SOLUTION OF A LOW PRANDTL NUMBER NATURAL

CONVECTION BENCHMARK BY A LOCAL MESHLESS METHOD

Gregor Kosec
*
and Boidar arler

*
Joef Stefan Institute
Laboratory for Parallel and Distributed Computing
Jamova 39, SI-1000 Ljubljana, Slovenia
and
University of Nova Gorica
Laboratory for Multiphase Processes
Vipavska 13, POB 301, SI-5001 Nova Gorica, Slovenia
e-mail: gkosec@ijs.si, phone: +386 1 477 3701; fax: +386 1 477 3111

University of Nova Gorica

Laboratory for Multiphase Processes
Vipavska 13, POB 301, SI-5001 Nova Gorica, Slovenia
and
Centre of Excellence for Biosensorics, Automation and Process Control
Laboratory for Advanced Materials Systems
Velika pot 22, SI-5250, Solkan, Slovenia
e-mail: bozidar.sarler@ung.si, phone: +386 5 3653 544, fax:+368 5 3315 359


Abstract
Purpose Solution of a highly nonlinear fluid dynamics in a low Prandtl number regime,
typical for metal like materials, as defined in the Call for contributions to a numerical
benchmark problem for 2D columnar solidification of binary alloys (Bellet, et al., 2009). The
solution of such a numerical situation represents the first steps towards understanding the
instabilities in a more complex case of macrosegregation.
Approach - The involved temperature, velocity and pressure fields are represented through
the local approximation functions which are used to evaluate the partial differential operators.
The temporal discretization is performed through explicit time stepping.
Originality - The solution procedure is formulated completely through local computational
operations. Besides local numerical method also the pressure velocity is performed locally
with retaining the correct temporal transient.
Findings - The performance of the method is assessed on the natural convection in a closed
rectangular cavity filled with a low Prandtl fluid. Two cases are considered, one with steady
state and another with oscillatory solution. It is shown that the proposed solution procedure,
despite its simplicity, provides stable and convergent results with excellent computational
performance. The results show good agreement with the results of the classical finite volume
method and spectral finite element method.
Gregor Kosec and Boidar arler






2







Key words Newtonian incompressible fluid, low Prandtl number, natural convection,
rectangular cavity, primitive variables, meshless methods, local radial basis function
collocation method, multiquadrics, local pressure-correction.
Paper type Research paper

1 INTRODUCTION
The computational modelling of solidification has become an important research subject due
to its pronounced influence in better understanding of nature as well as in the development of
the advanced technologies. The main computational difficulties in tackling solidification
systems stem from the moving interface between the solid and the liquid phase, respective
jump of the physical properties (density, viscosity, etc.) across the interface as well as
possible jumps in heat flux, species flux and concentration at the interface. The jumps are
accompanied by the fast thermal diffusion compared to the species diffusion, and involved
convection phenomena. The convection phenomena arise due to the thermal and solutal
gradients in the melt as well as convection, triggered through the movement of the interface
between the solid and the liquid phase with different densities. The situation could be
additionally complicated by the surface tension governed (Marangoni) convection, and
movement of the dispersed solid phase, that might be present in the melt. The respective flow
patterns can be laminar, periodic or turbulent. The computational treatment of solidification
phenomena can be conducted on different scales that might range from the dendritic growth
on the micron scale (Lorbiecka and arler, 2010), grain growth on the millimeter scale, and
system level on the meter scale (Vertnik, 2011). It is a long term vision to be able to solve the
grain size and distribution of the solidified system, microsegregation (chemical
inhomogeneities) on the scale of the solidified grain, and macrosegregation (chemical
inhomogeneities) on the system level, together with a range of segregations on the length
scales between micro and macro.

There have been many computational studies performed up to now to simulate the
macrosegregation. However, a precise answer on how accurate the results are (verification of
macrosegregation modelling) is quite a novel issue, proposed in a recent benchmark test by
(Bellet, et al., 2009). This benchmark test, that logically follows the melting exercise,
proposed by (Gobin and Le Qur, 2000) consists of two parts. I. A separate preliminary
single phase exercise: concerned with the convective problem in the absence of solidification
and in conditions close to those met in solidification processes. Two problems are considered
for the case of laminar natural convection: I. a transient thermal convection for a pure liquid
metal with a Prandtl number of the order of 10
-2
, and I. b double-diffusive convection in an
Gregor Kosec and Boidar arler






3







enclosure for a binary liquid metallic mixture with a Prandtl number on the order of 10
-2
and a
Lewis number on the order of 10
-4
. II. The simulation of the full solidification process: First a
specified minimal solidification model is proposed with II. a Pb-18%Sn and II. b Sn-10%Pb
alloys. The objective is to compare the numerical solutions obtained by different contributors
(verification). Then different physical solidification models may be compared to check the
features that allow for the best possible prediction of the physical phenomena (validation). It
is essential for proper simulation of macrosegregation to be able to simulate the transient
natural convection and the double diffusive flow structure. The part I.a of the proposed
benchmark is tackled in the present work, which by the best of knowledge of the present
authors represents the first results achieved on this benchmark.

Natural convection heat transfer problems are fully determined through Rayleigh and Prandtl
dimensionless numbers. Rayleigh number stands for the ratio between the buoyant to the
viscous effects, and depends also on the geometry and boundary conditions. The Prandtl
number, defined as the ratio of the momentum diffusivity to the thermal diffusivity, is an
intrinsic thermophysical property of a fluid, which is not directly dependent on the problem
configuration. The Prandtl number varies over several orders of magnitude from 10
-3
(liquid
metals) to 10
5
(functional oils) in common fluids. Ratio between Prandtl and Rayleigh
numbers is also known as Grashof number, which characterises the flow regime. There are
three important values of the Grashof number which demarcate different flow bifurcations in
a specific natural convection arrangement. The first critical value denotes the onset of steady
convection from the conduction regime, the second one the onset of the periodic convection
from the steady convection, and the third one the onset of the turbulent convection from the
periodic convection. Subsequently, the effects of geometry, boundary conditions and fluid
properties continue to attract the computational fluid dynamics community in search for more
and more accurate as well as diverse simulations of natural convection (Amimul, 2011).

Before tackling the macrosegregation as a consequence of the solidification of binary
substance, a proper solution of natural convection, double diffusive convection and freezing,
is needed. Detailed analyses and accurate solutions of different stages of solidification will
maximize the possibilities to identify the differences between different numerical solutions at
the point when they might originate. A comparison study of solution of macrosegregation,
computed with the finite volume method and the finite element method (with quite different
qualitative and quantitative response of different numerical method simulations) already
exists, however there was until now no effort given to reasonably systematic identification of
the origin of the differences in both solutions (Ahmad, et al., 1998).

Gregor Kosec and Boidar arler






4







The defined (Bellet, et al., 2009) macrosegregation cases with Pb-18%Sn and Sn-10%Pb give
as a consequence results with mesosegragates which seem, at the moment, difficult to
calculate in a discretisation and numerical method independent way. We have recently
defined a related macrosegregation case, free of mesosegregates, and most probably for the
first time demonstrated mesh independent and numerical method independent results (Kosec,
et al., 2011). The numerical methods used were the finite volume method and the local radial
basis function collocation method (Kosec and arler, 2008).

The transient natural convection of the Low-Prandtl-Number fluids in 2D was probably for
the first time studied in archival literature by Mohamad and Viskanta (Mohamad, 1991). Their
work follows a number of technical reports, cited in their paper. They numerically caught
periodic oscillations at the critical Grashof numbers. Prior to the oscillatory flow, the steady
state solutions with an oscillatory transient period were numerically predicted. The effect of
boundary conditions and time marching schemes on the Mohamad and Viskanta reference
solutions has been further studied by (Cless, 1996a, Cless, 1996b) .A three dimensional study
of Rayleigh-Benard convection of a low-Prandt-Number fluid in shallow three dimensional
cavity heated from below was studied in (Nakano, 1998). All mentioned studies used the
finite difference method solution procedure. The work of Mohamad and Viskanta was
recalculated by the finite element method in 1999 (Sammouda, 1999). A problem of a low-
Prandtl-number natural convection in volumetrically heated rectangular enclosures was
studied for different situations by (Arcidiacono, 2001a, Arcidiacono, 2001b, Di Piazza, 2001).
A comparison between the finite volume method and the spectral Chebyshev method (Xin,
2002), for low Prandtl number natural convection in a shallow cavity was given by (Zalonik,
et al., 2005). The transition from steady to oscillatory flow for a very low Prandtl number
fluid (Pr= 0.008) was studied for rectangular enclosures with aspect ratios (length/height) of
0.25, 0.4, 1.0, and 2.0 by (Crunkleton, 2006) with the finite volume method and SIMPLE
pressure correction algorithm. A graph of Rayleigh number of first bifurcation of natural
convection was recently given for a range of Prandtl numbers from 0.1 to 1.0 in (Xiaohua,
2009). Tesso and Piva (Pesso and Piva, 2009) recently calculated the low Prandtl natural
convection in a square cavity, caused by large differences in the working fluid, characterised
by the Gay-Lussac number (Ga) in addition to the Ra and Pr. They used a finite volume based
commercial software for obtaining the steady state results for a range of Pr from 0.0071 to
0.71, Ra from 10 to 10
8
and Ga from 0 to 2. Finally the authors propose a heat transfer
correlation for the tackled range of situations.

The meshless or sometimes also named meshfree or mesh reduction methods represent a class
of numerical methods where an arbitrarily distributed set of nodes, without any additional
Gregor Kosec and Boidar arler






5







topological relations between them, is used. Such meshless methods represent a promising
technique to avoid problems with polygonisation. There exist several meshless methods such
as Element free Galerkin method, the Meshless Petrov-Galerkin method, the point
interpolation method, the point assembly method, the finite point method, smoothed particle
hydrodynamics method, reproducing kernel particle method, Kansa method (Monaghan,
1988, Kansa, 1990a, Kansa, 1990b, Atluri and Shen, 2002b, Atluri and Shen, 2002a, Chen,
2002, Liu, 2003, Gu, 2005, Fasshauer, 2006, terk and Trobec, 2008, Trobec, et al., 2009).
However, this work is focused on one of the simplest classes of meshless methods in
development today, the local point interpolation (Wang and Liu, 2002) Radial Basis Function
(Buhmann, 2000) Collocation Method (RBFCM) (arler, 2007). Undoubtedly, these methods
can be of great advantage in solving solidification processing problems. In the present paper
we use a local variant of RBFCM (arler, 2007), the Local Radial Basis Function Collocation
Method (LRBFCM). The main advantage of the local approach is that the spatial disretization
problem is simplified to solving small systems instead of large global systems, which might
become unstable for increasing number of computational nodes. The LRBFCM approach was
already successfully applied to several thermofluid situations, ranging from the laminar to
turbulent situations (arler and Vertnik, 2006, Vertnik and arler, 2006, Divo and Kassab,
2007, Kosec and arler, 2008).
2 GOVERNING EQUATIONS
The considered problem domain is a closed air-filled square-shaped cavity with differentially
heated vertical walls with temperature difference T A and insulated horizontal walls. The
non-permeable and no-slip velocity boundaries are assumed. The problem is described by
three coupled PDE equations: mass (1), momentum (2), and energy conservation (3)
equations, where all material properties are considered to be constant. The Boussinesq
approximation (4) is used for the treatment of body force in the momentum equation. The
natural convection is thus described by the following system of equations
0 V = v , (1)
( ) ( ) P
t

c
+ V = V +V V +
c
v
vv v b , (2)
( )
( ) ( )
p
p
c T
c T T
t

c
+ V = V V
c
v , (3)
| |
ref
1 ( )
T
T T | = b g . (4)
Gregor Kosec and Boidar arler






6







with
ref
, , , , , , , , ,
p T
P T c T | v g and b standing for velocity, pressure, temperature,
thermal conductivity, specific heat, gravitational acceleration, density, coefficient of thermal
expansion, reference temperature for Boussinesq approximation, viscosity and body force,
respectively. The case is characterized by three dimensionless numbers
3 2
Ra =
T H p
T
T c |

A O g
(5)
Pr
p
c

= , (6)
A
H
W
O
=
O
, (7)
standing for Rayleigh number, Prandtl number and domain aspect ratio. The ratio between
Rayleigh and Prandtl is often referred as Grashof number, defined as
Ra
Gr=
Pr
T
. (8)
H
O

stands for domain height and
W
O for domain width (Figure 1).

Figure 1: The natural convection benchmark test schematics.
Gregor Kosec and Boidar arler






7









The introduced physical model does not have a closed form solution and in order to solve it a
numerical approach has to be employed. A reference eponymous work in this field is that by
De Vahl Davis (de Vahl Davis, 1983).
3 SOLUTION PROCEDURE
In this paper we use a novel local meshless numerical method with local pressure velocity
coupling. A general idea behind the local meshless methods is the use of local sub clusters of
domain nodes (Figure 2), named local support domains, with local basis functions for the
approximation of fields. With the selected support domain, an approximation function is
introduced as a sum of weighted basis functions
1
( ) ( )
Basis
N
n n
n
u o
=
= +

p p , (9)
where , , and
Basis n n
N u o + stand for the approximation function, the number of basis
functions, the approximation coefficients and the basis functions, respectively. The basis
could be selected arbitrarily (e.g., monomials, radial basis function, etc.), however in this
paper Hardys Multiquadrics (MQs)
( ) ( ) ( )
2
/ 1
n n
n C
o + = + p p p p p , (10)
with
C
o standing for the free shape parameter of the basis function, are used, based on the
results of the study by Franke (Franke, 1982). By taking into account all support domain
nodes and equation (9) the approximation system is obtained. In this paper we use collocation
(the number of support nodes is the same as the number of the basis functions). An arbitrary
spatial differential operation ( ) L can be applied on the approximation function in the
following way
( )
1
( )
Basis
N
n n
n
L L u o
=
= +

p p . (11)
In general, the system (9) has to be solved only when the influence domain topology changes
and therefore the computation can be optimized by computing
1
in a pre-process.
Gregor Kosec and Boidar arler






8







Furthermore, the computation of the coefficients and the evaluation of differential operators
can be combined. All information about the numerical approach and the local nodal topology
can be stored in a predefined vector, which has to be re-evaluated only when the topology of
the nodes changes. The differential operator vector (
L
m
_ ) is introduced as
( )
1
1
( ) ( )
N
L
m nm n
n
L _

=
= + +

p p

(12)

The introduced formalism holds in general and therefore the general notation for partial
differential operator ( ) L is used. However, in the present work, only operators
/ p
c
c c
and
2
V
are employed.
2
2
1
2
1
( ) ( )
N
m nm n
n
p
c
c
_
V
=
c
= + +
c

p p , (13)
/ 1
1
( ) ( )
N
p
m nm n
n
p
c
c
_
c c
=
c
= + +
c

p p , (14)
The structured formulation is convenient since most of the complex and CPU demanding
operations are performed in the pre-process phase. For all inner temporal loop operations only
N floating point operations (FLOPS) are need for evaluation of an arbitrary partial differential
operator. The implementation of the Dirichlet boundary condition is straightforward. In order
to implement Neumann and Robin boundary conditions a special case of interpolation is
needed. In these boundary nodes the function directional derivative instead of the function
value is known and therefore the equation in the interpolation system changes to
1
( ) ( )
Basis
N
BC n n n
n
a b u o
=
c | |
= + + +
|
c
\ .

p p
n
. (15)

In the presented numerical framework the computation of Neumann and Robin boundary
conditions can be simplified through the usage of the differential operator vector. Consider
the Neumann boundary condition
BC
a b
u
u u
c
+ =
cn
, (16)
Gregor Kosec and Boidar arler






9







/
2
0 /
0
Sub
N
BC m m
m
a
a b
u _ u
u
_
c c
=
c c

=
+

n
n
, (17)
where
0
u stands for boundary node. Equation (17) simplifies to Neumann boundary condition
computation if b is set to zero. Such approach makes the Neumann and the Robin boundary
condition computation straightforward and CPU effective, again only N flops are needed to
evaluate it, without any kind of special computational treatment on or near boundaries.

For the temporal discretization we use a two-level explicit time stepping
( ) ( )
0
0 0 0 0 0 0 0
D S
t
u u
u u

= V V V +
A
v , (18)
where zero-indexed quantities stand for the values at the initial time, and , D S for general
diffusion coefficient, and source term, respectively. The time step is denoted with t A . The
pressure-velocity coupling is performed through the correction of the intermediate velocity
( )
v
( ) ( )
0 0 0 0 0 0
( )
t
P

A
= + V +V V + V v v v b v v (19)
The equation (19) did not take into account the mass continuity and respective corrections
need to be applied
1

m m +
= + v v v , (20)
1

m m
P P P
+
= + ,

(21)
where , and m v P stand for iteration index, velocity correction and pressure correction,
respectively. By combining the momentum and the mass continuity equations, the pressure
correction Poisson equation emerges
2

m
t
P

A
V = V v . (22)
Gregor Kosec and Boidar arler






10







Instead of solving the global Poisson equation exactly, the pressure correction is guessed from
the divergence of the intermediate velocity.
2

m
P
t

= V
A
v . (23)
The proposed assumption makes solving of the pressure velocity coupling iteration local.
Such an approach is very CPU efficient, as it needs only one computation for each pressure
correction. With the computed pressure correction the pressure and the velocity can be
corrected as
1

m m
P P P ,
+
= + , (24)
where , stands for the relaxation parameter. The iteration is performed until the criterion

V
c V < v is met in all computational nodes. The approach is similar to the artificial
compressibility method (ACM) (Massarotti, et al., 1998, Rahman and Siikonen, 2008) and in
the framework of the Finite Difference Method to the SOLA approach (Hong, 2004).
However, the proposed approach retains the correct time transient which is not the case in
SOLA and ACM approaches. The free fluid flow situations have been tackled by ACM in
(Traivivatana, et al., 2007) and the flow in porous media in (Malan and Lewis, 2011). In the
present paper we are particularly interested in proper transient response of the computations.
The proposed solution procedure is effective from computational point of view as all
numerical operations are completely local. Despite several degrees of freedom over the spatial
discretization, its complexity remains comparable to Finite Difference Method or Finite
Volume Method.
Gregor Kosec and Boidar arler






11








Figure 2: Schematic representation of meshless numerical principle. The differential
operations in a circled node are performed only through the consideration of a local influence
domain.
4 NUMERICAL RESULTS
The results of the benchmark tests are assessed in terms of streamfunction and cavity Nusselt
number, with dimensionless variables, defined as

y
x
x y
W H
p
p
p p = =
O O
, (25)

x W p y H p
x y
v c v c
v v


O O
= = , (26)

C
H C
T T
T
T T

, (27)

2
p H
t t
c

=
O
. (28)
Gregor Kosec and Boidar arler






12







( )
1
0
( )
x y
v dp =
}
p p , (29)

( )
Nu( ) ( ) ( )
x
x
T
v T
p
c
= +
c
p
p p p . (30)

The Nusselt number is computed locally on five nodded influence domains, while the
streamfunction is computed on the one dimensional influence domains representing each row,
where all nodes in the row are used as an influence domain. The streamfunction is set to zero
in south west corner of the domain ( ) 0, 0 0 = . The presented solution procedure is first
verified through the comparison of results against previously published data. The similar
oscillatory low Prandtl number flow is considered, where a tall cavity with height/width
aspect ratio A=4 is filled with Pr=0.0137 (Al4.5%Cu alloy) fluid and simulated at
5
Ra=2.81 10 . Current numerical approach is compared against finite volume method and
Chebishev spectral method (Zalonik, et al., 2005). In present paper the results are presented
in terms of hot side mean Nusselt number and its transformation into frequency domain
(Figure 4). From Figure 4 it can be seen that our present numerical approach shows good
agreement with more standard numerical solutions. In the present work computation is
performed on 40497 uniformly distributed nodes. This case has already been solved in (Divo
and Kassab, 2007, Kosec and arler, 2009). We additionally present the frequency analysis of
the case in the present paper. In Figure 3 the streamlines and the temperature contour plots for
tall cavity test at different times are presented.

Gregor Kosec and Boidar arler






13








Figure 3: Streamlines (stream step 0.2) and temperature contours (contour step 0.1) for tall
cavity test.


Figure 4: Comparison case: Left: hot side mean Nusselt number as a function of
dimensionless time. Right: hot side mean Nusselt number as a function of dimensionless
frequency.

The numerical setup of present cases is presented in Table I together with some characteristic
Gregor Kosec and Boidar arler






14







results of numerical solution. Dimensionless pressure velocity coupling relaxation parameter
is set to the same numeric value as a dimensionless time step in all computations. The
dimensionless pressure velocity coupling criterion
V
c is set to 0.5 for all computations. Free
RBF shape parameter
C
o is set to 90 in all computations. In our computations Case 1 results
in steady state while Case 2 produces oscillatory solution. In Table I the maximum
streamfunction and hot side mean Nusselt numbers in steady state are presented for Case 1.
For Case 2 the values averaged over | |
5, 7 t e are stated. Figure 4 depicts excellent agreement
of the present method with the FEM and spectral method. Solution of Case 1 is depicted in
Figure 5, where the streamlines and the temperature contour plots are plotted on the left
figure. The temporal development of the hot side mean Nusselt number is presented on the
right figure. In Figure 6 the streamlines and the temperature contour plots for Case 2 at
different times during one oscillation are presented. In Figure 7 the hot side mean Nusselt
number temporal development and in Figure 8 its representation in the frequency domain are
presented, where f stands for dimensionless frequency. One can observe from Figure 6 an
almost same maximum flow intensity during oscillations. One can see from Figure 7 that
increase of the number of the computational nodes also increases the amplitude of the
oscillations. Both cases show good convergent behaviour with respect to the number of the
discretization nodes (Figure 9). In Case 2 it is evident that using too coarse nodal distribution
the important part of the solution can be missed as with the coarsest computation the strongest
oscillation is not captured. The oscillation is more pronounced with the increasing number of
the nodes. The oscillations are fully developed after initial transient which ends roughly at
4.5 t .

Table I: Discretization parameters and principal results. Since Case 2 does not exhibit steady
state, a mean value over one oscillation period is given.
Ra Pr A N
D
, dt ,
max

hot side
mean
Nu

Case 1

4
10

0.01 1 437
4
10

4.3358 2.0615
S
t
e
a
d
y

s
t
a
t
e






v
a
l
u
e
s


2597
5
5 10

4.4735 1.9496
10197
5
10

4.5896 1.9523
40397
6
5 10

4.6328 1.9572
160797
6
10

4.6444 1.9584
Case 2
4
5 10

0.01 1 437
5
5 10

6.7741 3.3319





M
e
a
n









v
a
l
u
e
s


2597
5
5 10

7.0020 2.8477
10197
5
10

7.1591 2.7831
40397
6
5 10

7.2786 2.7946
160797
6
10

7.3123 2.7991
Gregor Kosec and Boidar arler






15










Figure 5: Case 1: Left: Steady state streamlines (stream step 0.5) and temperature contours
(contour step 0.1). Right: hot side mean Nusselt number as a function of dimensionless
time.

Gregor Kosec and Boidar arler






16









Figure 6: Case 2: Streamlines (stream step 0.8) and temperature contours (contour step 0.1)
as a function of dimensionless time during one oscillation period.

Gregor Kosec and Boidar arler






17







Figure 7: Case 2: Hot side mean Nusselt number as a function of dimensionless time.

Figure 8: Case 2: Hot side mean Nusselt number as a function of the number of the domain
nodes in frequency domain.

Figure 9: Maximal streamfunction (left) and hot side mean Nusselt number (right) as a
function of the number of domain nodes, where all quantities are normalized to one at the
finest calculation.
Gregor Kosec and Boidar arler






18







The oscillations in Case 2 are result of a balance between the buoyancy and shear stress
forces.
5 CONCLUSIONS
In the present paper we demonstrate the application of extremely simple and intuitive
meshless numerical approach towards solution of transient thermo-fluid problems. The first
two numerical problems from the spectra of recent solidification oriented benchmark call are
treated. The low Prandtl number natural convection in a closed domain is solved with the
proposed numerical solution procedure. Two cases are considered; one with steady state and
another with oscillatory solution. In both situations we show good convergence behaviour.
Until now, no other reference solution exists of the treated cases. In addition, we compare the
present numerical approach on the treated oscillatory case and show excellent agreement with
more common global numerical approaches. We show that in oscillatory case the coarse nodal
distributions miss relevant physical behaviour. In our recent related work we research
application of adaptive nodal distribution in connection with the method used in this paper
(Kosec, 2011). More involved solidification benchmark call tests will be attempted by this
adaptive numerical approach. The future work will also be focused on a parallel
implementation of current solver as we already achieve good speedups with straightforward
OpenMP based parallelization. Further steps will be taken in a GPU and MPI based
parallelization schemes.

Acknowledgment:
The research was funded through Slovenian state research projects J2-4120, P2-0379 and P2-
0095. The Centre of Excellence for Biosensors, Instrumentation and Process Control is an
operation financed by the European Union, European Regional Development Fund and
Republic of Slovenia, Ministry of Higher Education, Science and Technology. The financial
support is kindly acknowledged.
6 REFERENCES
Ahmad, N., Combeau, H., Desbiolles, J. L., Jalanti, T., Lesoult, G., Rappaz, J., Rappaz, M.
and Stomp, C. (1998), "Numerical simulation of macrosegreation: a comparison
between finite volume method and finite element method predictions and a
confrontation with experiments", Metallurgical and Materials Transactions, Vol. A29,
pp. 617.

Amimul, A. (2011), Convection and Conduction Heat Transfer, Intech, Rijeka.
Gregor Kosec and Boidar arler






19








Arcidiacono, S.; Ciofalo, M. (2001a), "Low-Prandtl number natural convection in
volumetrically heated rectangular enclosures III. Shallow cavity, AR = 0.25",
International Journal of Heat and Mass Transfer, Vol. 44, pp. 3035-3065.

Arcidiacono, S.; Di Piazza, I.;Ciofalo, M. (2001b), "Low-Prandtl number natural convection
in volumetrically heated rectangular enclosures II. Square cavity, AR = 1",
International Journal of Heat and Mass Transfer, Vol. 44, pp. 537-550.

Atluri, S. N. and Shen, S. (2002a), The Meshless Method, Tech Science Press, Encino.

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

Bellet, M., Combeau, H., Fautrelle, Y., Gobin, D., Rady, M., Arquis, E., Budenkova, O.,
Dussoubs, B., Duterrail, Y., Kumar, A., Gandin, C. A., Goyeau, B., Mosbah, S. and
Zaloznik, M. (2009), "Call for contributions to a numerical benchmark problem for 2D
columnar solidification of binary alloys", International Journal of Thermal Sciences,
Vol. 48, pp. 2013-2016.

Buhmann, M. D. (2000), Radial Basis Functions, Cambridge University Press, Cambridge.

Chen, W. (2002), "New RBF collocation schemes and kernel RBFs with applications",
Lecture Notes in Computational Science and Engineering, Vol. 26, pp. 75-86.

Cless, C.M.;Prescott, PJ., (1996a), "Effect of time varying thermal boundary conditions on
oscillatory natural convection of a low-Prandtl-number fluid", Numerical Heat
Transfer, Vol. A29, pp. 645-669.

Cless, C.M.;Prescott, PJ., (1996b), "Effect of time marching schemes on predictions of
oscillatory natural convection ", Numerical Heat Transfer, Vol. A29, pp. 575-597.

Crunkleton, D. W.; Narayanan, R.; Anderson, T. J. (2006), "Numerical simulations of
periodic flow oscillations in low Prandtl number fluids", International Journal of Heat
and Mass Transfer, Vol. 49 pp. 427438.

Gregor Kosec and Boidar arler






20







de Vahl Davis, G. (1983), "Natural convection of air in a square cavity: a bench mark
numerical solution", International Journal for Numerical Methods in Fluids, Vol. 3,
pp. 249-264.

Di Piazza, I.; Ciofalo, M. (2001), "Low-Prandtl number natural convection in volumetrically
heated rectangular enclosures II. Slender cavity, AR = 4", International Journal of
Heat and Mass Transfer, Vol. 43, pp. 3027-3051.

Divo, E. and Kassab, A. J. (2007), "Localized meshless modeling of natural-convective
viscous flows", Numerical Heat Transfer, Vol. B129, pp. 486-509.

Fasshauer, G. (2006), "Radial basis functions and related multivariate meshfree
approximation methods: Theory and applications - Preface", Computers &
Mathematics with Applications, Vol. 51, pp. 1223-1366.

Franke, J. (1982), "Scattered data interpolation: tests of some methods", Mathematics of
Computation, Vol. 48, pp. 181-200.

Gobin, D. and Le Qur, P. (2000), "Melting from an isothermal vertical wall, synthesis of a
numerical comparison excercise", Computer Assisted Mechanics and Engineering
Sciences Vol. 7, pp. 289-306.

Gu, G.R. Liu; Y.T. (2005), An Introduction to Meshfree Methods and Their Programming,
Springer, Dordrecht.

Hong, C. P. (2004), Computer Modelling of Heat and Fluid Flow Materials Processing,
Institute of Physics Publishing, Bristol.

Kansa, E. J. (1990a), "Multiquadrics - a scattered data approximation scheme with application
to computational fluid dynamics, part I", Computers and Mathematics with
Applications, Vol. 19, pp. 127-145.

Kansa, E. J. (1990b), "Multiquadrics - a scattered data approximation scheme with application
to computational fluid dynamics, part II", Computers and Mathematics with
Applications, Vol. 19, pp. 147-161.

Kosec, G. and arler, B. (2008), "Solution of thermo-fluid problems by collocation with local
Gregor Kosec and Boidar arler






21







pressure correction", International Journal of Numerical Methods for Heat & Fluid
Flow, Vol. 18, pp. 868-882.

Kosec, G. and arler, B. (2009), "Solution of phase change problems by collocation with local
pressure correction", CMES: Computer Modeling in Engineering & Sciences, Vol. 47,
pp. 191-216.

Kosec, G., Zalonik, M., arler, B. and Combeau, H. (2011), "A Meshless Approach Towards
Solution of Macrosegregation Phenomena", CMC: Computers, Materials, & Continua,
Vol. 580, pp. 1-27.

Kosec, G.;arler, B. (2011), "H-adaptive local radial basis function collocation meshless
method", Computers, Materials & Continua, Vol. 26, pp. 227-254.

Liu, G.R. (2003), Mesh Free Methods, CRC Press, Boca Raton.

Lorbiecka, A. Z. and arler, B. (2010), "Simulation of dendritic growth with different
orientation by using the point automata method", CMC: Computers, Materials, &
Continua, Vol. 18, pp. 69-103.

Malan, A. G. and Lewis, R. W. (2011), "An articial compressibility CBS method for
modelling heat transfer and uid ow in heterogeneous porous materials",
International Journal for Numerical Methods in Engineering, Vol. 87, pp. 412-423.

Massarotti, N. , Nithiarasu, P. and Zienkiewicz, O. C. (1998), "Characteristic-based-split
(CBS) algorithm for incompressible flow problems with heat transfer", International
Journal of Numerical Methods for Heat and Fluid Flow, Vol. 8, pp. 969.

Mohamad, A. A, ; Viskanta, R. (1991), "Transient natural convection of low-Prandtl-number
fluids in a differentially heated cavity", International Journal of Numerical Methods in
Fluids, Vol. 13, pp. 61-81.

Monaghan, J. J. (1988), "An Introduction to SPH", Computer Physics Communications, Vol.
48, pp. 89-96.

Nakano, A.; Ozoe, H; Churchill. S. W. (1998), "Numerical computation of natural convection
for a low-Prandtl-number fluid in a shallow rectangular region heated from below",
Gregor Kosec and Boidar arler






22







Chemical Engineering Journal, Vol. pp. 175-182.

Pesso, T. and Piva, S. (2009), "Laminar natural convection in a square cavity: Low Prandtl
numbers and large density differences", International Journal of Heat and Mass
Transfer, Vol. 52, pp. 10361043 %U
http://www.sciencedirect.com/science/article/pii/S0017931008004067.

Rahman, M. M. and Siikonen, T. (2008), "An articial compressibility method for viscous
incompressible and low Mach number ows", International Journal for Numerical
Methods in Engineering, Vol. 75, pp. 1320-1340.

Sammouda, H.;Belghith, A.; Surry, C. (1999), "Finite element simulation of transient natural
convection of low-Prandtl-number fluids in heated cavity", International Journal of
Numerical Methods for heat and Fluid Flow, Vol. 9, pp.

arler, B. and Vertnik, R. (2006), "Meshfree explicit local radial basis function collocation
method for diffusion problems", Computers and Mathematics with Applications, Vol.
51, pp. 1269-1282.

arler, B. (2007), From global to local radial basis function collocation method for transport
phenomena, Springer, Berlin, pp. 257-282.

terk, M. and Trobec, R. (2008), "Meshless solution of a diffusion equation with parameter
optimization and error analysis", Engineering Analysis with Boundary Elements, Vol.
32, pp. 567-577.

Traivivatana, S., Boonmarlert, P., Thee, P., Phongthanapanich, S. and Dechaumphai, P.
(2007), "Combined adaptive meshing technique and characteristic-based split
algorithm for viscous incompressible ow analysis", Applied Mathematics and
Mechanics, Vol. 28, pp. 1163-1172.

Trobec, R., terk, M. and Robi, B. (2009), "Computational complexity and parallelization of
the meshless local Petrov-Galerkin method", Computers & Structures, Vol. 87, pp. 81-
90.

Vertnik, R. and arler, B. (2006), "Meshless local radial basis function collocation method for
convective-diffusive solid-liquid phase change problems", International Journal of
Gregor Kosec and Boidar arler






23







Numerical Methods for Heat & Fluid Flow, Vol. 16, pp. 617-640.

Vertnik, R.; arler, B.; Seni, B. (2011), "Solution of macrosegregation in continuously cast
billets by a meshless method", The 3rd International Conference on Advances in
Solidification Processes.

Wang, J. G. and Liu, G. R. (2002), "A point interpolation meshless method based on radial
basis functions", International Journal for Numerical Methods in Engineering, Vol.
54, pp. 1623-1648.

Xiaohua, W.; Yingjie, W.;Xinrong, S. (2009), "Numerical investigation of the first bifurcation
for natural convection of fluids enclosed in a 2D square cavity with Pr lower than 1.0",
Energy Conversion and Management Vol. 50, pp. 2504-2512.

Xin, S.;Le Quere, P. (2002), "An extended chebyshev pseudo spectral benchmark for the
8:1 differentially heated cavity", International Journal of Numerical Methods in
Fluids, Vol. 40, pp. 981-98.

Zalonik, M., Xin, S. and arler, B. (2005), "Verification of a numerical model of
macrosegregation in direct chill casting", International Journal of Numerical Methods
for Heat & Fluid Flow, Vol. 18, pp. 308-324.