1: \section{Scattering channels}\label{S:sec2}
2: Any extension of the TLM method that includes heat transfer, fluid flow,
3: or particle current, for instance, involves scattering channels other than
4: transmission lines. The latter, for a non vanishing real part of the
5: characteristic line impedance, inherently impose wave propagation between
6: cells. Degenerate lines, with a purely imaginary impedance,
7: still work in diffusion models, cf. \cite{Ch1}, chap.7. Other types of
8: transport or modes of propagation, such as for example the relativistic
9: charged particle current treated in \cite{He5}, are very unnaturally and
10: more or less imperfectly modeled using transmission lines. There is good
11: reason to get rid of lines in such and other cases within an extended
12: framework.
13:
14: A first step towards the definition of more general scattering channels in
15: TLM has been undertaken in replacing transmission lines with abstract
16: projections into in- and outgoing field components, cf. \cite{He5}. It was
17: postulated in this paper that the propagating fields ('link quantities')
18: allow of a decomposition into a direct sum
19: \begin{equation}\label{2.1}
20: z = z_{in} \oplus z_{out} \quad ,
21: \end{equation}
22: $z_{in}$ and $z_{out}$ representing the incident and outgoing fields,
23: respectively. Moreover, it is essential in our understanding of TLM
24: that the latter have a merely operational meaning in that only the
25: total field $z$ enters the dynamical model equations (cf. sections
26: \ref{S:sec3},\ref{S:sec4}).
27: In singular cases, a physical interpretation can yet still be given
28: to $z_{in}$, $z_{out}$ on the basis of a special analysis,
29: e.g. \cite{He4}, Corollary \nolinebreak 2. \\
30: For the Maxwell field model the technical passage from the transmission
31: line formulation to the projection operator setting is outlined in
32: \cite{He5}, Appendix ~A.
33:
34: In the classical TLM setup the connection map simply transfers
35: without further modifications the quantities outgoing from a cell
36: into quantities incident at neigbouring cells or rejected at some
37: totaly reflecting electric or magnetic wall.
38: This is in perfect harmony with the behaviour of a propagating
39: electromagnetic field, the components of which are tangential to the cell
40: boundary (as the link quantities always are in a classical TLM cell,
41: cf. \cite{He1}) and that is thus not subject to refractive scattering
42: at the cell face, even if the medium changes there.
43:
44: The situation is clearly not thus simple for arbitrary propagating quantities.
45: To circumvent any modeling restrictions, the connection cycle of a
46: non-trivial DSC scheme comprises cell interface scattering from the outset.
47: Nodal and cell face scattering thus enter a kind of duality relation
48: that becomes visible, for instance, in an apparent symmetry of the model
49: equations in their most general form (\ref{3.16}, \ref{4.3n}, \ref{4.3p}).
50: Nodal and cell boundary scattering may in fact be of equal importance and
51: sometimes boundary scattering plays even the leading r\^ole in a DSC
52: algorithm.
53:
54: In the generalized setup, just as in the traditional TLM framework,
55: scattering channels interconnect a {\it node\/}, viz. a suitably defined
56: {\it centre\/} of a mesh cell, with {\it ports\/} at the cell boundary.
57: The channels are yet no longer represented by transmission lines.
58: With respect to a computed physical field in D-dimensional configuration
59: space, they simply form a pair of scalar or vector valued distributions,
60: transposed over a distance in space, which test the field within the
61: cell and on its boundary.
62: A DSC scattering channel will thus be defined, precisely, as a pair of
63: continuous linear functions (${\,p \,, \; p \sptilde\,}$) which act
64: on a class of (suitably smooth real or complex) vector fields $Z$
65: in configuration space, such that $p$ has its support on a cell
66: face and ${p \sptilde}$ is connected to $p$ via pull back into the node,
67: i.e.: Given any notions of \emph{centre} of cell and face, as well as
68: the spatial translation
69: ${s: \mathbb{R}^D \rightarrow \mathbb{R}^D}$
70: that shifts the centre of a cell (\emph{node\/}) into the centre
71: of the face where $p$ has its support, then the {\it nodal image}
72: $p\sptilde$ of $p$ is defined as the distribution
73: \begin{equation}\label{2.2}
74: (\, p\sptilde \, , \; Z \, ) \, :\, = \, ( \, p \circ s \, , \; Z \,) \,
75: = \, ( \, p \,, \; Z \circ s^{-1} \, ) \; ,
76: \end{equation}
77: and the pair (${ \, p,\; p\sptilde\,}$) is called a {\it scattering channel}.
78: Equivalently, a scattering channel can of course be identified with
79: (${\, p \, , \; s\,}$) or even simply with the \emph{port p\/}, the pertinent
80: shift and nodal image then being tacitly understood.
81: The concept should in fact not be handled in too rigid a fashion - ~ and
82: there is no need to do so.
83: In certain applications the support of the port distribution may be
84: extended over a neighbourhood of a face, or the node distribution may
85: rather be thought of as a mean over the entire cell (in the way familiar
86: from finite volume methods).
87: \input{fig1.tex}
88: Needless to say that the stressed duality between nodal and cell boundary
89: scattering is not to be misunderstood in the narrow sense of category
90: theory.
91: Here, it refers simply to the observation that a set of propositions are
92: valid, modulo symmetry in certain terms, in the two scattering situations
93: - \nolinebreak which of course reflects the paired distribution concept
94: of scattering channel and the already mentioned symmetry of the pertinent
95: model equations in their most general form.
96: A parallel symmetry then clearly characterizes the structure of the
97: reflection and connection maps that solve these equations.
98:
99: Cell boundary scattering is, for the rest, not thus new an option:
100: Already in the TLM model for superconducting boundary \cite {He0}
101: cell face s-parameters and boundary stubs have been introduced for
102: solving the discretized London equations, cf. also \cite{He4}.
103:
104: Despite the abolition of transmission lines, by their replacement
105: with abstract scattering channels the computed ('physical') fields
106: can still be represented -- in the way familiar from the classical
107: TLM method -- as sums of \emph{in- and outgoing} scalar or vector fields
108: \begin{equation}\label{2.3}\centering
109: z \; \; = \; \; z_{in} \, + \, z_{out} \quad .
110: \end{equation}
111: No physical interpretation or propagation property is, however, in general
112: ascribed to $z_{in,\, out}$.
113: In fact, these quantities are merely operationally defined by means of the well
114: known \emph{Johns cycle} of iteration
115: \begin{equation}\label{2.4}\centering
116: \begin{split}
117: \begin{aligned}
118: z_{in} \; &= \; ( \; \mathcal{C} [z_{out}] + e \; )&&\longleftarrow && \\
119: z^{p} \; &= \; z_{in} + z_{out} && &&\; \, \uparrow \\
120: &\, \downarrow \; t + \tau/2 && &&t + \tau \\
121: z_{out} \; &= \; \mathcal{R} [z_{in}] && &&\; \, \uparrow \\
122: z^{n} \; &= \; z_{in} + z_{out} &&\longrightarrow && \qquad \, .
123: \end{aligned}
124: \end{split}
125: \end{equation}
126: $\mathcal{R}$ and $\mathcal{C}$ denote the node and cell face propagators
127: (or so-called reflection and connection maps - the latter including
128: now cell boundary scattering, and $e = e(t)$ induces any excitation.
129: Note again that $z_{in}$, $z_{out}\,$ are so far \emph{purely operational}
130: quantities,
131: i.e. only the total fields $z$ enter the model equations, while $z_{in}$,
132: $z_{out}$ are {\it in general\/} bare of any physical meaning (a physical
133: interpretation in terms of an energy flow still exists within the classical
134: Maxwell field TLM model, cf. \cite{He4}, Corollary \nolinebreak 2).
135:
136: As will be seen in the next section, the structures of the propagators
137: ${\,\mathcal{R}\,}$ and ${\,\mathcal{C}\,}$ are very similar in the
138: general DSC scheme, thus reflecting the dual r\^ole that nodal and
139: boundary scattering play therein.
140: In a sense, precised in section \nolinebreak \ref{S:sec3},
141: ${\,\mathcal{R}\,}$ and ${\,\mathcal{C}\,}$ are the discrete
142: \emph{convolution integrals\/}
143: that in every Johns cycle strictly solve the model equations.
144: In fact, the Johns cycle can be looked at as basically a two-step
145: convolution method for solving certain types of explicit finite
146: difference equations in time.
147:
148: Note that the model equations can in principle be directly solved
149: inasmuch as they provide complete recurrence relations, cf.
150: sections \nolinebreak \ref{S:sec3}, \ref{S:sec4}.
151: The scattering approach (using Johns' cycle of convolutions) offers,
152: however, important advantages. Thus, it provides clear cut reliable
153: stability criteria that make the DSC algorithm unconditionally stable
154: under very general circumstances ~\cite{He7}.
155: