Proceedings of ICFD 10:

Tenth International Congress of Fluid Dynamics

December 16-19, 2010, Stella Di Mare Sea Club Hotel, Ain Soukhna, Red Sea, Egypt

ICFD10-EG-30I1

MESHLESS METHODS: A NEW APROACH FOR MODELING FLUID FLOW, HEAT
TRANSFER, AND SPECIES TRANSPORT

D.W. Pepper
Professor and Director, Nevada Center for Advanced Computational Methods, University of Nevada Las Vegas, NV 89154

ABSTRACT

A meshless method is used to solve the governing equations
for fluid flow, heat transfer, and species transport. A
MATLAB format is utilized to program the method. Radial
basis functions are employed to solve for lid-driven cavity
flow, natural convection in a square enclosure, flow with
forced convection over a backward facing step, and species
transport within an enclosure. Results are compared with
COMSOL, a commercial finite element model, and FLUENT,
a commercial CFD finite volume-based model.

INTRODUCTION

Although many numerical and analytical schemes exist for
solving engineering problems, the meshless method is a
particularly attractive method that is receiving attention in the
engineering and scientific modeling communities. Finite
difference/volume, finite element and boundary element
methods have been historically used to model a wide variety
of engineering problems in complex geometries, generally
requiring extensive meshing. The meshless method is simple,
accurate, and requires no meshing.
The need to accurately simulate various physical processes in
complex geometries is important, and has perplexed modelers
utilizing conventional numerical schemes for many years.
Today, advances in numerical schemes and enhanced
hardware have lead to many commercial codes that can
employ Herculean efforts to solve complex stress-strain, heat
transfer, fluid flow, and other nearly intractable problems
[1,2]. Recently, advances in the development and application
of meshless techniques show they can be strong competitors to
the more classical numerical approaches. Liu [3] discusses
meshfree methods, implementation, algorithms, and coding
issues for stress-strain problems, and includes Mfree2D, an
adaptive stress analysis software package available for free
from the web.
1
Atluri and Shen [4] describe the meshless
method in more detail, including much in-depth mathematical
basis.

In the rapidly developing branch of meshless numerical
methods, there is no need to create a polygonization, neither in

1
http://www.nus.edu.sg/ACES/software/meshless2D/webfiles/
webpageMFree.htm
the domain nor on its boundary, essentially avoiding meshing
problems. A number of mesh reduction techniques such as the
dual reciprocity boundary element method [5], meshfree
techniques including the dual reciprocity method of
fundamental solutions [6], and meshfree local Petrov Galerkin
methods (MLPG) [4,7], have been developed for transport
phenomena and solution of the Navier-Stokes equations. In
addition, there exist other variations of meshless methods,
including kernel methods, moving least square method,
partition of unity methods, and radial basis functions. The
meshless method illustrated here is based on radial basis
functions (RBFs) and Kansas approach [8], and is becoming
widely popular.
BACKGROUND
Meshless methods are uniquely simple, yet provide solution
accuracies for various classes of problems that rival those of
finite elements and boundary elements without requiring the
need for mesh connectivity. Ease in programming, no domain
or surface discretization, no numerical integration, and similar
formulations for 2-D and 3-D make these methods very
attractive.
Development in using RBFs as a meshless method approach
for approximating partial differential equations has accelerated.
Kansas method [8], which is a domain-type meshless method,
was developed by directly collocating RBFs, especially
multiquadric approximations (MQ). The use of MQ was first
developed by Hardy [9] as an interpolation method for
modeling the earths gravitational field. Franke [10] published
a review paper on 2-D interpolation methods and ranked MQ
as the best based on its accuracy, speed, storage requirements,
and ease of implementation. Problems including the 1-D
nonlinear Burgers equation, heat transfer, and free boundary
problems are discussed in Li et al, [11].
Meshless methods utilizing RBFs create mesh-free algorithms
that are significantly simpler to employ than FDM/FVM, FEM,
and BEM approaches, and truly eliminate the need for meshes
requiring connectivity and optimization. Figure 1 shows an
arbitrary domain discretized using 3-noded triangular elements,
boundary elements, and a meshless method. An internal mesh
is required in the FEM (Fig. 1a) and linear elements are
needed along the boundary in the BEM (Fig. 1b). Both
methods require the use of efficient matrix solvers to obtain
values at the prescribed nodes, which can become resource
Copyright 2010 by ICFD 10
limiting and time consuming. The meshless method, with
arbitrarily distributed interior and boundary points, require no
mesh as illustrated in Fig. 1 (c).
(a)

(b)
(c)
Figure 1. Irregular domain discretized using (a) 3-noded
triangular finite elements, b) boundary element, and (c)
meshless method.

This paper focuses on the simplest class of mesh-free methods
being employed today known as radial basis function (RBF)
methods [8,9].


GOVERNING EQUATIONS

Assuming incompressible laminar flow with convective heat
transfer effects, the non-dimensional forms of the governing
equations for conservation of mass, momentum, and energy
can be written as [12]

V 0 V- = (1)

2
visc
V 1
V V p C V B
t
c
+ -V = V + V +
c
(2)

2
T
T
V T C T Q
t
c
+ -V = V +
c
(3)

2
C
C
V C C C S
t
c
+ -V = V +
c
(4)

where the body force in equation (2) is defined as B = PrRaT
in the y-direction for natural convection problems. For all
other cases, B = 0. The coefficients in equations (1)-(4) are
defined as (1) lid-driven cavity flow: C
visc
= 1/Re, (2) natural
convection: C
visc
= Pr, C
T
= 1, (3) forced convective flow: C
visc

= 1/Re, C
T
= 1/Pe, and (4) species transport, C
visc
= 1/Re and
C
C
= D (diffusion). S and Q are source/sink terms. The
Reynolds number, Rayleigh number, Prandtl number and
Peclet number are defined as

3
h c
g (T T )L VL VL
Re , Ra , Pr , Pe
| u
= = = =
ou o o


PRESSURE-VELOCITY COUPLING

A local pressure-velocity coupling (LPVC) algorithm is used
to account for pressure. The method represents a local variant
of already developed global solutions for coupled fluid flow
problems. A simple finite difference approximation is adopted
to calculate the time derivative and the Navier-Stokes
equations solved iteratively. The LPVC algorithm, where
pressure correction is estimated from local mass continuity
violation, is used to drive an intermediate velocity towards a
divergence-free velocity. Details regarding the formulation of
the intermediate pressure and velocity, and the correction steps,
are discussed in Pepper et al [12] and Kalla [13].

Steady-state is achieved if the criteria (where denotes V, T,
or C)

n 1 n
n
n
; 0
+
|
| |
< c | =
|
(12)

is met at all nodes. If the criteria are not met, the calculation
returns to the first step. For example, T
0
and T would denote
temperature at time t
0
and t
0
+t, respectively.

MESHLESS METHOD USING RBFs

There are many radial basis functions (RBFs) which have been
suggested and applied in various numerical schemes [14,15].
One of the most popular is the multiquadrics (MQ). Other
RBFs are thin-plate splines, Gaussian, and Inverse MQ.

MQs were initially proposed by Hardy [9]. Franke [10]
studied RBFs and found that MQs generally perform better
than others for the interpolation of scattered data. The
exponential convergence of MQ makes it superior to other
RBFs [16].

Since multiquadrics (MQ) are infinitely smooth functions,
they are often chosen as the trial function for | (some form of
RBF), i.e.,

2 2 2 2 2
j j j j
(r ) r c (x x ) (y y ) c = + = + + (13)

where c is a shape parameter provided by the user. The shape
parameter c strongly influences the accuracy of the MQ-RBF
method. The key factor in obtaining accurate results by the
RBF method is the MQ matrix. The choice of the shape
I
O
I
O
I
O
parameter c has been a topic of discussion in the community
of RBF researchers. Utilizing numerical experimentation on
functions of two variables, it was found there is an optimum
value of c at which the accuracy of the scheme is a maximum
and remains constant over a range of c.

In Kansas method [8], a function is first approximated by an
RBF, and its derivatives are then obtained by differentiating
the RBF. Although RBFs were initially developed for
multivariate data and function interpolation, their truly
meshfree nature has motivated researchers to employ them in
solving PDEs, especially for higher dimension problems.

The momentum equation (Eq. 2) is discretized using a linear
combination of RBFs and can be expressed in the form

N N
n 1 n
j j i i j j i i
j 1 j 1
N N
n 2 n
visc j j i i j j i i
j 1 j 1
N N N
n n n
j j i i j j i i j j i i
j 1 j 1 j 1

V (x , y ) V (x , y )
t C V (x , y ) P (x , y )
V (x , y ) V (x , y ) B (x , y )
[
]
+
= =
= =
= = =
= +
A V V
V +




i = 1,2,..,N
I
(14)

Discretized forms for the pressure and velocity correction
equations (denoted by arc overlays) are

I
2
N N
n 1 n
j j i i j j i i
j 1 j 1
i 1, 2, ...N
L

P (x , y ) V (x , y ),
t
+
= =
=

= V
A

(15)
I
N N
n 1 n
j j i i j j i i
j 1 j 1
i 1, 2...N
t
V (x , y ) P (x , y ),
+
= =
=
A
= V

(16)
Intermediate pressure and velocity correction equations can be
written as (caret overbar denotes correction)

N N N
n 1 n n
j j i i j j i i j j i i
j 1 j 1 j 1
P (x , y ) P (x , y ) P (x , y )
+
= = =
= +|

,
i = 1,2,,N
I
(17)
N N N
n 1 n n
j j i i j j i i j j i i
j 1 j 1 j 1
t

V (x , y ) V (x , y ) P (x , y ),
+
= = =
A
= | V

i = 1,2,,N
I
(18)

The energy equation (Eq. 3) is also discretized using a linear
combination of RBFs and can be expressed as

N N
n 1 n
j j i i j j i i
j 1 j 1
N N N
n 2 n n
T j j i i j j i i j j i i
j 1 j 1 j 1
T (x , y ) T (x , y )
t C T (x , y ) V (x , y ) T (x , y )
+
= =
= = =
= +
A V V
(
(



,
i = 1,2,,N
I
(19)

The concentration equation (Eq. 4) becomes

N N
n 1 n
j j i i j j i i
j 1 j 1
N N N
n 2 n n
C j j i i j j i i j j i i
j 1 j 1 j 1
C (x , y ) C (x , y )
t C C (x , y ) V (x , y ) C (x , y )
+
= =
= = =
= +
(
A V V
(



,
i = 1,2,,N
I
(20)
where N
I
denotes the total number of interior points and N
denotes total number of points, t denotes the time step,
superscript n+1 is the unknown value to be solved, and
superscript n is the current known value.

RESULTS

1. Lid-Driven Cavity Flow

The lid-driven cavity is one of the most frequently employed
benchmark cases to evaluate accuracy and feasibility of
numerical algorithms and commercial CFD software. Many
papers are available in the literature [17].

The boundary conditions for flow in a lid-driven cavity (0 x
1, 0 y 1) include a top lid that is moving at a unit
horizontal velocity with no-slip conditions on all other walls.

The computational results for various Reynolds numbers for
the lid-driven flow in a square cavity were compared with
results obtained by COMSOL and FLUENT. Both uniform
and random point distributions of 31 X 31 were used for the
RBF approximations. Figure 2 shows the comparison of
velocity vectors in the square cavity for Re = 100 using the
meshless method with velocity vectors using COMSOL and
FLUENT. The meshless results are in excellent agreement
with the two commercial packages.


(a) (b)


(c)
Figure 2. Velocity vectors in the square cavity using (a)
Meshless (b) COMSOL (c) FLUENT

2. Natural Convection in an Enclosure

Natural convection in a square enclosure is a very popular
benchmark problem which has been studied extensively for
over 40 years. Many papers continue to appear in the literature
utilizing various numerical techniques. Researchers usually
compare their simulation results with the benchmark solutions
obtained by De Vahl Davis [18], who employed a finite
difference scheme with a stream function/vorticity formulation.

The boundary conditions for natural convection in a
differentially heated square enclosure (0 x 1, 0 y 1)
are described in Fig. 3 as

Figure 3. Natural convection within an enclosed cavity

The domain of the problem is a closed square entity filled with
air (Prandtl number = 0.71) with differentially heated walls
and insulated horizontal walls. With constant initial
temperature, pressure and velocity set to zero, steady-state is
achieved using a transient solution. Results for various
Rayleigh numbers for the natural convection in a square cavity
are compared with those of COMSOL and FLUENT. The
mesh and nodal distributions are shown in Fig. 4 (a,b,c).


(a) (b)

(c)
Figure 4. Point distribution for natural convection in a square
cavity (a) COMSOL mesh, (b) FLUENT mesh, and (c)
meshless node distribution

Figure 5 shows the comparison of velocity vectors in the
square cavity for Ra = 10
3
using the meshless method with
velocity vectors from COMSOL and FLUENT. Meshless
results are in excellent agreement.


(a) (b)

(c)
Figure 5. Velocity vectors for natural convection for Ra = 10
3

in a square cavity using (a) COMSOL (b) FLUENT (c)
Meshless

For the case of Ra = 10
4
, velocity profiles on the vertical and
horizontal lines through the cavity geometric center are plotted
in Fig. 6 (a,b) and compare closely with corresponding results
from COMSOL and FLUENT.

(a) (b)

Figure 6. Velocity profiles for Ra = 10
4
along (a) vertical and
(b) horizontal central lines

In Fig. 7, meshless results showing temperature contours
ranging from 0 to 1 with 0.1 as the interval for Ra = 10
4
are
compared with results from COMSOL and FLUENT.
Meshless results are again in excellent agreement.


(a) (b) (c)
Figure 7. Isotherms for Natural convection in a square cavity
for Ra = 10
4
using (a) Meshless (b) COMSOL (c) FLUENT

3. Flow with Forced Convection over a Backward Facing
Step

Two-dimensional flow over a backward facing step is a well
known benchmark case that has been studied extensively over
many years the problem is easy to set up with known
(expected) results at various Reynolds numbers. Early
research work for this problem focused on the fluid pattern,
and many numerical simulations for this case can be found in
the literature. Gartling [19] examined this problem for
assessing outflow boundary conditions. In 1992, Blackwell
and Pepper [20] suggested flow over the backward facing step
with heat transfer as an ASME benchmark test problem.
Twelve numerical simulations were presented.

The boundary conditions for this problem are described as

For inlet flow:

1
0, for 0 y
2
u(y)
1
8y(1 2y), for y 1
2
v(y) 0
1 1
T(y) [1 (4y 1)][1 (4y 1)] for y 1
3 2
T(y) 1
0 for 0 y
x 2
s s
=
< s
=
= < s
c
= s s
c



`


)


On upper and lower walls:

u(y) v(y) 0
32
T n
5
= =
V =

where n is the outward unit vector normal to the domain
boundary.

For outlet flow:
p = 0


Figure 8. Problem configuration for forced convection in a
backward facing step

Figure 8 shows the configuration for forced convection over
the 2-D backward facing step. A constant heat flux is
introduced into the upper and lower channel walls
immediately downstream of the step. The purpose of this
particular set of conditions is to evaluate the change in
temperature along the upper and lower surfaces as initially
heated flow proceeds down the channel. Ideally, the
temperature gradient approaches a constant value with
increasing horizontal distance from the step. Flow over the
two-dimensional backward facing step is simulated for Re =
800 and Pr = 0.71. Distribution of interior nodes and boundary
nodes are shown in Fig. 9. Velocity vectors adjacent to the
step are shown in Fig. 10 (a,b,c).

(a)

(b)

(c)
Figure 9. Meshes for backward facing step solution (a) 284
nodes for the meshless method, (b) COMSOL mesh of 388
elements (c) FLUENT mesh of 284


(a)

(b)

(c)
Figure 10. Velocity vectors for backward facing step using (a)
Meshless (b) COMSOL (c) FLUENT

Velocity profiles at x = 7 and at x = 15 are shown in Fig. 11.
Meshless results compare closely with those obtained by
COMSOL and FLUENT.

(a) x = 7 (b) x = 15
Figure 11. Velocity profiles for Re = 800

Temperature profiles at x = 7 and at x = 15 are shown in Fig.
12. The comparisons of temperature are in excellent
agreement.

(a) x = 7 (b) x = 15
Figure 12. Temperature profiles for Re = 800

Temperature contours for Re = 800 are shown in Fig. 13. The
isotherms are nearly identical for all three models.


(a)

(b)

(c)
Figure 13. Isotherms for backward step flow using (a)
Meshless (b) COMSOL (c) FLUENT

4. Species Transport
Coupling the meshless method with a Lagrangian particle
technique, contaminant transport can be quickly simulated and
graphically displayed. The technique is particularly appealing
for evaluating contaminant transport as well as risk assessment
associated with homeland security issues.

The threat of chemical and biological agents being dispersed
within a building has become a reality, and is an important
cornerstone of the homeland security initiative and
Department of Homeland Security instigated by the US
government. This issue became evident when a letter
contaminated with anthrax was sent to former US Senator
Tom Daschles office in October 2001. Senator Daschles
office resided in the Hart Senate Office Building, which is a 9-
story complex located near the Capital Building in
Washington, DC. Fumigation and cleanup of the building took
approximately three months and cost about $14M. Traces of
anthrax were found in other rooms; however, it is unknown
exactly how the aerosolized spores dispersed from the
envelope to other parts of the building.

Contaminant dispersion traces are affected by both the
contaminant source location and the indoor ventilation
conditions. Figure 14(a,b) shows particulate transport within a
simple domain with ceiling ventilation and variable outflow
boundary conditions (door opening). The concentration
exhausts into the ceiling vent with the right door closed. As
the door is opened, the plume bends towards the door this is
due to the change in ventilation outflow.

(a) exterior door closed

(b) exterior door opened
Figure 14. Transport of particulates with exit door open and
closed

Figure 15 (a) shows the configuration of a two-room office
complex [21,22]. Different locations of contaminant sources
are denoted by the dots. Particles are released from a pollutant
source in the secretarys outer office. The meshless nodal
configuration is shown in Fig. 15 (b). Due to the outer office
door being opened, along with ventilation within the office
complex, the contaminant spreads into the inner office.

(a) Two-room office








(b) meshless nodal pattern
Figure 15. Office complex layout

Figure 16(a,b) shows the particle paths within the two rooms
for cases a and b. As can be seen from the particle dispersion
patterns, the pollutant is transported and diffused by the
ventilation pattern in the office complex. Source location is
particularly important as the pollutant can travel to either side
of the managers desk within the inner office. When first
responders arrive at an incident location, it is important that
they be aware of the trajectory of the spreading contaminant.
For example, the manager in the inner office should move to
the lower right corner of his desk in case a (Fig. 16a), but
should move to the upper left corner of his room in case b (Fig.
16b) until reached by a rescue team.

Figure 16. Indoor particle dispersion within office: (a) case a -
upper table and (b) case b - lower table.

CONCLUSIONS
A meshless method has been developed for solving fluid flow,
heat transfer, and species transport equations. Radial basis
functions (RBFs) are used to approximate a function and its
derivatives. Example problems related to recirculating flow in
a cavity, natural convection in a square enclosure, flow with
forced convection over a backward facing step, and species
transport within an enclosure were solved. Results have been
compared with well-known benchmark solutions. From the
comparisons made, the meshless method is very effective in
solving transport-based problems. The number of points
required to obtain comparable accuracy is much less than
mesh-based methods, and is an economical alternative for
fluid flow, heat transfer, or species transport problems.

Numerical implementation was done in MATLAB. Using only
a one step pressure correction, the algorithm needs only a
small number of calculations per iteration cycle. The
combined procedure makes the algorithm fast and robust.
Excellent agreement was achieved using model results
obtained by COMSOL and FLUENT.

NOMENCLATURE

B 0; PrRaT
c Shape parameter
C
C
Diffusion coefficient
C
visc
Pr; 1/Re
C
T
1; 1/Pe
g Gravity
L Reference length
n unit vector
N
I
Number of internal nodes
p Pressure
P
e
Peclet number, Pe = VL/
P
R
Prandtl number, Pr = /
Re Reynolds number, Re = VL/
Ra Rayleigh number, Ra = g(T
h
-T
c
)L
3
/
r
j
Radial dimension
S body force
t Time
T Temperature
T
c
Cold (or reference) temperature
T
h
Hot temperature (heated wall)
u Horizontal (x) velocity
v Vertical (y) velocity
V Velocity vector
x Spatial vector (x, y, z)
y Vertical direction
Thermal diffusivity
Coefficient of thermal expansion
t time step

error for temperature residual
Molecular viscosity
Dynamic viscosity
Density
Variable denoting V, T, or C
Radial basis function
Del operator
p Pressure gradient
ACKNOWLEDGEMENTS

I am very appreciative of the efforts provided by Dr. Xiuling
Wang, Purdue University Calumet, and Dr. David Carrington,
Los Alamos National Laboratory. I also wish to acknowledge
the work of Nagamani Kalla, former graduate student, and the
advice and comments from Dr. Bozidar Sarler, University of
Nova Gorica, Dr. Edward Kansa, University of California
Davis, and Dr. Alain Kassab, University of Central Florida.

REFERENCES

[1.] Lewis, R. W., Morgan, K., Thomas, H. R., and
Seetharamu, K. N., The Finite Element Method in Heat
Transfer Analysis, J. Wiley & Sons: Chichester, UK, 1996.
[2.] Huang, H-C. and Usmani, A. S., Finite Element Analysis
for Heat Transfer, Springer-Verlag: London, UK, 1994.
[3.] Liu, G. R., Mesh Free Methods: Moving Beyond the
Finite Element Method, CRC Press, Boca Raton, FL,
2002.

[4.] Atluri, S.N. and Shen, S., The meshless local Petrov-
Galerkin (MLPG) method: a simple & less-costly
alternative to the finite element and boundary element
method, Computer Modeling in Engineering & Sciences,
vol. 3, pp. 11-52, 2002.

[5.] Sarler, B and Kuhn, G., Primitive variables dual
reciprocity boundary element method solution of
incompressible Navier-Stokes equations, Engineering
Analysis with Boundary Elements, vol. 23, pp. 443-455,
1999.

[6.] Sarler, B., Towards a mesh-free computation of transport
phenomena, Engineering Analysis with Boundary
Elements, vol. 26, pp. 731-738, 2002.

[7.] Lin, H. and Atluri, S. N., Meshless local Petrov Galerkin
Method (MLPG) for convection-diffusion problems,
Computer Modeling in Engineering and Sciences, vol. 1,
pp. 45-60, 2001.

[8.] Kansa, E. J., Multiquadrics a scattered data
approximation scheme with application to computational
fluid dynamics, part I, Computers and Mathematics with
Applications, vol. 19, pp. 127-145, 1990.

[9.] Hardy, R. L., Multiquadric equations of topography and
other irregular surfaces, J. Geophy. Res., vol. 76, pp.
1905-1915, 1971.

[10.] Franke, R., A critical comparison of some methods for
interpolation of scattered data, TR NPS-53-79-003,
Naval Postgraduate School, 1979.
[11.] Li, J. Hon, Y. C., and Chen, C. S., Numerical
comparisons of two meshless methods using radial basis
functions, Engineering Analysis with Boundary
Elements, Elsevier Science Ltd., 26, pg. 205-225, 2002.
[12.] Pepper, D. W., Carrington, D. B., and Gewali, L., A
Web-based, Adaptive Finite Element Scheme for Heat
Transfer and Fluid Flow. ISHMT/ASME 4
th
Conf. on
Heat and Mass Transfer, Jan. 12-14, Pune, India, 2000.

[13.] Kalla, N. K., Solution of Heat Transfer and Fluid Flow
Problems using Meshless Radial Basis Function
Method, M.S. thesis, UNLV, 62 p., 2007.

[14.] Buhman, M. D., Radial Basis Functions, Cambridge
University Press, Cambridge, 2000.

[15.] Chen, W., New RBF collocation schemes and kernel
RBFs with applications, Lecture Notes in
Computational Science and Engineering, vol. 26, pp. 75-
86, 2002.

[16.] Franke, C., Schaback R., Solving partial differential
equations using radial basis functions, Applied
Mathematics and Computations, vol. 3, pp. 73-82, 1998.

[17.] Gresho, P.M., A modified Finite Element method for
solving the time-dependent incompressible Navier-
Stokes Equations, Part 2: Applications, Int. J. Numer.
Meth. Fluids, vol.4, pp. 619-640, 1984.

[18.] De Vahl Davis, G., Natural convection of air in a
square cavity: A bench mark numerical solution, Int. J.
Numer. Meth. Fluids, vol. 3, pp. 249-264, 1983.

[19.] Gartling, D. K., A Test Problem for Outflow Boundary
Conditions Flow over a Backward-Facing Step, Int. J.
Numer. Meth. Fluids, vol. 11, pp. 953-967, 1990.

[20.] Blackwell, B. F. and Pepper, D. W., Benchmark
Problems for Heat Transfer Codes, HTD, ASME WAM,
vol. 21, pp. 190-198, 1992.

[21.] Pepper, D. W. and X. Wang, Modeling Indoor
Contaminant Dispersion, ICCES, Vol. 1, No. 1, pp. 1-5,
2007.

[22.] Wang, X. and D. Pepper, A high-order numerical
method for interior contaminant dispersion, 46
th
AIAA
Aerospace Sciences Meeting and Exhibit, Jan. 7 10,
Reno, Nevada, 2008.