The physical field around the exit planes of the CMT window can be seen
as a superposition of the outgoing guided modes of the bus core with
their constant amplitudes, and a remainder, that, when expanded
in the modal basis associated with the straight waveguide, is orthogonal to
the guided waves. The notions of ``vanishing interaction'' or
``decoupled'' fields, as used e.g. for the motivation of the assumptions
underlying the abstract framework of Section 1.4, are to be
concretized in precisely this way: The guided waves in the straight
core are stationary, iff projections onto the mode profiles at growing
propagation distances lead to constant amplitudes ,
(apart from the
phase changes according to the undisturbed propagation of the
respective modes).
Now the coupled mode theory formalism is limited to the few non-orthogonal
bend and straight modes included in the CMT ansatz, which are overlapping in
the regions of the input and exit ports A and B of the coupler. Consequently,
when the CMT procedures try to approximate both the guided and radiative
part of the real field, the optimum approximations may well be superpositions
with non-stationary coupled mode amplitudes
of the modes of the
bus waveguide. Indeed, as observed in Sections
3.4.1 and 3.4.2, the projected amplitudes
(or the related scattering matrix elements
) become stationary, when viewed as a function of the
exit port position
, while at the same time the associated CMT solution
(or the elements
of the transfer
matrix) exhibit an oscillatory behaviour. Still, in the sense of the
projections, one can speak of ``non-interacting, decoupled'' fields.
That justifies the limitation of the computational window to
-intervals
where the elements of
S (not necessarily of
T) attain
constant absolute values around the input and output planes.
In conclusion, it is at least partly misleading to stick to the familiar notion of ``mode evolutions'' computed by the CMT approach. If one abandons that viewpoint and regards the CMT procedures as just ``a'' method that generates an approximate field solution inside the computational window, then applying the projections to extract the external mode amplitudes appears perfectly reasonable.