the solution of such problems. Because the numerical model has to preserve the main qualitative properties...
Збережено в:
| Дата: | 2006 |
|---|---|
| Автор: | |
| Формат: | Стаття |
| Мова: | English |
| Опубліковано: |
Інститут програмних систем НАН України
2006
|
| Теми: | |
| Онлайн доступ: | https://nasplib.isofts.kiev.ua/handle/123456789/1574 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Цитувати: | Preservation of the qualitative properties in the vertical advection-diffusion DEM sub-model of the air pollution process / M.A.T. Elshebli // Проблеми програмування. — 2006. — N 2-3. — С. 648-653. — Бібліогр.: 11 назв. — англ. |
Репозитарії
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
nasplib_isofts_kiev_ua-123456789-1574 |
|---|---|
| record_format |
dspace |
| spelling |
nasplib_isofts_kiev_ua-123456789-15742025-02-10T00:22:25Z Preservation of the qualitative properties in the vertical advection-diffusion DEM sub-model of the air pollution process Elshebli, M.A.T. Прикладне програмне забезпечення The analysis of the transport of air pollutants is one of the most widely investigated problems and it has great practical importance. The operator splitting method is a widely used tool for the solution of such problems. Because the numerical model has to preserve the main qualitative properties of the original model, we need such methods which preserve them as well. In this paper we investigate this question by use of the operator splitting approach. 2006 Article Preservation of the qualitative properties in the vertical advection-diffusion DEM sub-model of the air pollution process / M.A.T. Elshebli // Проблеми програмування. — 2006. — N 2-3. — С. 648-653. — Бібліогр.: 11 назв. — англ. 1727-4907 https://nasplib.isofts.kiev.ua/handle/123456789/1574 004.75 en application/pdf Інститут програмних систем НАН України |
| institution |
Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| collection |
DSpace DC |
| language |
English |
| topic |
Прикладне програмне забезпечення Прикладне програмне забезпечення |
| spellingShingle |
Прикладне програмне забезпечення Прикладне програмне забезпечення Elshebli, M.A.T. Preservation of the qualitative properties in the vertical advection-diffusion DEM sub-model of the air pollution process |
| description |
The analysis of the transport of air pollutants is one of the most widely investigated problems and it has great practical importance. The
operator splitting method is a widely used tool for the solution of such problems. Because the numerical model has to preserve the main
qualitative properties of the original model, we need such methods which preserve them as well. In this paper we investigate this question
by use of the operator splitting approach. |
| format |
Article |
| author |
Elshebli, M.A.T. |
| author_facet |
Elshebli, M.A.T. |
| author_sort |
Elshebli, M.A.T. |
| title |
Preservation of the qualitative properties in the vertical advection-diffusion DEM sub-model of the air pollution process |
| title_short |
Preservation of the qualitative properties in the vertical advection-diffusion DEM sub-model of the air pollution process |
| title_full |
Preservation of the qualitative properties in the vertical advection-diffusion DEM sub-model of the air pollution process |
| title_fullStr |
Preservation of the qualitative properties in the vertical advection-diffusion DEM sub-model of the air pollution process |
| title_full_unstemmed |
Preservation of the qualitative properties in the vertical advection-diffusion DEM sub-model of the air pollution process |
| title_sort |
preservation of the qualitative properties in the vertical advection-diffusion dem sub-model of the air pollution process |
| publisher |
Інститут програмних систем НАН України |
| publishDate |
2006 |
| topic_facet |
Прикладне програмне забезпечення |
| url |
https://nasplib.isofts.kiev.ua/handle/123456789/1574 |
| citation_txt |
Preservation of the qualitative properties in the vertical advection-diffusion DEM sub-model of the air pollution process / M.A.T. Elshebli // Проблеми програмування. — 2006. — N 2-3. — С. 648-653. — Бібліогр.: 11 назв. — англ. |
| work_keys_str_mv |
AT elsheblimat preservationofthequalitativepropertiesintheverticaladvectiondiffusiondemsubmodeloftheairpollutionprocess |
| first_indexed |
2025-12-02T03:32:15Z |
| last_indexed |
2025-12-02T03:32:15Z |
| _version_ |
1850365797267931136 |
| fulltext |
Прикладне програмне забезпечення
© K. Georgiev, E. Donev, 2006
ISSN 1727-4907. Проблеми програмування. 2006. №2-3. Спеціальний випуск 648
UDC 004.75
PRESERVATION OF THE QUALITATIVE PROPERTIES IN THE
VERTICAL ADVECTION-DIFFUSION DEM SUB-MODEL OF THE AIR
POLLUTION PROCESS
Maryam A.T. Elshebli
Department of Applied Analysis, Eötvös Loránd University
H-1117 Budapest, Pázmány P. s. 1/c, Hungary
The analysis of the transport of air pollutants is one of the most widely investigated problems and it has great practical importance. The
operator splitting method is a widely used tool for the solution of such problems. Because the numerical model has to preserve the main
qualitative properties of the original model, we need such methods which preserve them as well. In this paper we investigate this question
by use of the operator splitting approach.
Key-words: air pollution modelling, operator splitting, numerical solution.
Introduction
The transport of air pollutants is one of the most widely investigated phenomena and it is of great practical
importance (Prussov and Dorosenko 2003, Zlatev 1995). The numerical handling of this problem needs many efforts
due to the extremely big size of the unknown in the discretized models. The physical model describes of the
simultaneous effect of several different sub-processes. The mathematical equivalents of these processes are operators
describing the sub-processes. They are as a rule simpler than the whole spatial differential operator. Their separation
serves as the basic idea of the operator splitting method.
The preservation of the main qualitative properties of the continuous model is a natural requirement for the
global numerical models. In order to get such a discretization method we guarantee the preservation in each sub-model.
Clearly, this serves as a sufficient condition for the global preservation.
The application of operator splitting for the air pollution model results in an advection-diffusion sub-model, as
a separate task. For the solution of such a problem we need numerical methods. There are several robust numerical
methods and some of them are very powerful (c.f. Prussov and Dorosenko 2006). However, usually the convergence of
these methods are investigated only. In this paper we investigate the qualitative properties, namely, the preservation of
the maximum principle in the discrete model.
This paper is organised as follows.
In Section 2 we describe the mathematical model of the air pollution process. In the Section 3 we formulate the
operator splitting methods. We give the algorithm of the most widely used operator splitting methods like the sequetial,
Marchuk-Strang splittings and the weighted sequential splitting. We also mention some further splittings like the
additive splitting and the iterated splitting. In Section 4 we analyse the problem of how to choose the sub-operators in
the operator splitting process. We show that this choice can be performed differently in the air pollution model given in
Section 2. In Section 5 we formulate the requirement of the qualitative property preservation for the global
discretization method and we show how the operator splitting idea can be used to this aim. We analyse in detail one
sub-model in the DEM (Danish Eulerian Model), namely, the vertical advection-diffusion decomposition part. We
examine the preservation property of the maximum-minimum principle (the so-called discrete maximum-minimum
principle) and we give the condition of the choice of the discretization parameter for the bilinear finite element space
and Θ-method time discretization method. Finally we conclude how non-negativity preservation is related to this
property.
1. Air pollution modelling
In this section we present the mathematical model of the air pollution process.
Let ),( tcc jj x= denote the concentration of the j-th air pollutant, and c the vector function of these
functions. Then the time evolution of the vector c can be described mathematically by the system of partial differential
equations (Zlatev, 1995)
=
∈+−+∇∇+−∇=
∂
∂
)()0,(
],0()()()(
0 xx
u
cc
TtcRcEcKc
t
c σ
(1)
Прикладне програмне забезпечення
649
where ),( txuu = is a vector function describing the wind velocity, ),( tKK x= is the diffusion coefficient
function, ),( tEE x= is the function of emission, ),( txσσ = describes the deposition and R defines the chemical
reactions of the pollutants. The initial function c0(x) is given. Using these notations, the terms in equation (10) have the
following physical meaning. The first term on the right-hand side describes the transportation due to the velocity field,
which is called advection. The second term expresses the turbulent diffusion, the third term the emission, the fourth term
describes the deposition and the last term defines the chemistry of the pollutants.
We can see that (1) is a system of partial differential equations which is non-linear due to the chemistry term. The
analytical solution of such a system is impossible. In order to simplify the task, we use the idea of operator splitting.
The number of chemical species involved in an air pollution model sometimes reaches 200, or even more,
which results in a huge system of partial differential equations. The analytical solution of such a problem is obviously
impossible to find. Hence we have to treat it numerically. We note that in case of semi-discretization usually the number
of spatial grid points equals many millions. This means that the system of ordinary differential equations obtained after
spatial discretization is extremely big, hence the use of any numerical method developed for systems of ODE’s is rather
complicated. Moreover, the model equations contain terms that have different physical meanings and so different
mathematical properties (e.g. linear, non-linear, stiff and non-stiff). Therefore, it is impossible to find such a universal
numerical method which would perform well when applied directly to the original system. The application of operator
splitting allows us to treat the different physical terms separately.
Operator splitting method (OSM) is a kind of problem decomposition: we divide the spatial differential
operator of the global system into a few simpler operators and solve the corresponding problems one after the other, by
connecting them through their initial conditions.
The simpler systems which are obtained in this manner, and are sometimes called sub-systems, might have
some special properties that can be exploited in the numerical solution. The sub-systems are usually easier to treat
numerically than the whole system.
Splitting can be performed in several ways. We expect the method to be accurate as well as efficient enough.
The latter property depends on the number of computations and the possibility of performing the computations in
parallel. Taking into account the latter requirement, we made attempts to construct a new splitting scheme which does
not require a lot of computational work, and is parallelizable on the operator level.
2. Description of the OSM
In the sequel the frequently used splitting methods are described and compared. We describe the methods only
for two operators (i.e., n = 2), however, the generalisation for n operators is straightforward. (For more details, see
Hundsdorfer and Verwer, 2003, Zlatev, 1995, and Dimov at al., 2006.)
Hence, we consider the ACP
=
∈+=
,0
],,0(
0u)u(
T tBu(t) , Au(t)
dt
du(t)
(2)
where now A and B are given operators. In fact, the OSM yields a time-discretization method where instead of solving
the continuous (in time) problem (2) directly, we seek the split (discretized) solution on the grid points of the mesh
ωτ = {t j = jτ, j = 0, 1,…M}, where Mτ = T. (3)
Here τ > 0 denotes the splitting time step.
In the following we summarise some widely used splitting methods.
2.1 Sequential splitting. The scheme of this method is the following. As a first step, we solve the system with
operator A using the initial condition of the original problem, and then, applying the obtained solution at time τ as an
initial condition, we solve the system with operator B.
The solution obtained in this way is considered as the splitting solution in τ. This procedure is performed
cyclically in the following way:
−=−
−∈=
− ),)1(())1((
],,)1(( ,
2
1
1
)1(
1
ττ
ττ
kuku
kkt(t)Au
dt
(t)du
)(
k
)(
k
k
)(
k
(4)
Прикладне програмне забезпечення
650
=−
−∈=
),())1((
],,)1(( ,
12
)2(
2
ττ
ττ
kuku
kkt(t)Bu
dt
(t)du
)(
k
)(
k
k
)(
k
(5)
for k = 1, 2,…, m, where 0
)2(
0 0 u)(u = . The split solution at time t = kτ is defined as
),) )2( (kτu(kτu ksp = for k = 1, 2,…, M. (6)
It can be shown that the sequential splitting is a first-order method (see e.g. Havasi et al., 2001).
2.2 Strang-Marchuk splitting. Using this method, at each time step we begin and end the computation with
operator A (we apply it over a distance τ/2 twice) and put B to the middle (we apply it over a distance τ once) as
follows:
−=−
−−∈=
− ),)1(())1((
],)(,)1(( ,
3
1
1
2
1)1(
1
ττ
ττ
kuku
kkt(t)Au
dt
(t)du
)(
k
)(
k
k
)(
k
(7)
−=−
−∈=
),)(())1((
],,)1(( ,
2
112
)2(
2
ττ
ττ
kuku
kkt(t)Bu
dt
(t)du
)(
k
)(
k
k
)(
k
(8)
=−
−∈=
),())((
],,)(( ,
2
2
13
2
1)3(
3
ττ
ττ
kuku
kkt(t)Au
dt
(t)du
)(
k
)(
k
k
)(
k
(9)
for k = 1, 2,…, M, where 0
)3(
0 0 u)(u = . The splitting solution at t = kτ is defined as
),) )3( (kτu(kτu ksp = for k = 1, 2,…, M. (10)
This method has second-order accuracy.
2.3 Weighted sequential splitting. This method can be obtained by symmetrizing the sequential splitting in
the following way: in each time step we apply sequential splitting both in the order A → B as follows:
),)1(())1((
],,)1(( ,
1
)1(
1
−=−
−∈=
ττ
ττ
kuku
kkt(t)Au
dt
(t)du
sp
)(
k
k
)(
k
(11)
=−
−∈=
),())1((
],,)1(( ,
12
)2(
2
ττ
ττ
kuku
kkt(t)Bu
dt
(t)du
)(
k
)(
k
k
)(
k
(12)
and B → A as follows:
−=−
−∈=
),)1(())1((
],,)1(( ,
1
)1(
1
ττ
ττ
kukv
kkt(t)Bv
dt
(t)dv
sp
)(
k
k
)(
k
(13)
=−
−∈=
).())1((
],,)1(( ,
12
)2(
2
ττ
ττ
kvkv
kkt(t)Av
dt
(t)dv
)(
k
)(
k
k
)(
k
(14)
The splitting solution at t = kτ is defined as
)1)) )2()2( (kτθ)v((kτθu(kτu kksp −+= , for k = 1, 2,…, M, (15)
where 00 u)(usp = , and [0,1]∈θ is some fixed weight parameter.
Прикладне програмне забезпечення
651
The weighted sequential splitting is a second-order method for 5.0=Θ , otherwise it has first order. An important
property of the weighted sequential splitting is that it can be parallelized in a natural way (on the operator level). The
theoretical investigation of the method can be found in Csomós et al. (2003), and an application of the method in a one-
column transport-chemistry model is described in Botchev et al. 2003.
2.4 Other kinds of splittings. There are also other OSM’s, which has been recently developed. We want to
mentioned two methods of them, namely, the additive splitting and the iterated splitting.
The additive splitting is similar to the sequential splitting with the following only difference: in both sub-problems (4)
and (5) we use the same initial value ),)1( τk(usp − i.e., the split solution at the previous time level. If we denote by
))( (kτu A
k and ))( (kτu B
k the corresponding solutions, then the split solution at the new time level is defined as
))1())) )()(
τ(ku(kτu(kτu(kτu sp
B
k
A
ksp −−+= for k = 1, 2,…, M. (16)
The method has first order accuracy. The main advantage of this OSM is its easy parallelization. (For more details we
refer to Gnandt, 2005.)
The iterated splitting suggests the following algorithm: on the interval [tk-1,tk] we solve the following sub-problems
consecutively, for i = 1,3,5, … 2m+1.
),)1(())1((
],,)1(( ,)11()1(
−=−
−∈+= −
ττ
ττ
kuku
kkt(t)Bu(t)Au
dt
(t)du
sp
(i)
k
kk
(i)
k
(17)
),)1(())1((
],,)1(( ,
1
)11()1(
1
−=−
−∈+=
+
+
+
ττ
ττ
kuku
kkt(t)Bu(t)Au
dt
(t)du
sp
)(i
k
kk
)(i
k
(18)
where the initial function u0(t) is any fixed function for each iteration, and 0
)3(
0 0 u)(u = . The split solution at t = kτ is
defined as
),) )12( (kτu(kτu m
ksp
+= for k = 1, 2,…, M. (19)
The order of the accuracy of this OSM depends on the number of the inner iteration, and equals 2m+1. The main
advantages of the constructed OSM are the following. Each of the sub-problems result in a consistent approximation of
the exact solution. Moreover, using the built-in iteration, we can achieve (at least, theoretically) arbitrary high order for
the local splitting error.
3. Choice of the sub-operators
For some given complex mathematical model the sub-operators iA can be chosen in different ways. In the
particular case of the air pollution modelling, we sketch the following two basic possibilities.
3.1 Physical decomposition. It is natural to define the different sub-operators on the base of the separate
physical processes, namely, we can define the following operators
♦ )(1 ccA u−∇= - the advection operator,
♦ )(2 cKcA ∇∇= - the diffusion operator,
♦ ccA σ−=3 - the deposition operator,
♦ EcA =4 - the emission operator,
♦ )(5 cRcA = - the chemistry operator.
3.2 DEM decomposition. This kind of decomposition is used in the Danish Eulerian Model (DEM) and is
called DEM decomposition.
Прикладне програмне забезпечення
652
♦ )(
2
1
1 cucA i
i
i∑
=
∂−= - the horizontal advection operator,
♦ )(
2
1
2 ckcA ii
i
i ∂∂=∑
=
- the horizontal diffusion operator,
♦ ccA σ−=3 - the deposition operator,
♦ )(4 cREcA += - the emission and chemistry operator,
♦ )()( 333335 ckcucA ∂∂+∂= - the vertical transport operator.
The main advantage of the DEM decomposition is its high flexibility for 2D problems because only the last operator
contains the vertical part.
We note that in the DEM splitting the operators A1 and A2 are two-dimensional operators and -which is crucial in our
later analysis- the operator A5 is a one-dimensional operator. Hence the numerical investigation of the latter is simple.
4. Qualitative analysis in the vertical advection-diffusion DEM sub-model
Using the OSM, the split sub-problem for the vertical sub-operators 5A means the following one-dimensional
partial differential equation of parabolic type
],)1(( ,
,( 2
333 ττ kktckcu
t
t)xc −∈∂+∂=
∂
∂
(20)
where 3∂ denotes the partial derivative w.r.t. the vertical variable z. Here, for simplicity, we assume that the advection
and diffusion coefficients are constant. For this case the maximum principle can be proved, which states the following:
on some bounded domain the unknown function c(x,t) attains its maximum and minimum on the parabolic boundary,
i.e., either for t = 0 or for Ω∂∈x where Ω∂ denotes the boundary of the bounded space domain Ω. It is worth
preserving this property for the numerical solution, i.e., to guarantee the following property for the fully discretized
numerical model: whenever we introduce the mesh Ω∆z,∆t on the solution domain ΩT = Ω×(0,T), the discretization of the
problem (20) should take its maximal and minimal value at the corresponding discrete parabolic boundary points of the
mesh Ω∆z,∆t. If
n
ic denotes the approximation at the mesh-point (zi,tn) (where zi is the i-th space mesh-point and tn is the
n-th time level, respectively) then the maximal and minimal value of the numerical solution are achieved either for n =
0, or { }∂∈ Ni where the indices { }∂N correspond to the space boundary points.
If we use either the finite difference approximation or the linear finite element method to the semi-discretization for the
space variable and the usual Θ-method for the time discretization (which is, in fact the generalized trapezoidal formula),
then we can show the following sufficient condition which guarantees the discrete maximum-minimum principle.
THEOREM. Assume that the space discretization parameter is sufficiently small. Then for the bilinear finite element
approximation combined with the Θ-method the discrete maximum principle holds if the conditions
)3( 2
3
2
uhk
h
t
+Θ
≥∆ (21)
and
)6)(1( 2
3
2
uhk
h
t
−Θ−
≤∆ (22)
are fulfilled.
The proof is complicated and it can be found in Elshebli, 2006.
We note that it usually happens in the qualitative analysis of finite element approximations that there are both upper and
lower bounds for the time-step, which means that ∆t can be chosen neither too small nor too large. Such a kind of error
bounds can be found in some works, e.g. Faragó et al. 2004 and 2005.
The conditions (21) and (22) imply that this criterion can be applied only in the case where the lower bound is less than
the upper bound. This requirement implies that for the choice of Θ the requirement is
Прикладне програмне забезпечення
653
9
6
3
2
3
k
uhk −
≥Θ . (23)
This shows that for negative u the condition
3
2≥Θ (24)
should be satisfied.
It is worth mentioning that many other important qualitative properties can be shown from the discrete
maximum principle. One of the most crucial requirements is the non-negativity preservation property, which means the
following: if the output functions of the continuous problem are non-negative, then the solution of the problem is also
non-negative. The notion of the discrete non-negativity preservation is clear: if the vectors which correspond to the
initial function and the boundary condition are non-negative then the numerical solution should also be non-negative.
It can be shown that in both the continuous and discrete cases the discrete non-negativity preservation property and the
discrete maximum principle are equivalent. Hence, the conditions (21) and (22) together guarantee the non-negativity
property in the discretized model, too.
1. Botchev, M., Faragó, I., Havasi, Á., 2003, Testing weighted splitting schemes on a one-column transport-chemistry model. Int. J. Environmental
Pollution 22, 3-16.
2. Dimov, I., Faragó, I., Havasi, Á., Zlatev, Z., 2006, Different splitting techniques with application to air pollution models, Int. J.
Environmental Pollution (to appear)
3. Elshebli, M., 2006, Discrete maximum principle for bilinear finite elements on uniform square meshes (submitted)
4. Faragó.I., Horváth,,R. and Korotov, S., 2004, Discrete maximum principle for Galerkin finite element solutions to parabolic problems on
rectangular meshes, Numerical Mathematics and Advanced Applications, Springer Verlag, Berlin, 298-307.
5. Faragó.I., Horváth,,R. and Korotov, S., 2005, Discrete maximum principle for linear parabolic problems solved on hybrid meshes, Appl. Numer.
Math. 43, 249-264.
6. Gnandt, B.2005, A new operator splitting method and its numerical investigation, Advances in Air Pollution Modeling for Environmental
Security, 54, Kluwer, Amsterdam, 229-241.
7. Havasi, Á., Bartholy, J., Faragó, I., 2001, Splitting method and its application in air pollution modeling. Idıjárás 105, 39–58.
8. Hundsdorfer, W., Verwer, J. G., 2003, Numerical solution of time-dependent advection-diffusion-reaction equations. Springer, Berlin
9. Prusov, V., Dorosenko, A., 2003, Modelling and forecasting atmospheric pollution over region, Annales Univ. Sci. ELTE 46, 27-45.
10. Prusov, V., Dorosenko, A., 2006, On the numerical solution of the one-dimensional convection-diffusion equation, Int. J. of Environmental.
Pollution (to appear)
11. Zlatev, Z., 1995, Computer treatment of large air pollution models. Kluwer, Amsterdam.
|