The realization theory of discrete time linear representation systems is established. We have already established the realization theory of discrete and continuous time dynamical systems in (1) and (2). The main theorems in (1) and (2) are as follows. For any causal input/output map
A# of a Black-Box, there are canonical (reachable & distinguishable) dynamical systems which realize
A#. A dynamical system δ is a collection of Ω-module (
X, φ), an initial state
x0∈X, and a readout map
h: X→Y. If δ
1 and δ
2 are canonical realization of
A#, then there is a unique dynamical system isomorphism
T: δ
1→δ
2.
In this paper, we restrict the set
Y of output values to a linear space over a field
K, and consider only a subclass of dynamical systems, i.e., the class of dynamical systems δ=((
X, δ),
x0, h) such that (
X, φ) is a linear Ω-module,
x0∈X, and
h: X→Y is a linear map. Such dynamical systems δ are called linear representation systems (L. R. S.). We define that δ is quasi-reachable if the linear hull of the reachable set from
x0 is
X.
The main theorems state that for any causal (non-linear) input/output map, there exist canonical (quasi-reachable & distinguishable) L.R.S. realizations, and that if δ
1 and δ
2 are canonical L.R.S. realizations, then there exists a unique L.R.S. isomorphism
T: δ
1→δ
2. To prove the theorems, we have converted a L.R.S. δ=((
X, φ), x0, h) to a sophisticated L.R.S. Σ=((
X, φ), G, H), where (
X, φ) is an
A(Ω)-module,
G: (A(Ω), _??_)→(
X, φ) is a linear input map, and
H: (
X, φ)→(
F(Ω, Y), Sl) is a linear observation map.
View full abstract