170
Front-tracking methods
the
instability which develops as x increases and (4.20a) enters a closed
loop because it becomes very sensitive
to
rounding errors. Goodling and
Khader (1974) incorporated their finite-difference form of (4.18) into
the
system of equations
to
be
solved for an arbitrary value of
u~!n+1
which
is
then updated by (4. 17a). Gupta and Kumar (1981a) in a study of a
convective boundary condition
at
the
fixed end found this latter method
fails
to
converge as x increases because it
is
too sensitive
to
a small
change in
u",n+1'
Gupta and Kumar's (1981a) results are in good agree-
ment with those obtained by the other variable time-step methods and by
a Goodman's (1958) integral method (see §3.5.4).
Yuen and Kleinman (1980) used another way of updating
at
by
successive approximation.
Gupta and Kumar
(1981b) adapted the method of Douglas and Gallie
(1955)
to
solve
the
oxygen diffusion problem (see §1.3.10).
Other
authors
have
to
introduce interpolation
or
extrapolation procedures for calculat-
ing the total absorption time at which the moving boundary, marking
the
innermost penetration of oxygen, reaches
the
outer fixed surface.
In
Gupta and Kumar's method,
the
total absorption time emerges from
the
final step in
the
normal computing procedure. Their results for the
position of
the
moving boundary are compared with others in Table 4.1:
Hansen and Hougaard (1974) predicted the total absorption time
to
be
between 0.1972 and 0.1977. Gupta and Kumar (1981b) obtained 0.1973
which
is
close
to
the
accurate value of 0.197434 obtained by Dahmardah
and Mayers (1983) (see §5.1).
Kumar (1982) discussed in detail various variable time-step methods
and their application
to
problems mentioned above and
to
others includ-
ing phase-change with non-uniform initial temperatures, time-dependent
boundary conditions,
the
dissolution of a spherical gas bubble in a liquid,
and the freezing of liquid inside and outside a cylinder.
For
most
problems, numerical results obtained by several methods were compared.
4.3.2. Variable
space
grid
Murray and Landis (1959) kept
the
number of space intervals between
x = 0 and x = s(t), i.e. between a fixed and a moving boundary, constant
and equal
to
I,
say, for all time (Fig. 4.2). Thus, for equal space intervals,
8x =
s(t)/l
is different in each time step. The moving boundary is always
on
the
lth
grid line. They differentiated partially with respect
to
time
t,
following a given grid line instead of
at
constant
x.
Thus, for the line i 8x
they had
(4.21)
Murray and Landis assumed
that
a general grid point
at
x moved