Beruflich Dokumente
Kultur Dokumente
www.elsevier.com/locate/amc
Department of Mathematics, Faculty of Arts and Science, Inonu University, 44069 Malatya, Turkey
Abstract
In this paper, the enthalpy method based on suitable nite dierence approximations
has been applied to the one-dimensional moving boundary problem with a Neumanntype boundary condition known as the Stefan problem. The numerical results obtained
by the hopscotch technique are compared with the exact solution of the problem. It is
shown that all results are found to be in very good agreement with each other, and the
numerical solution displays the expected convergence to the exact one as the mesh size is
rened.
2002 Elsevier Inc. All rights reserved.
Keywords: Enthalpy method; Moving boundary; Stefan problem; Hopscotch technique
1. Introduction
Many important heat conduction or diusion problems arising in science,
engineering and industry involve a phase change or a moving boundary. A
common example of such problems is the problem of melting of ice that was
rst treated by Stefan [1] and after whom this kind of problems are widely
referred as Stefan problems. Since the moving boundary (interface) separating
liquid and solid regions of the material changes with time, its location which is
not known a priori and so needs to be determined as a part of the solution.
Corresponding author.
E-mail address: skutluay@inonu.edu.tr (S. Kutluay).
0096-3003/$ - see front matter 2002 Elsevier Inc. All rights reserved.
doi:10.1016/S0096-3003(02)00846-9
322
However, such problems are highly non-linear and their exact solutions seem
to be inconceivable except in some very simple cases. The exact solutions of the
one-dimensional Stefan problems have been surveyed by Hill [2].
In view of the diculty in obtaining exact solution, numerical methods are
far more common. As a result, in recent years, these non-linear problems have
motivated considerable research by many authors. To tackle dicult phase
change problems, many numerical methods such as nite dierence, nite element and heat balance integral methods [311] have been and are still being
developed. A comprehensive review of a wide range of these methods may be
found in [12].
In this paper, an enthalpy formulation based on suitable nite dierence
approximations has been applied to one-dimensional Stefan problem with a
Neumann condition at the xed boundary suggested by Homan [13].
oT
o2 T
K 2;
os
ov
oT
h~Tw ;
ov
T Ss; s Tf ;
v 0; s > 0;
v Ss; s > 0
where q is the liquid density (assumed constant), c is the specic heat capacity,
K is the liquid thermal conductivity (assumed constant), and h~ is the prescribed
function of s > 0.
The location of the moving interface is governed by the heat balance
equation known as Stefan condition
Lq
dS
oT
K
;
ds
ov
v Ss; s > 0
323
where L is the latent heat of fusion. The term on the right hand side of Eq. (1) is
called the heat ux which represents heat ow per unit area per unit time of an
isothermal surface (that is, a surface of a constant temperature) in the decreasing temperature direction. Therefore, the heat ux is a positive quantity
since the heat ow is prescribed in the positive v direction, 0 < v < 1.
Initially there is no liquid region which implies the condition
S0 0;
s0
In the above equations, the variables and heat parameters are in terms of
physical units. It is convenient to non-dimensionalise the problem by introducing the change of variables
v
x ;
S
s ;
js
;
2
T Tf
;
Tw
K
qc
subject to
oU
expt;
ox
U x; t 0;
x 0; t > 0
x st; t > 0
x st; t > 0
subject to
s0 0;
t0
tP0
324
The exact solution (4) can be used to compare the numerical solutions and can
also be used to initialise the numerical scheme at some t > 0.
oH
o2 T
K 2;
os
ov
325
H cTf
;
cTw
E<0
Utilising Eq. (7) and using Eqs. (3) and (6) then becomes
E a; E > a
U
0;
06E6a
Hence the non-dimensional enthalpy formulation of Eq. (7) describing the heat
ow becomes
oE o2 U
2 ;
ot
ox
subject to
oU
expt; x 0; t > 0
ox
U ; t 0; x ; t > 0
10
where > st. Thus, standard nite dierence schemes may now be applied to
the whole region of the problem. Since we are interested in the temperature
distributions in the liquid region, the temperature U x; t can easily be found by
virtue of Eq. (8) after computing setting the values of E.
4. Method of solution
The explicit nite dierence scheme of the problem provides a relatively
straightforward expression for the determination of the unknowns temperatures at any time step, but it limits the size of the time step.
The implicit dierence replacement is stable for all sizes of time step, thus
there is no size restriction on time step. However, it requires the solution of a
system of simultaneous algebraic equations.
The oddeven Hopscotch algorithm proposed by Gourlay [19] combines
explicit and implicit nite dierence replacements at alternate mesh points in
such a way that the overall scheme does not require the solution of a system of
algebraic equations, that is, it is overall explicit. The method is also unconditionally stable.
326
Eij1
Eij
j
rUi1
2Uij
j
Ui1
;
i0
i 1; 2; . . . ; N 1
11
kek1
j
N 1
1 X
1 Ui ;
N i1
U xi ; tj
e e1 ; e2 ; . . . ; eN 1
327
Table 1
Values of the temperature distribution as predicted by the numerical and exact solutions at tf 0:9
and various mesh sizes for r 0:4
Explicit
h 0:1
Hopscotch
h 0:1
h 0:05
h 0:01
Exact
solution
1.50022
1.26667
1.05554
0.86450
0.69127
0.53341
0.38833
0.25320
0.12486
0.00926
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.49909
1.26552
1.05438
0.86336
0.69014
0.53239
0.38746
0.25258
0.12658
0.00925
1.48121
1.24725
1.03559
0.84415
0.67101
0.51424
0.37172
0.24081
0.11822
0.00874
1.46407
1.23004
1.01828
0.82667
0.65329
0.49641
0.35448
0.22608
0.10935
0.00108
1.45960
1.22554
1.01375
0.82212
0.64872
0.49182
0.34986
0.22140
0.10517
0.0
kek1
0.08376
0.05139
0.01444
Table 2
Predicted location of the interface at tf 0:9 for r 0:4
Hopscotch
Exact solution
Error (%)
0.1
0.05
0.01
0.9003703
0.9000937
0.8999822
0.0411
0.0104
0.0001
0.9
328
Table 3
Values of the temperature distribution as predicted by the numerical and exact solutions at tf 0:9
and various mesh sizes for r 2:0
Explicit
Hopscotch
h 0:1
h 0:1
h 0:05
h 0:01
Exact
solution
Method unstable
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.50235
1.26642
1.05157
0.85846
0.69746
0.55766
0.40828
0.26326
0.14908
0.04756
1.48867
1.25465
1.04185
0.84934
0.67776
0.52343
0.37981
0.24727
0.12183
0.04277
1.46450
1.23045
1.01868
0.82707
0.65369
0.49681
0.35487
0.22651
0.10969
0.00108
1.45960
1.22554
1.01375
0.82212
0.64872
0.49182
0.34986
0.22140
0.10517
0.0
kek1
0.11545
0.06128
0.01545
Table 4
Predicted location of the interface at tf 0:9 for r 2:0
Hopscotch
Exact solution
Error (%)
0.1
0.05
0.01
0.9095119
0.9002139
0.8999707
0.1056
0.0237
0.0003
0.9
the number of iteration corresponding to the same mesh sizes decreases when
the stability parameter r increases.
In conclusion, we have demonstrated that the enthalpy approach can be
used eectively in problems with a Neumann condition at the xed boundary
and that the hopscotch scheme is an ecient solver for such problems since
hopscotch dierence approximation produces the expected converging numerical solutions for r > 0.
References
[1] J. Stefan, Uber die theorie der eisbildung imbesondee uber die eisbindung im polarmeere, Ann.
Phys. U. Chem. 42 (1891) 269286.
[2] J. Hill, One-dimensional Stefan Problem: An Introduction, Longman Scientic and Technical,
Essex, 1987.
[3] J. Crank, Two methods for the numerical solution of moving boundary problems in diusion
and heat ow, J. Mech. Appl. Math. 10 (1957) 220231.
[4] J. Crank, R.S. Gupta, A moving boundary problem arising from the diusion in absorbing
tissue, J. Inst. Math. Appl. 10 (1972) 1933.
329