Professional Documents
Culture Documents
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 43, NO. 11, NOVEMBER 1998
AbstractIn the finitely recursive process (FRP) model of discrete event systems (DES), concepts about processes and process
operators have been introduced. An infinite set of process expressions or functions can be built recursively through function
composition using a few elementary operators. Given any process
realization, it is important to find out whether the process is
bounded, i.e., whether it has a finite state realization. In the
FRP setting this translates to the problem of finding out whether
the set of post-process expressions is finite or not. In Cieslak
and Varaiya (1990) it has been shown that the boundedness
problem is undecidable for general FRPs. This paper investigates
the decidability of the problem for subclasses of FRP. In Inan
and Varaiya (1988), it was conjectured that the set of functions
that can be recursively generated using the parallel composition
operator (PCO) and different change operators [i.e., without
using the Sequential Composition Operator (SCO)] will be finite
and FRPs constructed over this set of functions will naturally
be bounded. In the present work a counterexample has been
provided to disprove the conjecture about the finiteness of the
above set of functions. However, using a suitable post-process
computation procedure, it has been shown here that the FRPs,
built recursively over this set of functions, are bounded.
Index TermsBoundedness, discrete-event systems, finitely recursive processes, process algebra models.
terminating
process
NOMENCLATURE
Fixed finite collection of event symbols.
Set of finite length strings (also
called traces) formed with elements
of
Null string.
String containing a single event
String containing
Concatenation of the string
Followed by string
Language
Prefix-closed
Prefix closed if
Family of prefix-closed languages
over
Length of the string
repeated times.
Manuscript received October 25, 1996; revised December 2, 1997. Recommended by Associate Editor, E. K. P. Chong. This work was supported in part
by the University Grants Commissions Young Teachers Career Development
Award to S. Mukhopadhyay.
S. Bose was with the Department of Electrical Engineering, Indian Institute
of Technology, Kharagpur 721302, India. He is now with Siemens Information
Systems Limited, Bangalore, India.
A. Patra and S. Mukhopadhyay are with the Department of Electrical
Engineering, Indian Institute of Technology, Kharagpur 721302, India.
Publisher Item Identifier S 0018-9286(98)08494-3.
00189286/98$10.00 1998 IEEE
Global
change
operator
process
space
over
Syntactic
transformations
over
Closure
relation
over
1515
1516
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 43, NO. 11, NOVEMBER 1998
and
II. PRELIMINARIES
In this section, we shall briefly introduce the FRP model
of DESs and formulate the boundedness problem over this
model. However, the reader is requested to go through [7] or
[8] for detailed discussions on this model.
is the set
Let be a finite collection of event symbols.
For
of all possible event strings formed with elements from
then
example if
where denotes the null string. For any two strings
and
denotes the concatenation of the strings
and
Also,
Note that
(followed by)
The prefixes of
in [7] the null string was denoted as
will be
and
Let
be the set of
the string
where each process
processes formed over events from
is expressed formally as a 3-tuple
1) The set
is the prefix-closed set of traces
(event sequences) that can take place in
2)
is the alphabet function, so that
is the set of events
can execute or block after it has
generated the event sequence
3)
is the termination function where
represents successful termination of after
generation of the event sequence in
is a subprocess of
(denote it as
if
and
and
and
are equal,
if
, and
is the set of
processes which never terminate successfully.
describes the behavior of
after
The post process
has been generated in it:
a string
Definition 2.2
the
events and
Definition 2.3
of events and
the
: Given a process
and collection
is defined as follows:
is not in
Definition 2.4
composition
: Given
is defined as
and
the sequential
if
if
1517
Among the above operators, some are unary and some are
binary in nature. Each binary (unary) operator is actually a
function over two (one) processes. More complex functions
in multiple process arguments can be composed as process
expressions by repeated application of these operators. The
following definitions will be quite useful in representing such
a set of functions.
First, we define
a set of functional symbols (also called a
signature [5]) corresponding to the different process operators,
as follows.
Definition 2.6 (Signature): The signature set
if
otherwise.
Intuitively,
behaves initially as
Once
termiThe sequential composinates successfully, it behaves as
tion operator is a particularly important operator as it allows
recursive and re-entrant description.
between two proThe parallel composition operator
and
generates a new process (denoted by
cesses
which depicts the concurrent operation of two processes.
: The projection of a string
Definition 2.5
on a process (denoted by
) is defined as follows:
and
undefined if
if
if
Then, for processes
is defined as:
If
and
then
iff
and
Finally
and
if
otherwise.
The process
after generating an event sequence
can execute an event from
iff both
and
execute the event synchronously. Otherwise, such
an event will be blocked because of lack of participation from
belongs to any of
one of the components. Otherwise, if
then, it can take
if it takes place in
since
cannot block
place in
such an event.
Note that
is an -argument function,
is the corresponding symbol, while
is a constant
process.
Definition 2.8 (Projection Symbols): For any
The
function simply yields its th argument. Note
and
with
that for any three positive integers
1518
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 43, NO. 11, NOVEMBER 1998
the function
belongs to both
and
However, in the first case it is an
-ary
function and in the second case it is an -ary function.
over a signature
After having a family of functions
for some
one can define recursively the infinite class
and the corresponding set of
of functional symbols
from
to as follows.
functions
): It is defined recursively as:
Definition 2.9 (
and
implies
where
has an arity
Definition 2.10
and
),
3)
5)
6)
7)
So the process
can be expressed as
if
is
For any
first
is expressed in notationally simplified form,
symbol is then removed, and finally
its outermost
is written as
The same rule is applied
and
Thus
for
is simplified as
For
if is
or
first
and
are expressed in notationally
simplified form and the outermost symbols are removed.
or
respectively.
Then is written as
and
are
After simplification, since both
associative in nature [7], whenever the removal of
the braces does not cause any confusion regarding the
scope of any of the functions, they will be removed.
will be expressed as
For example
As an example,
will be written as
A distinction between syntax and semantics is needed becan represent
cause different functional symbols in
the same function. For example, the syntactically distinct
and
(for some
)
expressions
for
and
and
4)
and
if
and
and
and
such that for every
there
with
and vice-versa.
exists
The above definition implies that
is syntactically
and
Also clearly
equivalent to both
is an equivalence relation. Because of the notational
implies that
simplification we can conclude that
A similar statement
none of the has a structure
also.
holds for
Definition 2.12 (Semantic Equivalence): Given
and are semantically equivalent,
if
and
Thus
and
are syntactically distinct but semantically equivalent.
Again it is easy to see that is an equivalence relation
Moreover is finer than i.e.,
over
implies
but not vice-versa.
Definition 2.13 (Length of Expressions and Subexpressions):
the length of
and set of subexFor
are defined recursively as follows.
pressions of
and
for
1)
or
2)
and
for
or
.
and
3)
for
or
The aim of defining all these operators is to provide a finite
The following
and recursive description of processes from
definitions and results show how it is achieved.
Definition 2.14 (Mutually Recursive Processes): A finite
is called a family of mutually recursive
set
if
processes (MRP) with respect to
we have
for some
Theorem 2.1:
is a family of MRP w.r.t.
iff
is the unique solution of
the recursive equation with consistent initial condition
(1)
where
that is
of
1519
is guarded,
(2)
and
is a suitable
with each
choice of initial alphabet and termination, such that
and
Proof: For the proof, see [7].
Definition 2.15 [Finitely Recursive Processes (FRP)]: A
is said to be a finitely recursive process
process
if it can be represented
(FRP) with respect to
where
is
as
is said to be a
of the form (2) and
realization of the FRP
Definition 2.16 (Algebraic Process Space): The collection
is denoted as
of all possible FRPs w.r.t
The algebraic process space
is defined as
in
is of the form of (2), i.e.,
describing
Here
(Note:
if
and
if
if
if
Finally for
with realization
the set of postprocess expressions
is computed recursively
as follows.
1)
2)
Fact 2.1: It is easy to check that
is closed
under post-process operation, that is, if
with realization
for some
then for any
for some
Example 2.2 (Post-Process Expressions): Refer to the three
in Example 2.1.
for these
realizations of the process
three cases are as follows.
for realization (1).
1)
for re-
2)
alization (2).
1520
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 43, NO. 11, NOVEMBER 1998
of
is finite. In this definition boundedness is viewed as a
property of the particular realization of the FRP. Referring to
Examples 2.1 and 2.2, we then find that based on the second
definition, is bounded in its first realization but unbounded
in its second and third realizations.
it
In case of the second realization of the process
should be noted that the different syntactic expressions
are
This fact leads us
all semantically equivalent to
to formulate a definition of boundedness which is closer (in
spirit) to the first (ideal) definition of boundedness. Given
and
let
some FRP with realization
be some syntactic transformation over elements of
such that for any
Then, the simple post-process expression computation
procedure mentioned earlier can be modified by applying
this transformation along with syntactic substitution and one
can hope that many redundancies, like those present in second
in Example 2.2, will be taken care of by
realization of
the set of post-process
this transformation. Formally,
expressions, is computed as follows.
1)
2)
where
is computed by the recursive procedure
mentioned earlier. Finally, boundedness is defined as
below.
Definition 2.18 (Boundedness): An FRP
with realization
is said to be bounded under
the post-process expression computation procedure involving
(having the property that
some syntactic transformation
for any
and
if
is finite.
Note that, in this definition, boundedness becomes a property of the realization of the process with respect to a syntactic
substitution used in the computation of post-process expressions.
The following definitions and results will be useful in
defining suitable syntactic transformations to take care of
functional redundancies among function symbols.
First we present the concept of minimum present and
minimum absent alphabet in function symbols. Because of
and the
operators, given some
the
it is possible to have a lower
function
bound on the minimum set of events that is present (or
absent) everywhere in the alphabet along the traces of
irrespective of which particular
is chosen as the argument
Formally the definitions are as follows.
of
Definition 2.19 (Minimum Alphabets):
Minimum Present Alphabet:
Then
But
then
Whereas
where
In the above example, if
Now
1521
Then
If
but
for
any
and
then
s and
and
s.
let
s.
as
1522
of
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 43, NO. 11, NOVEMBER 1998
is a distinct member
): Given
where
Since we are dealing
for any process, if the set of traces
with processes from
contains only the null string then the process must be
for some
as
if
if
Thus, in case of a
transformation
function expression involving
removes redundancy in individual argument expressions.2
In case of change operators the following transformations
take place.
1) Multiple applications of the same types of change ops and
s) are converted into a sinerators
gle application of the same operator. For example,
will be converted into
2) Redundant event symbols in the parameters of
and
are removed with the help
of minimum present and absent alphabets. Thus
is converted into
If
is empty, the corresponding
operator symbol
and
is removed. Similarly, in case of
if
Also
will
be
converted
into
where
and
4) The
over the
if
if
if
To illustrate the transformation
we give an example
here.
be the collection of the
Example 3.2: Let
following equations:
1523
TABLE I
SYNTACTIC TRANSFORMATION
^ := (B 0
Note that in Column 3, B
~ := (
0 C ) [ C^
\ C \ F (g)) [ (C 0 F (g)) C
00
^ := (B 0
and C
=
\ C \ F (g)).
0 F (CF (f )):
follows:
are subsets of
Finally
1)
2)
Note that
together include arbitrary process expressions with two characteristic features: the first being the
fixed order among the change operator symbols, and the
in distributed form
second is the appearance of
over
The following lemma formally expresses the properties of
Lemma 3.1: For any
and
we have
1)
2)
3)
4)
5)
6)
7)
8)
Proof: The claims can be proved together with the help
of structural induction on the construction of
The proof is mechanical but tedious as it involves exhaustive
enumeration and analysis of all possible structures that
might possess. The formal proof is given in [1].
CF 1
~ := (B 0
In column 5, B
0 C ) [ (B \ C \ F (g))
0
and
[within
1524
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 43, NO. 11, NOVEMBER 1998
expression, even if
turns out to be a singleton.
cannot be a variant of any component
Thus
expression even if its component set is a singleton containing
Referring to Example 3.2, we find that
three components, which are
has
and
is empty, then
and
For any functional symbol from
in order
to achieve the above kind of reductions, we need to identify
its minimal (in terms of minimum present alphabet) trace generating subexpression(s), (called component expression(s))
genwhich, when applied (as functions) on
erate the same trace as the original functional symbol. For
example for the expression
the component expressions will be
and
Definition 3.3 (Component and Variant Expressions): For
and
the set of component
is the set of expressions
expressions of namely
such that
is obtained as an interleaving of the
and the expressions are minimal in the
traces of
sense of minimal present alphabet. Formally
if
if
if
and
Remark 3.1 (Tree Representation): Any
can be expressed in the form of a tree
such
the leaf nodes
that the root node contains the expression
contain the variants of different component expressions of
, and the intermediate nodes contain the subexpressions of
which are not variants of any component expression of
Formally the tree is constructed recursively as follows:
2) If any node contains some
1) The root node contains
or
expression of the form
or
such that
then the node is not expanded further as the expression itself is
a variant of some component expression; and 3) if it contains
then
some expression such that
children are created, one each containing each
denotes the set of nodes in
and
denotes the set of leaf nodes. For any node
denotes the expression contained in it.
Example 3.3: Consider the processes in Example 3.2.
Clearly,
Let
is in
If we construct
it will have
a root node containing and two immediate children nodes,
and
containing
respectively. Each of these nodes will have an identical set
and
of two children (leaf) nodes containing
respectively. The tree generated is shown in Fig. 1.
In order to remove redundant arguments of
we now
identify a set of conditions under which a functional symbol
can merge into another symbol
to form a single symbol,
such that
for
say,
but
We say
merges into
and express it as
where
is the binary merging relation. We also
require that after merging, the resultant expression, namely
has the same structural properties as mentioned
should be an
in Lemma 3.1. In other words,
To keep these conditions of
element of
merging general and amenable to easy computation, we use
in the form of
and
only local information from
This information is local as it can be obtained without
computing any post process expression.
): Given
Definition 3.4 (Merging:
if any of the following is satisfied.
and
1)
2)
(say)] and
or
or
1525
= ( 1 jj
4)
3)
and
In the above cases the result of merging
into expression
namely
is (to be defined below)
for the argument
where
and
is rewritten as follows:
From definition of
1526
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 43, NO. 11, NOVEMBER 1998
to
when
where
and
For
merged to form
the formal proof the reader is referred to [1].
Note that in case 3) of Definition 3.4 if
is a proper subset of
(which also equals
) then
is defined, but
is not defined. In case
and
these two sets are equal, both
will be defined. However, they will be syntactically distinct even though
Example 3.4: Consider
Clearly
It is then obvious
as defined in the second
that
have identical trace,
condition, and
termination and initial alphabet. Also, for any string
other than the empty string
Hence
From the final form of
it can be easily seen that
belongs to
as it satisfies the first six
conditions of Lemma 3.1. From the second condition, it is
easy to see that
(if
) or
(otherwise). From the above it is also obvious that
is
reflexive and transitive.
We now proceed to prove the lemma for case 3) of Definition 3.4. Because of the special structure obtained after
given any
the
application of
can be considered as the set of trace
elements of
This is because
is
generating components of
formed via interleaving of traces of processes corresponding to these individual component expressions. The
symbol and the different
and
symbols
only add to the blocking capability of different components
For example, consider the symbol
joined by
which clearly belongs
For any three sets X; Y; Z we have X 0 Y = (X [ Z ) 0 (Y [
Z 0 X )): Here we take X = F (h) [ C1i ; Y = B2i and Z = C1j :
4 Note: For any three sets X; Y ; Y ; (X 0 Y ) [ (X 0 Y ) = X 0\2
1 2
1
2
i=1 Yi :
We take X = F (h) [ C11 [ C12 and Yi = (B2i [ (C1j 0 (F (h) [ C1i ))):
5 Note: Y = (B t [ B ai [ (C j 0 ( (h) [ C i ))) and B t \ (B ai [
i
2
2
1
2
2
1
F
j
(C1 0 (F (h) [ C1i ))) = for both i = 1; 2: Then Y1 \ Y2 =
j
(B2t [ (\i2=1 (B2ai [ (C1 0 (F (h) [ C1i ))))):
6 Note: If Y X; Y = Y [ Y ; Y \ Y = then (X 0 Y ) [ Z =
1
2 1
2
(X 0 Y1 [ (Y2 0 Z )) [ (Z \ X \ Y1 ) [ (Z 0 X ): Here X = F (h) [
C11 [ C12 ; Y1 = B2t ; Y2 = B2a ; Z = C21 [ C22 :
3 Note:
expression
when
and
The expression is in
for the in Example 3.2. It can be merged
into
the
function
either
or
However,
further merging among subexpressions in both cases would
result in the unique expression
As demonstrated in Example 3.4, after one obtains
further merging
an expression
may be possible among the components of
To exhaust all merging possibilities we define a
as follows. For any
recursive merging procedure
defines a
and
set such that for
such that
and
Definition 3.5 (Recursive Merging Procedure ):
then
If
If
then
given below.
Step M.1: Let
elements among
and
1)
2) Partition
such that
are in the same
and
partition iff both
Let there be partitions and the th partition
be
3)
is a singleton and
4)
1527
and
as follows:
such that
or
such that
but
by RL-
such that
Step M.3:
compute
(say).8
and
Step M.5: If
with
Finally if
least one
go to
Otherwise
is
of
and at
nonempty, then
where
RL-1: Computation of
Step M.1 (RL-1):
if
if
by RLExample 3.5: Once again consider the expression presented
in Example 3.4. Let
when
and
The
with respect to the
expression is in
presented in 3.2. Here we show the steps that
follows to
prevent further merging in the resultant expression.
RL-0 (Recursion Level 0): We have the computation of
.
RL-2: Computation of
9 By Definition 3.4, M (g ; g ) is in C
(
n (Gjj ; 5~ )) and M apm 1 2
F1
plies Mm (1; 1) recursively. In spite of that, we had to apply CF 1 (1)
in this case to take care of the possibility that when M (jjm
i=1 vi ) turns
out to be a singleton, h[[+C ]][0B ][+C ] will not have the structure as
described in Lemma 3.1. For example, without the application of the final
set containing
by RL-
RL-3: Computation of
Step M.1 (RL-3):
1528
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 43, NO. 11, NOVEMBER 1998
and
where
and
if
are not syntactically equiva-
is present in
as the variant
and in
it is present as the variant
In the case of
and
it is just the opposite. At this stage it should be
in
can
noted that each variant of the component
if
where
is defined as follows:
if
if
and
is a singleton.
To find
note that
for
but we have the equation shown at the bottom of the page.
according to the algorithm
Next we compute
1529
After partitioning
and doing both-ways merging among
we get
elements of
then
such that
and
Also
At the end of the while loop,
Among the expressions of
the first element (i.e.,
and
in any order and then applying
of
and
those which merge in to
are
So
together imply
d)
3) For
Proof Outline: Intuitively, the results of the lemma are easy
to see from the following facts.
1) From Lemma 3.2 we see that the merging operation preserves the components, the semantics for the argument
, and results in an expression in
2) After applying
over the arguments of outermost
s in any subexpression, no further merging among
s is possible.
the arguments of the outermost
3) During computation of Closure, one extracts all the
different variants of the components that are present in
So after applying
an expression and puts them in
over elements of
it is obvious that each variant of
identical component gets replaced by a unique variant of
the same component in the resultant expressions, while
preserving the semantics.
ensures that this
4) Finally, the recursive procedure
is applied in every subexpression level
Closure and
of a component/variant. Formal proof of the lemma is
given in [1].
Definition 3.7 (Transformation
): Let
such that
Lemma 3.4: Given
satisfies the following: 1)
2)
3)
Proof: The proof is straightforward from Lemmas 3.3
and 3.1.
For two reasons, the expressions obtained after applying
cannot be considered to be in some normal form. First,
the resultant expressions are not semantically equivalent in
general; they are semantically equivalent only for the argument
This is because we use local information in the
and
in the transformation. Moreover,
form of
the transformation does not ensure that any two function
expressions that are semantically equivalent in general or for
will be converted into a unique expression.
the argument
Thus, some redundant function expressions in general, may
remain even after applying
Consider the infinite sequence of functions
, presented
in the counterexample 3.1. Given some processes
such that
it can be easily seen
Thus, given some FRP realization
that
though transformation
removes a substantial
1530
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 43, NO. 11, NOVEMBER 1998
amount of redundancy,
remains an infinite
set in general.
with
However, for any arbitrary FRP
we will show that while the set of postrealization
is infinite, (see Case 2 of Examples 2.1
process expressions
computed by the procedure for comand 2.2), the set
putation of post-process expressions involving transformation
is finite.
): For
Definition 3.8 (Computation of
with realization
the computation proceis the same as presented in previous section. The
dure of
is
set of postprocess expressions
now computed recursively as follows:
If
then
The finiteness of
follows from two important
observations.
Observation 1: During the computation of the post
operators do
process expressions, we find that 1)
not accumulate giving rise to new components; 2) because of
redundant application of change operators
application of
are avoided; 3) new components are generated only when
are
post-processes of any subexpression of the form
computed.
As a result, components of any expression from
are from a finite set
defined below.
in
Definition 3.9: Give any process
with realization
we define a set of components
as follows:
where the
IV. CONCLUSION
In the present work we have discussed the boundedness
and the change operators.
property of FRPs built over
alone does not guarantee
We have shown that removal of
finiteness of the underlying function space. However, using
suitable syntactic transformations along with the computation
procedure for the post-process expressions, it has been possible
and the change
to show that the FRPs built over
operators are bounded. It can also be shown that if
is removed along with
the underlying set of functions
becomes finite [1].
1531
REFERENCES
[1] S. Bose, Finitely recursive process models for discrete event systemsAnalysis and extension, Ph.D. dissertation, Dept. Electrical
Engineering, I.I.T. Kharagpur, 1995.
[2] R. A. Cieslak, Undecidability results and real time semantics for finitely
recursive processes, Ph.D. dissertation, Dept. E. E. and C. S., Univ.
California, Berkeley, 1988.
[3] R. A. Cieslak and P. P. Varaiya, Undecidability results for deterministic
communicating sequential processes, IEEE Trans. Automat. Contr., vol.
35, pp. 10321039, Sept. 1990.
, Simulating finitely recursive processes in LISP, Discrete Event
[4]
Dynamic Syst.: Theory and Appl., vol. 1, pp. 373392, 1992.
[5] M. Hennessy, Algebraic Theory of Processes. Cambridge, MA: M.I.T
Press, 1988.
[6] C. A. R. Hoare, Communicating Sequential Processes. New Delhi,
India: Prentice-Hall, 1985.
[7] K. Inan and P. P. Varaiya, Finitely recursive process models for discrete
event systems, IEEE Trans. Automat. Contr., vol. 33, pp. 626639, July
1988.
[8]
, Algebras of discrete event models, in Proc. IEEE, Special
Issue on Discrete Event Dynamic Systems, vol. 77, no. 1, pp. 2438,
Jan. 1989.
[9] R. Milner, Communication and Concurrency. London, U.K.: Prentice
Hall Int., 1989.
[10] J. S. Ostroff and W. M. Wonham, A framework for real-time discrete
event control, IEEE Trans. Automat. Contr., vol. 35, pp. 386397, Apr.
1990.