Methods using variable interchange
353
which a = 1, b = 5/3, H =
1,
h = 1/6, results obtained by the three-
dimensional algorithm in
x,
y,
z coordinates with
&f>
= 5y = 1/12 and
5z = 1/15 compared favorably with those for
the
cylindrical formulation
(8.83).
No formal study
of
the convergence of the iterative solution of the
non-linear finite difference equations proved possible. Instead, Ozis
(1981) quoted numerical experiments which demonstrated reasonable
convergence of some
of
the
numerical values obtained for the three-
dimensional formulation including the position of the free surface, start-
ing from parabolically interpolated x-values.
About
100 iterative cycles
were necessary
to
satisfy a convergence limit
of
Ix;:t,!-xu,kl.;;;10-
3
•
(ii) Multi-variable interchange. Jeppson (1972) gave an inverse formula-
tion
of
a three-dimensional dam problem in which
he
interchanged all
three
cartesian coordinates, x,
y,
z,
with a potential function and two
selected streamline functions.
In
the same paper Jeppson gave several
references to corresponding inverse solutions
of
two-dimensional inviscid
and porous flow problems, including some axisymmetric cases.
In
three
dimensions, Jeppson introduced two stream functions
1/1
and
1/1'
defined
to
be
orthogonal surfaces normal to equipotential surfaces and tangential'to
the
velocity vector such
that
their intersections define the streamlines of
the
flow.
The
potential function
</>
is
defined in
the
usual way.
The
inverse
formulation
is
in the
</>,
1/1,
1/1'
space in which three non-linear first-order
partial differential equations
are
to be solved for
the
three new dependent
variables
x =
x(</>,
1/1,
1/1'),
y =
y(</>,
1/1,
1/1'),
Z =
z(</>,
1/1,
1/1'),
subject to appro-
priately rewritten boundary conditions.
In
order
to
proceed with a finite-
difference solution Jeppson (1972) combined these equations by differen-
tiation, making the assumption that certain relatively small quantities
were known and remained
cons~ant
so
that
separate equations for
x,
y,
z
could
be
used in different planes in the transformed space.
He
discussed
factors influencing the choice of the approximated equations.
To
illustrate his method, Jeppson (1972) obtained solutions for three-
dimensional seepage through a dam with a partial toe drain, as in Fig.
8.13a.
The
transformed
</>,
1/1,
1/1'
space
is
shown
in
Fig. 8.13b, and a
graphical
flow
pattern in Fig. 8.13c. More detailed sectional pictures are
given in Jeppson's paper. No attention was paid
to
the singUlarities and
stagnation regions which occur near the bottom and vertical sides of the
dam
at
the
drain
end
of
the
flow.
In
the iterative procedure adopted, the
interior values were allowed
to
settle
to
one set
of
boundary conditions
before
the
latter were themselves improved iteratively. A line
SOR
method was used.
On
the face of it, this inverse method calls for more personal judgment
in
setting
up
the equations and more computational effort than the