332
Numerical solution
of
free-boundary problems
known exactly. Furthermore, by defining
f(c)={(1+c
2
)}!-.j1O,
we find
that
C(k+l)
= C(k)-f(C(k»)/f(c(k»), which
is
the Newton iterative formula for
the
solution of f(c) = 0 from the initial guess
c(o)..
Since f(c)
is
convex for
c~o
and
f(O)
<0
it follows that for any initial guess c(O)<O the sequence
of the trials
r(k)
converges quadratically
to
the
true free boundary. Thus
the simple problem lends support
to
Garabedian's claim (1956) that his
rewriting of
the
boundary conditions leads
to
quadratic convergence and
points
to
the
connection with Newton's method. Cryer (1968) found that
numerical
result~
for a more complicated problem in stellar evolution
behaved like those for the model problem which was chosen
to
resemble
it.
8.3. Boundary singularities
Singularities commonly occur
on
a fixed boundary in boundary-value
problems. They may be associated with sudden changes in the direction of
the boundary, as
at
a re-entrant corner,
or
with mixed boundary condi-
tions. Much attention has been paid
to
these singularities and methods
include mesh refinement for both finite differences and finite elements,
and
the
use of modified finite-difference approximations
or
singular finite
elements in which the local analytical form of
the
singularity is somehow
incorporated. Various other techniques are based
on
integral equations,
power series, dual series, Fourier series, and removal of
the
singularity.
Conformal transformation methods have proved particularly efficient and
highly accurate for
the
solution
of
elliptic problems.
Good
summaries
together with
other
basic references are given by Fox (1979) for finite
differences, by Wait (1979) for finite elements, Delves (1979) for global
and regional methods, and by Scheffler and Whiteman (1979) for
conformal-mapping techniques. Many other references
to
these and other
methods are quoted by Furzeland
(1977b).
In
addition
to
such generally occurring singularities, free-boundary
problems often have a singularity
at
the separation point where
the
free
boundary meets a fixed boundary, as, for example,
at
point D in Fig. 2.1
for the simple dam problem.
Aitchi~on
(1972) used complex variable
methods
to
determine
the
shape of
the
free boundary near the separation
point and incorporated the local analytical solution into a finite-difference
scheme in order to improve its accuracy.
Starting with
the
dam problem defined by equations (2.7-11) in Chap-
ter
2, but taking
Yl
= 1,
Y2
= d, and
Xl
= L in Fig. 2.1, Aitchison (1972)
moved the origin of
the
(x,
y) coordinates
to
D and added a constant
to
c(J
so that
cfJo
=
O.
In terms of the stream function
"',
the condition
ac(Jlan
= 0
on
DF
becomes '" = 0,
and
the problem is
to
determine the complex variable