![]() |
Figure 1 shows the flow variable and the flux
function
for
a constant-velocity medium. Both are two-dimensional functions, and
are drawn as vectors on the 2-D grid, with
, the horizontal
time gradient, pointing in the horizontal direction, and
,
the vertical time gradient, pointing in the vertical direction. The length
of the vectors corresponds to the magnitude of the flow.
Equation (3) describes the conservation of flow:
for each grid cell, the equation balances the change in horizontal flow
with the change in vertical flow.
Thus, the total flow through each grid cell, the vector sum
of
and
, is constant (Figure 1c).
The magnitude of the total flow is s, which
denotes the preserved ``substance'' for this problem.
(In many
fluid-mechanics problems, the flow variable is velocity, and the preserved
scalar quantity is energy.)
However, for inhomogeneous media s is not constant, and no classical solution exists. In these media, the flux function F is discontinuous at interfaces where velocity contrasts occur; flow is ``destroyed'' or ``created'' in grid cells along these interfaces. Possible nonclassical solutions to the problem would therefore also be discontinuous: gradient components of the traveltime field would locally obey Snell's law, describing rays that break when passing a velocity contrast.
As discussed before, nonclassical solutions are generally not accessible by finite-difference methods. Instead, viscosity solutions have to be computed, where a numerical ``viscous'' quantity is added to s, which is then preserved by the equation. (In fluid-mechanics problems with energy as the conserved quantity, viscosity allows dissipation of energy.) In computational fluid dynamics, many methods have been developed for computing these viscosity solutions (see Roache, 1976). The next section discusses one particular method.