The detailed expressions of the determinants of
are given in appendix A. The six-component vector
is obtained by combining the matrix
and the vector
in the same way as equations (A.12) for
and
. The computation is done for each layer for bottom to top up to the first one at the free surface. As
and
(appendix A) are equal (from equation (3.31)), it is obvious that
. From bottom to top, the two components are always equal (equation (3.30)) and we can reduce the number of components to five rather than six. Also,
like
is the only imaginary component and this feature is preserved across the layered medium. Thus, the 6 components of
reduce to 5 and the matrix
to 5x5 components.
To speed up the computation, we slightly modified the Dunkin's original formulation to reduce the total number of operations, preferring subtractions, additions and multiplications to divisions. The
and
functions are calculated in the same way as for Love case, including the real and imaginary cases (equation (3.22), section 3.1.3). A frequency factor of
(equation (A.12), appendix A) has been introduced in
to avoid unscaled vector at low frequencies. For each layer, values of sub-determinants are scaled to fit in the range between
to
to avoid overflow when propagating across a stack with many layers. Compared to Herrmann's formulation (1994) in the same conditions (not in its original Fortran code but already translated in C++), this implementation reduces by 25% the time consumption.
In figure 3.3, the values taken by
for all couples
are shown in the case of a two-layer model: 1350 and 250 m/s for
and
, respectively in the first 25 m thick layer, and 2000 and 1000 m/s in the half-space.