Dual views on throughput properties of continuous

mono-T-semiflow Petri nets: firing rate and initial marking

variations
Manuel Navarro-Gutiérreza,1 , Antonio Ramı́rez-Treviñoa , Manuel Silvab
a
CINVESTAV–IPN, Unidad Guadalajara,
Av. del Bosque 1145, Zapopan , 45019, Jalisco, México.
email: {mnavarro, art}@gdl.cinvestav.mx
b
Instituto de Investigación en Ingenierı́a de Aragón (I3A), Universidad de Zaragoza,
Marı́a de Luna 1, 50018 Zaragoza, Spain
email: silva@unizar.es

Abstract
This work studies behavioral properties of the equilibrium throughput in Timed Continuous
Petri nets. It presents novel results that are grouped in three parts. The first one focuses
on the relationships among properties as monotonicity, continuity, and deadlock-free mono-
tonicity of the equilibrium throughput when either the firing rate or the initial marking
varies, for the class of Mono-T-Semiflow nets. The second one relates previous properties
with some structural objects named configurations; it is shown that for both, firing rate
and initial marking variations, the occurrence of undesired behaviors (non-monotonicities,
discontinuities) is due to configurations whose associated set of places does not contain the
support of a P-semiflow. Finally, based on the structural characterization, sets of firing rates
and initial markings for avoiding undesired behaviors are derived.

1. Introduction
Performance evaluation is a major topic in the area of Discrete Event Systems (DESs). It
provides, for instance, the amount of finished jobs per unit time in a manufacturing system
[1, 2], the interaction between populations or species in chemical reaction networks [3], the
patient flow in an hospital department [4], among others. Performance dependency on the
events occurrence and on the initial state of the system suggests that these two aspects could
be conveniently adjusted to improve this performance.
Among different formalisms to represent DESs, Petri Nets (PNs) are a widely accepted
paradigm by the scientific community for modeling, analyzing and control of DESs because of
their powerful mathematical background and their nice graphical representation [5]. Unfor-
tunately, they suffer of the state explosion problem as any other expressive DES formalism,

Preprint submitted to Some Journal September 19, 2018

particularly when they are heavily populated (marked in PN terms). This is an obstacle that
has been tackled with a classical relaxation: the fluidification of the system; this leads to
the concept of Continuous PNs (CPNs) [6, 7]. They allow to make computational problems
decidable or much tractable in practice. Furthermore, when time is also considered, Timed
CPNs (TCPNs) make possible that diverse performance measures (throughput, response
time, average marking) can be approached.
Different timing interpretations have been introduced from diverse perspectives (finite
server semantics, FSS; infinite server semantics, ISS; product semantics, PS; see [7]). This
work focuses in ISS since it has been shown that they best approximate interesting classes
of DESs; particularly, the class of Mono-T-Semiflow-Reducible MTSR nets [8]. These nets
allow to model cooperative and competitive relationships that arise in manufacturing sys-
tems, biochemical networks, or health care systems [1, 3, 4]. Even more, the equilibrium
throughput of a TCPN system for this class of nets can be interpreted as a performance
measure (finished jobs per unit time, interaction between populations) that depends on the
firing rates and on the initial marking of the system.
Properties like monotonicity or continuity of the equilibrium throughput in function
of the initial marking and the firing rates [8, 9], and deadlock-freeness as function of the
firing rates [10], have been earlier discussed. Although they suggest that monotonicity,
continuity and deadlock-freeness are related, there is no work analyzing their relationships.
Monotonicity means that the increasing of either the initial marking or the firing rates leads
to a non-slower system (a desired behavior), i.e. more resources and raw material in the
system, or faster machines, should not reduce the finished jobs per unit time. Deadlock-
freeness means that no blocking situation can occur in the system; a property that can be
reinterpreted as persistence in chemical reaction networks [3]. Continuity indicates that no
abrupt changes of the equilibrium throughput are possible in the system when firing rates
or the initial marking vary (again a desired system property). Moreover, [11] suggests that
discontinuities of the equilibrium throughput in TCPN systems are closely related to poor
approximations of the mean marking or throughput of the corresponding stochastic Petri
net at that point.
This work takes advantage of the equilibrium throughput dependency on the firing rates
and on the initial marking to establish relationships among the three properties mentioned
before. It points out the similarities between the effects of varying the initial marking and
the firing rates in the equilibrium throughput, an expected behavior because of the duality
between places and transitions. Moreover, it makes an extensive use of structural theory
to characterize these behavioral properties using structural objects of the net called config-
urations; this directly results in a computational improvement for the system analysis and
allows to better understand why undesired behaviors (discontinuities and non monotonic-
ities) appear in the system. It leverages on this characterization to derive conditions that
guarantee the desired properties. This work complements the results reported in [9] and
[12].
The remaining of this paper is structured as follows. Section 2 presents first some basic
concepts of Petri nets, continuous and timed systems; it also highlights some homothetic
properties of firing rate and initial marking variations. Section 3 describes the class of
2
nets under study. Relationships between monotonicity, continuity, and deadlock-free mono-
tonicity with respect to firing rates and initial marking variations are explained in Section
4. Using structural objects of the net, these properties are characterized under a common
framework in Section 5. Section 6 presents how the firing rates and the initial marking
can be restricted to ensure the desired properties. Some concluding remarks and research
directions are briefly discussed in Section 7.

2. Basic Concepts and notations

Notation: N, Q and R stands for natural, rational and real numbers, respectively. Given a
set of numbers A, A≥0 (resp. A>0 ) denotes the set of nonnegative (resp. positive) numbers
of A. Given a matrix M of size |A| × |B| with A and B sets of indices, the submatrix
M [A0 , B 0 ], with A0 ⊂ A and B 0 ⊂ B, denotes the restriction of M to rows indexed by A0
and columns indexed by B 0 . The support of a vector z is the set of indices corresponding
to its non null values and is denoted by kzk.

2.1. Petri net structure and system

Definition 2.1. A Petri net (PN) structure is a bipartite graph defined as a four tuple
N = hP, T, P re, P osti, where:

• P and T are finite non-empty disjoint sets of nodes named places and transitions,
respectively;

• P re : P × T → N ∪ {0} is the pre-incidence function specifying the weighted arcs

directed from places to transitions; and

• P ost : P × T → N ∪ {0} is the post-incidence function that specifies the weighted arcs
directed from transitions to places.

A subnet of N is a Petri net structure N 0 = hP 0 , T 0 , P re0 , P ost0 i where P 0 ⊆ P , T 0 ⊆

T are subsets of places and transitions of N , respectively; and P re0 = P re[P 0 , T 0 ] and
P ost0 = P ost[P 0 , T 0 ] are the pre- and post-incidence functions of N restricted to P 0 and
T 0 . The preset and postset of a node v ∈ P ∪ T are denoted as • v (set of input nodes)
and v • (set of output nodes), respectively. These definitions can be naturally extended:
let V ⊂ P ∪ T be a set of nodes, • V (respectively, V • ) denotes the union of the preset
(respectively, of the postset) of every node v ∈ V . A siphon is a set of places Θ such that
•
Θ ⊆ Θ• .

Definition 2.2. Particular structures of the nodes of a net.

1. A transition t such that |t• | > 1 (resp. |• t| > 1) is called a fork (resp. a join).
2. A place p such that |p• | > 1 (resp. |• p| > 1) is called a choice (resp. an attribution).
3. Two transitions ti and tj are in conflict relation if • ti ∩ • tj 6= ∅. This is a reflexive
and symmetric relation, but not transitive.

3
 4. The transitive closure of the conflict relation is an equivalence relation. The coupled
conflict set (or equivalence class) of transition t is denoted by CCS(t) and the quotient
set is SCCS.
5. Two transitions ti , tj are in proportional equal conflict (PEQ) relation if P re[P, ti ] =
kP re[P, tj ] for k > 0.

These structural concepts allow to define the following net subclasses.

Definition 2.3. Let N be a Petri net

• N is Join-Free (JF) if ∀t ∈ T , |• t| ≤ 1, that is, if there is no join transition in the
net.

• N is Choice-Free (CF) if ∀p ∈ P , |p• | ≤ 1, that is, if there is no choice place in the

net.

• N is Fork-Attribution (FA) if it is JF and CF.

• N is Continuous or Proportional Equal Conflict (PEQ) if all conflicts are proportion-

ally equal.

The token-flow matrix (incidence matrix if self-loop free) of a net N is defined as C =

P ost − P re. A column vector y 6= 0 is called a P-flow of N if y T C = 0; if the non-null
entries of such vector are positive, then it is called a P-semiflow. Similarly, a column vector
x 6= 0 is termed a T-flow of N if Cx = 0; if the non-null entries of that vector are positive,
then it is termed a T-semiflow. If there is a P-semiflow (T-semiflow) such that y > 0
(x > 0), the net is said to be conservative (consistent).

Definition 2.4. A net N is Mono-T-Semiflow (MTS) if it is conservative and consistent

with a unique minimal T-semiflow.

2.2. Continuous Petri net system

|P |
A marking is a mapping m : P → R≥0 that assigns to each place of N a non negative
real value.

Definition 2.5. A continuous Petri net (CPN) system is a net N together with an initial
marking m0 and it is denoted as hN, m0 i.

One characteristic of continuous Petri net systems is their evolution rule: allowing the
firing of transitions in positive real amounts, the reachable markings of the continuous system
are non-negative real values.
A transition t ∈ T is enabled at a marking m if ∀p ∈ • t, m[p] > 0. The enabling degree
of a transition t is defined as
 
m[p]
enab[t] = min . (1)
∀p∈• t P re[p, t]
4
An enabled transition can be fired in any real amount between the interval 0 ≤ δ ≤ enab[t].
Its firing leads to a new marking m0 = m + δC[P, t]. A marking m is reachable if there
exists a (finite or infinite) firing sequence σ, such that m can be reached from m0 . The set
of reachable markings, denoted as RS(N, m0 ), is composed of every reachable marking and
its elements satisfy the following fundamental equation

m = m0 + Cσ, (2)
|T |
where σ ∈ R≥0 is the firing count vector. Moreover, if there is no empty siphon at m0 and
the net is consistent (required conditions in any practical system), then the set of reachable
markings is equal to the set of solutions of Equation 2 [7].

Definition 2.6. It is said that a continuous PN system hN, m0 i is:

1. Bounded (B) if for every reachable marking m ∈ RS(N, m0 ) there is a constant b,
such that for all pi ∈ P , m[pi ] ≤ b;
2. Live (L) if for every transition t and for every marking m ∈ RS(N, m0 ) there exists
m0 ∈ RS(N, m) such that t is enabled at m0 .

The above are behavioral concepts depending on both, the net and the initial marking.
Structural properties are those in which m0 is abstracted, hence they depend only on N ,
the structure.

Definition 2.7. It is said that a PN N is:

1. Structurally Bounded (SB) if for every initial marking m0 , the CPN system hN, m0 i
is bounded;
2. Structurally Live (SL) if there exists an initial marking m0 such that, the CPN system
hN, m0 i is live.

2.3. Timed continuous Petri net systems

If the firing of transitions are timed, the marking of the continuous system evolves au-
tonomously through a trajectory within the set of reachable markings. In this sense, Eq.
2 depends on time, that is, m(τ ) = m0 + Cσ(τ ), and its time derivative gives the state
equation ṁ(τ ) = C σ̇(τ ). The derivative of the firing count vector is known as the flow
or throughput vector of the timed model f (τ ) = σ̇(τ ). Because of the good approximation
they provide [8], in this work the firing flow is defined using the infinite servers semantics
and, for each transition tj ∈ T , it is described by the equation
 
m[p]
f [tj ] = λ[tj ] min , (3)
∀p∈• tj P re[p, tj ]
|T |
where λ : T → R>0 is a mapping known as the firing rate vector, that assigns to each
transition of N a positive real value.

5
Definition 2.8. A timed continuous Petri net (TCPN) system is a CPN system together
with a firing rate λ and it is denoted as hN, λ, m0 i.

Due to the min operator in Eq. 3, the marking evolution of a TCPN system can be
represented by an affine positive piecewise linear system. The following four concepts are
provided to illustrate this fact.
1. A configuration C of a net N is a set of (p, t) arcs, one per each transition of the net,
such that p ∈ • t. The T-coverture of a configuration C is TC = {p | ∀t ∈ T, (p, t) ∈ C},
i.e., the places preceding the arcs of C. A configuration C has associated a configuration
|T |×|P |
matrix ΠC ∈ Q≥0 where

1
 if (pi , tj ) ∈ C
ΠC [tj , pi ] = P re[pi , tj ] (4)
 0 otherwise.

A configuration C is active at a marking m if ΠC m = enab(m).

2. A region RC is a (sub)state space in which a unique configuration C is active. Regions
constitute a partition —except on the borders— of the full polytope of reachable
markings.
3. The induced net of a configuration C is defined as the subnet NC = (TC , T, P re0 , P ost0 ).
4. An operation mode or regime ΣC of a TCPN system is the linear system (ṁ =
CΛΠC m, where Λ = diag(λ) is a diagonal matrix) that describes the marking evo-
lution while m evolves within region RC . At configuration C, CΛΠC is the dynamic
matrix.

Q The• set of all configurations of a net is represented as SC(N ). Its size is upper bounded by
∀t∈T | t|. It is possible that different configurations have associated the same T-coverture
when in a coupled conflict set exist different ways of cover its transitions with arcs that
have the same preceding places. Therefore, the set of configurations can be partitioned
(equivalence relation) according to the associated T-covertures.
Configurations and T-covertures are structural objects of the net N . They are related to
the regions of the CPN system hN, m0 i, and to the operation modes of the TCPN system
hN, λ, m0 i. Both structural objects allow to perform an analysis of the system using the
structure of the net, and are key for the results presented in upcoming sections.
The dynamic behavior of a TCPN system can be described by the following equations:
ṁ = CΛΠC m, m ∈ RC , C ∈ SC(N )
(5)
m(0) = m0 ,

which is reducible to an equivalent affine positive piecewise linear system [7].

An equilibrium marking me ∈ RC is a state of the TCPN system fulfilling ṁ =
CΛΠC me = 0; its associated equilibrium throughput is denoted by f e = ΛΠC me . Note
that the flow f e is a linear combination of the T-semiflows of the net, no matter which
configuration me activates.
6
 2 2 2
2 2

2 2 2
2 2

(a) A continuous Petri net (b) Configuration C4 high- (c) Configuration C5 high-
system hN, m0 i. lighted and its induced sub- lighted and its induced sub-
net. net.

Figure 1: (a) A continuous Petri net system; (b) and (c) configurations C4 and C5 and their corresponding
induced subnets.

Example 2.9. The net in Fig. 1(a) has the following |• t1 | · |• t2 | · |• t3 | = 1 · 3 · 2 = 6

configurations (with their corresponding T-covertures):
C1 = {(p1 , t1 ), (p1 , t2 ), (p3 , t3 )} with TC1 = {p1 , p3 },
C2 = {(p1 , t1 ), (p1 , t2 ), (p4 , t3 )} with TC2 = {p1 , p4 },
C3 = {(p1 , t1 ), (p2 , t2 ), (p3 , t3 )} with TC3 = {p1 , p2 , p3 },
C4 = {(p1 , t1 ), (p2 , t2 ), (p4 , t3 )} with TC4 = {p1 , p2 , p4 },
C5 = {(p1 , t1 ), (p3 , t2 ), (p3 , t3 )} with TC5 = {p1 , p3 },
C6 = {(p1 , t1 ), (p3 , t2 ), (p4 , t3 )} with TC6 = {p1 , p3 , p4 }.
As |• t1 | = 1, (p1 , t1 ) is an essential (i.e., always needed) cover. The induced net of configu-
ration C4 is depicted in Fig. 1(b); configurations C1 and C5 have the same T-coverture and
the same induced net which is shown in Fig. 1(c), thus in this net |ST (N )| < |SC(N )|.
 T
If the net is provided with the initial marking m0 = 2 0 2 0 , the P-semiflows
 T  T
1 1 0 0 and 0 0 1 1 lead to the following two token conservation laws:
m[p1 ] + m[p2 ] = m[p3 ] + m[p4 ] = 2.
Therefore, the reachable space of hN, m0 i, can be fully understood if projected to the m1 −m3
plane. It is formed by the corresponding six regions shown in Fig. 2(a).
 T
If the CPN system is provided with the firing rate vector λ = 6 3 1 , then the
marking evolution of hN, λ, m0 i is as depicted in Figure 2(b). It asymptotically converges to
the equilibrium marking me = [0 2 0 2]T , in which the equilibrium throughput is f e = [0 0 0]T
(i.e., it is a deadlock).

2.4. Homothecies and dynamic evolution in timed continuous systems

Since the dynamic behavior of a TCPN system can be described by Equation 5 —the
marking evolution of hN, λ, m0 i is the solution to that initial value problem— the following
well known homothetic properties are dynamically satisfied [7]:

• If m0 is multiplied by a constant k > 0, then the marking trajectory will be k times

bigger, but the speed will be the same; the throughput will be k times bigger.
7
 1 2 3

1 1
1

0.9 0.9

0.8 2 0.8

0.7 0.7

0.6 0.6

0.5 0.5
3
0.4
0.4

0.3
0.3

0.2
0.2

0.1
0.1

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0
0 0.5 1 6 8 10 12

(a) (b)

Figure 2: (a) The reachable space of hN, m0 i can be partitioned in regions using the structural objects of
the net (configurations). (b) The marking evolution of hN, λ, m0 i evolves under Σ4 , Σ6 and Σ5 to finally
reach a deadlock.

• If λ is multiplied by a constant k > 0, then the marking trajectory will be the same,
but k times faster; the throughput will be k times bigger, and k times faster.

Particularly, if an equilibrium throughput of a TCPN system is considered, then the

homothetic increment of the initial marking or of the firing rate vectors produce the same
increment on f e .

Property 2.10. Let f e be an equilibrium throughput of hN, λ, m0 i, then kf e is an equilib-

rium throughput of the system hN, kλ, m0 i, as well as of the system hN, λ, km0 i.

If the variation of the firing rates or of the initial marking is not homothetic, that is, if only
some components of λ or m0 are increased, then the equilibrium throughput may change
in a paradoxical/undesired way, particularly, presenting discontinuous and non-monotonic
value changes [13]. This work provides structural conditions to explain the relation between
λ and m0 with these phenomena, focusing on the study of a class of nets.

3. Mono-T-Semiflow reducible nets and subclasses

This section first recall a reduction rule for transitions in proportional equal conflict.
Furthermore, the class of Mono-T-Semiflow Reducible (MTSR) nets, the relevant classes
that are encompassed in it, and its modeling power are briefly discussed, together with some
performance properties that can be interpreted on it.

3.1. Proportional equal conflict reduction

When TCPN systems under infinite server semantics are considered, the proportional
equal conflicts can be merged keeping fixed the firing rates of the fused transitions. This
results in a reduced system in which the marking evolution is identical to the original one.
8
 Figure 3: The reduction of the P EQ = {t1 , t2 } to a single transition t12 [13].

Property 3.1 ([13]). Let hN, λ, m0 i be a TCPN system and ti , tj ∈ T be two transitions in
PEQ relation. The marking evolution of the system, m(τ ), is identical to the one produced
by the system with ti and tj merged into a single transition, according to the rule in Figure
3.

This property derives directly from Eq. 5 and the proportion between P re[P, ti ] and
P re[P, tj ], the marking dynamic behavior of the input and output places of transitions in
PEQ, is described by the same equation of that when ti and tj are merged as in Fig. 3.
Moreover, if the firing rates of transitions in PEQ relation are varied homothetically,
that is, in the same proportion, the variation of the corresponding transition in the reduced
system leads to equivalent behaviors. That gives the possibility to focus the attention of
this work on the following class of nets.

3.2. Mono-T-Semiflow Reducible nets

Before describing the class of nets under study, a classical concept on queuing network
theory is rewritten in Petri net terms [13]. The visit ratio of a transition tj with respect to
ti , v (i) [tj ], is the average number of times tj is fired for each firing of the reference transition
ti . When this visit ratio is analyzed for systems that have a steady-state, v (i) is proportional
to the equilibrium throughput f e .

Definition 3.2 ([13]). Let N be a consistent and conservative net and λ be a firing rate
vector. hN, λi is Mono-T-Semiflow Reducible if the following system has a unique solution:
 (1)

 Cv = 0,

P re[p, ti ]v (1) [ti ] P re[p, tj ]v (1) [tj ]

= , ∀ti , tj in PEQ relation, ∀p ∈ • ti (6)

 λ[ti ] λ[tj ]

 (1)
v [t1 ] = 1.

The class of MTSR nets encompasses an important spectrum of interesting Petri nets
subclasses (see Figure 4). It contains the structurally live and structurally bounded PEQ
nets (the weighted version of the free-choice nets), thus it represents a large and important
generalization of well known net subclasses, such as marked graphs and state machines. In
addition, MTSR nets allow to model cooperative and competitive relationships that arise,
9
 MTSR
MTS

SL&SB CF SL&SB PEQ

SL&SB FA SL&SB JF

Figure 4: Net classes contained in the MTSR net class.

among others, in manufacturing systems, biological and chemical processes, and multipro-
cessor systems.
If the reduction rule of Fig. 3 is applied to MTSR nets, then Mono-T-Semiflow nets are
obtained. An essential performance measure in a wide range of systems is the equilibrium
throughput, which in the case of MTS systems is proportional to the unique T-semiflow of
the net. Thus, it can be expressed as f e = ΛΠC me = αx, where x is the T-semiflow of the
net and α is a scalar that depends on the firing rates λ and on the initial marking m0 of
the system (me is a reachable marking from m0 ), as well as on the net structure. λ and
m0 are parameters that can be changed to modify the equilibrium throughput.
In this sense, the following definitions are introduced to explain the possible equilibrium
throughput changes, due to the variation of the firing rates and of the initial marking.

3.3. Performance properties of MTS TCPN systems

Definition 3.3. Let hN, λ, m0 i be an MTS TCPN system with m0 fixed. Its equilibrium
throughput is
• monotonic with respect to the firing rates, denoted as M (λ), if
∀(λ0 , λ00 ), λ0 ≤ λ00 =⇒ α(λ0 ) ≤ α(λ00 ),

• deadlock-free monotonic with respect to the firing rates, denoted as DF M (λ), if

∀(λ0 , λ00 ) with λ0 ≤ λ00 , α(λ0 ) > 0 =⇒ α(λ00 ) > 0,

• continuous with respect to the firing rates, denoted as C(λ), if for every λ, given  > 0,
there is δ > 0 such that
kλ − λ0 k2 < δ =⇒ kα(λ) − α(λ0 )k2 < 
10
 where k·k2 is the euclidean norm.

Equilibrium throughput monotonicity and continuity are behavioral properties that may
be interpreted as desirable characteristics for a system. For instance, the number of finished
jobs per unit time of a manufacturing system can be modeled as the equilibrium throughput
of a TCPN system, where the firing rates represent the working speed of the machines. Then
it is desirable that increasing the speed of a machine results in a non-slower system and that
this performance change occurs gradually.
These behaviors are also desired with respect to the initial marking, which may model
the buffer capacities and the resources that a manufacturing system has. Analogous to
Definition 3.3, the following one captures this idea.

Definition 3.4. Let hN, λ, m0 i be an MTS TCPN system with λ fixed. Its equilibrium
throughput is

• monotonic with respect to the initial marking, denoted as M (m0 ), if

∀(m00 , m000 ), m00 ≤ m000 =⇒ α(m00 ) ≤ α(m000 ),

• deadlock-free monotonic with respect to the initial marking, denoted as DF M (m0 ), if

∀(m00 , m000 ) with m00 ≤ m000 , α(m00 ) > 0 =⇒ α(m000 ) > 0,

• continuous with respect to the initial marking, denoted as C(m0 ), if for every m0 ,
given  > 0, there is δ > 0 such that

km0 − m00 k2 < δ =⇒ kα(m0 ) − α(m00 )k2 < 

where k·k2 is the euclidean norm.

The manufacturing system of Fig. 5(a) is presented to show the performance properties
of a TCPN system. In that process, raw material arrives to input buffer I; afterwards,
a resource of type R (robot) picks the material and places it into the machine M , and
the resource is released; machine M performs some operations over the material to obtain
a final product; a resource of type R unloads the machine delivering the product to the
output buffer O releasing the resource. The Petri net structure shown in Fig. 5(b) is an
MTS net that models this system. Place p1 represents the input buffer capacity, p2 the
loading machine state, p3 the raw material being processed by M , p4 the unloading machine
state, p5 the machine capacity and p6 the number of resources in the system. Transitions t1 ,
t2 , t3 and t4 represent the events of picking raw material from I, placing raw material into
M , picking final product from M and place final product in O, respectively.

11
 I

M R

(a) Manufacturing Process (b) Petri net

model

Figure 5: A simple manufacturing system and its model.

(a) Relationships among performance properties under firing rate varia-

tions.

(b) Relationships among performance properties under initial marking vari-

ations.

Figure 6: Equilibrium throughput properties of an MTS TCPN system from two alternative perspectives.

4. Population and Velocity: basic system performance properties from two al-
ternative perspectives
Through this section, it is analyzed how the equilibrium throughput changes under the
variation of the firing rates and of the initial marking. Throughput properties (monotonicity,
deadlock-free monotonicity, and continuity) with respect to λ and m0 are discussed and a
relationship between them is established.

4.1. Basic properties with respect to firing rates

Relationships among continuity, monotonicity, and deadlock-free monotonicity with re-
spect to the firing rates are summarized in Fig. 6(a).

Theorem 4.1 ([9]). Let hN, λ, m0 i be an MTS TCPN system with m0 fixed. If the equi-
librium throughput is monotonic with respect to λ, then it is continuous with respect to
λ.
12
 1.4 1.4

1.2 1.2

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 0 1 2 3 4 5 6 7 8

(a) Throughput with respect to λ1 (b) Throughput with respect to m0 (p1 )

Figure 7: Similarities between firing rate and initial marking variations: considering the net in Fig. 5(b),
(a) equilibrium throughput of the system under variation of λ[t1 ], λ = [∗ 1 1 1]T with fixed initial marking
m0 = [5 0 0 0 3 4]T ; (b) equilibrium throughput of the system under variation of m0 [p1 ], m0 = [∗ 0 0 0 3 4]T
with fixed firing rate λ = [2 1 1 1]T .

The converse is not true, Figure 7(a) shows the equilibrium throughput of the TCPN
system hN, λ, m0 i for the net N in Fig. 5(b), provided with the fixed initial marking
m0 = [5 0 0 0 3 4]T , and the firing rate λ = [∗ 1 1 1]T . The first element of λ is varied
evidencing that the equilibrium throughput of the system is non-monotonic with respect
to the firing rates; however, there is no firing rate variation that results in a discontinuous
change of the equilibrium throughput.
Property 4.2. Let hN, λ, m0 i be an MTS TCPN system with m0 fixed. If its equilibrium
throughput is monotonic with respect to λ, then it is deadlock-free monotonic with respect to
λ.
This property is directly derived from Definition 3.3. Nevertheless, the converse is not
true: using the same example of above, the system is non-monotonic with respect to the
firing rates (Figure 7(a)); however, there is no positive firing rate variation that results in a
zero value of the equilibrium throughput.
Concluding, the equilibrium throughput of a TCPN system can be continuous and
deadlock-free monotonic, with respect to the firing rates, and be non-monotonic. This
is because discontinuities and deadlock-free non-monotonicities represent a degradation of
the system performance, the latter being the worst case; thus, these are particular cases of
non-monotonicity.

4.2. Basic properties with respect to initial marking

Since f e also depends on the initial marking, continuity, monotonicity and deadlock-free
monotonicity can be studied with respect to m0 ; their relationships are summarized in Fig.
6(b).
13
Theorem 4.3. Let hN, λ, m0 i be an MTS TCPN system with λ fixed. If the equilibrium
throughput is monotonic with respect to m0 , then it is continuous with respect to m0 .

Proof. By contradiction. Let mv be a vector representing the direction in which the initial
marking is varied and produces a discontinuity in the equilibrium throughput of system
hN, λ, m0 i, that is,

• assume m0 is an initial marking such that f e = α1 x, and

+
• for the initial marking vector m+0 = m0 + δmv , the equilibrium throughput f e = α2 x
is obtained, being δ > 0 an infinitesimal increment and the difference between α1 and
α2 is a finite real value.

If the discontinuity goes from a higher value α1 = αH to a lower one α2 = αL , then the
equilibrium throughput is non monotonic, a contradiction; thus the theorem trivially holds.
Hence, remains to analyze when the throughput goes from α1 = αL to α2 = αH . Increasing δ
times the initial marking in the direction m0 − mv , the new initial marking m∗0 = (1 + δ)m0
is obtained. From the homothetic Property 2.10, the equilibrium throughput of the system
is f ∗e = (1 + δ)αL x. Thus, f ∗e − f +
e is negative, that is, the system is non-monotonic with
respect to m0 , a contradiction. 

The converse of this implication is not true. Figure 7(b) shows the equilibrium through-
put of the TCPN system hN, λ, m0 i for the net N in Fig. 5(b), the fixed firing rate
λ = [2 1 1 1]T , and the initial marking m0 = [∗ 0 0 0 3 4]T . The first element of m0
is varied evidencing that the equilibrium throughput of the system is non-monotonic with
respect to the initial marking; however, there is no initial marking variation leading to a
discontinuous change of the equilibrium throughput.

Property 4.4. Let hN, λ, m0 i be an MTS TCPN system with λ fixed. The equilibrium
throughput is monotonic with respect to m0 , then it is deadlock-free monotonic with respect
to m0 .

This property is directly derived from Definition 3.4. The converse of this implication
is true and it is proved later in this document, because its explanation requires of some
structural objects of the net, which are introduced in the following section.

4.3. Comparison of performance properties with respect to firing rates and initial marking
As already mentioned, the equilibrium throughput depends on both the firing rates and
the initial marking, this allows to identify some similarities between the variation of λ and
m0 . A comparison between Figures 6(a) and 6(b) shows the similarities among both kind
of variations, three of four arrows are identical; nonetheless, the one from DF M (m0 ) to
M (m0 ) in Fig. 6(b) is not equal to its counterpart in Fig. 6(a). Both diagrams are not
symmetric because λ and m0 variations are not exactly equal: the former is a change on the
parameters of the linear matrices that describe each operation mode of the piece-wise linear
14
system, consequently, it represents a variation of their eigenvalues; the latter represents
an initial condition modification of the piece-wise linear system, this does not modify the
eigenvalues of each operation mode, but it modifies the reachability space of the TCPN
system.
Referring again to Fig. 7: in one hand (Figure 7(a)), the manufacturing system perfor-
mance can be improved by increasing the activity of loading machine M (λ[t1 ] is increased).
Nonetheless, there exists a value where increasing this activity does not improve the system
performance, even worst it reduces the equilibrium throughput. On the other hand (Fig-
ure 7(b)), if the input buffer capacity provided to the system is increased (i.e., m0 [p1 ] is
increased), the system performance is enhanced until a maximum is reached; if it is still
increased, the performance is diminished and becomes a worst problem when a deadlock is
reached due to this increment. This points out that even when λ and m0 variations are not
exactly equal, they produce a similar change on the equilibrium throughput.
The underlying cause of the undesired behaviors is the design of the manufacturing
process, thus it is due to the Petri net structure. In this sense, it is important to remember
that the equilibrium throughput f e depends on λ and m0 as parameters that can be varied,
and also depends on the Petri net structure which is codified in the configuration matrix
ΠC . Structural objects such as configurations allow to comprehend why undesired behaviors
arise under initial marking and firing rate variations, they provide better explanations of
system malfunctioning. The next section is devoted to characterize those behaviors using
the structure of the net.

5. Performance properties characterized under a common framework

Previous section presented some relationships among behavioral properties of the equilib-
rium throughput subject to firing rates and initial marking variations. Now in this section,
these behavioral properties are related with structural objects of the net with the aim of
obtaining more knowledge and a better understanding of why undesired behaviors arise in
a TCPN system. Through this section, relationships shown in Figure 8 are addressed.
In order to introduce this section, consider the marking evolution of the TCPN system
in Example 2.9 (Figure 2). It starts at m0 = [2 0 2 0]T (the initial marking belongs to
region R4 ). As previously said, due to the firing rates λ = [6 3 1]T , the marking evolves
until me = [0 2 0 2]T is reached, with equilibrium throughput f e = 0. Alternatively, if the
firing rates are λ = [1 3 1]T , then the system evolves converging to the equilibrium marking
me = [ 12 2 8 6 T
7 7 7 7
] , with equilibrium throughput f e = 67 · [1 1 1]T . Thus, the equilibrium
throughput of this system is non DF M (λ), and consequently, it is non M (λ).
Taking a look at both equilibrium markings, mD T
e = [0 2 0 2] belongs to R5 ∩ R1 (the
L 12 2 8 6 T
border of both regions), and me = [ 7 7 7 7 ] belongs to region R4 . Regions R5 and R1
are associated with configurations C5 and C1 , while TC5 ≡ TC1 . Furthermore, there exists
no P-semiflow whose support is contained in TC5 , it will be shown that this is a necessary
condition for an equilibrium marking having a null equilibrium throughput. On the other
hand, region R4 is associated with configuration C4 and with T-coverture TC4 which contains

15
Figure 8: Relationships among performance properties w.r.t. λ and m0 and the configurations of MTS
TCPN systems. Continuity and monotonicity w.r.t. λ and their connection with the structural properties
have been studied in [9]. They are complemented in this work by studying these properties w.r.t. m0
improving results of [12]. Relationships between configurations and deadlock-free monotonicity with respect
to λ and m0 are also incorporated.

the support of the P-semiflow y = [1 1 0 0]T , this means that any equilibrium marking
belonging to this region has an equilibrium throughput greater than zero.
Due to the equilibrium throughput dependency on the configuration matrices, the fol-
lowing natural question arises: is there a connection between the structural objects of a net
(configurations, T-covertures) and the behavioral properties represented in Figures 6(a) and
6(b)? In order to answer this question, it is important to focus on the possible equilibrium
markings of the system and in the regions they belong.

5.1. Suitable and problematic configurations

An equilibrium marking may activate a configuration whose associated T-coverture either
contains the support of a P-semiflow or not. This makes possible to relate the net structural
objects with the performance behaviors of the TCPN system. In order to get information
about the possible behaviors, let us classify such structural objects.

Definition 5.1. Let N be an MTS net and C ∈ SC(N ) be a configuration of the net:

• C is suitable if its T-coverture contains the support of a P-semiflow. Otherwise, it is

called problematic,

• a problematic configuration C is nB-Problematic if its T-coverture contains the support

of a P-flow. Otherwise, it is called B-Problematic.

16
 B- and nB- notations in problematic configurations refer to the existence and nonexis-
tence of bifurcations, respectively. Discontinuities due to firing rate variations have been
explained through bifurcations in [14], and, additionally, taking advantage of the structure
in [9]. The following is a purely structural result (it only depends on N ) which relates the
two kind of problematic configurations.

Lemma 5.2. Let N be an MTS net. If there exists an nB-Problematic configuration, then
there exists a B-Problematic configuration.

Proof. Assume C is an nB-Problematic configuration. Thus, the support of a P-flow y is

contained in TC . In other words, the induced subnet of C has a P-flow. The set of places
kyk can be split in ky + k and ky − k (the positive and negative entries of y). Necessarily,
there are common output transitions to both set of places, ky + k• ∩ ky − k• 6= ∅, and common
input transitions, • ky + k ∩ • ky − k =
6 ∅; otherwise, y + and y − would be P-semiflows and
configuration C would be suitable.
For every p ∈ kyk, its associated row of the token-flow matrix C[p, T ] is a linear combi-
nation of the rows C[q, T ], with q ∈ kyk \ {p}. Hence, there exists a place pr ∈ kyk such
that p•r ⊂ ky + k• ∩ ky − k• . Otherwise stated, every output transition of pr is contained in
the set of common output transitions of ky + k and ky − k, which are necessarily join transi-
tions. Thus, this place pr can be removed with its input and output arcs, obtaining a subnet
associated with a configuration in which pr is not contained in its T-coverture.
This reasoning can be done until the removal of such places leads to a subnet with no
P-flows in which all transitions are covered. This means that a B-Problematic configuration
is obtained. 

Lemma 5.2 can be applied for the net in Figure 5(b). Configuration C2 is nB-Problematic,
the support of P-flow y = [−1 0 0 0 1 1]T is contained in TC2 = {p1 , p4 , p5 , p6 }. In this case,
the output transitions of place p1 are contained in the set of common output transitions of
ky + k and ky − k (i.e., {p1 }• ∩{p5 , p6 }• = t1 ). Thus, p1 can be removed together with its input
and output arcs, maintaining every transition covered. The resulting subnet is associated
to configuration C1 which is B-Problematic (see Figure 9). The converse of Lemma 5.2
is not true. For example, the net in Fig. 1(a) has six configurations, three of them are
B-Problematic and the other three are suitable configurations.
Using the configurations classification, some structural properties parametrized by equi-
librium markings can be defined as well.

Definition 5.3. Let hN, λ, m0 i be an MTS TCPN system.

• The system is suitable, denoted by S, if every reachable equilibrium marking me ac-

tivates a suitable configuration,

• The system is nB-Problematic, denoted by nB, if every reachable equilibrium mark-

ing me that does not activate a suitable configuration, activates only nB-Problematic
configurations.
17
 Figure 9: Subnets associated with configurations of net in Fig. 5(b).

• The system is B-Problematic, denoted as B, if there exists an equilibrium marking

me that does not activate a suitable configuration and activates a B-Problematic one.

The above definitions make possible to characterize the behavioral properties of a system
—continuity, monotonicity and deadlock-free monotonicity w.r.t. λ and m0 — using its
structural properties as a common framework.

5.2. Basic properties w.r.t. the firing rates and the structure
Monotonicity and continuity of the equilibrium throughput as a function of the firing
rates and its relation with the suitability property in Definition 5.3, have been previously
considered in [9]. Particularly, the implication S ⇒ C(λ) was derived, and the equivalence
between M (λ) and S was also presented. Those results are rewritten here for comparing
them with the new ones due to initial marking variations.

Theorem 5.4 ([9]). Let hN, λ, m0 i be an MTS TCPN system with m0 fixed. If the equilib-
rium throughput is discontinuous with respect to λ, then there exists an equilibrium marking
me that does not activate any suitable configuration.

Theorem 5.5 ([9]). Let hN, λ, m0 i be an MTS TCPN system with m0 fixed. The equilib-
rium throughput is non monotonic with respect to λ iff there exists an equilibrium marking
me that does not activate any suitable configuration.
18
 The sufficiency of this last result could be proven by expressing S in its equivalent form
¬S ⇐⇒ nB ∨ B, and by proving that nB =⇒ ¬M (λ) and B =⇒ ¬M (λ). This
representation clarifies the relationships among the performance and structural properties.
The top of Figure 8 shows in a contrapositive manner the relationships among continuity,
monotonicity and deadlock-free monotonicity with respect to λ presented in Fig. 6(a),
highlighting how they are related to structural properties.
Since ¬DF M (λ) =⇒ ¬M (λ) and ¬M (λ) ⇐⇒ nB ∨ B, it follows that if the
equilibrium throughput of a TCPN system is non deadlock-free monotonic with respect to
λ, then nB or B is true. In this regard, the following result refines this implication.

Theorem 5.6. Let hN, λ, m0 i be an MTS TCPN system with λ fixed. If the equilibrium
throughput is non deadlock-free monotonic with respect to λ, then there exists an equilibrium
marking me that does not activate a suitable configuration and activates a B-Problematic
one.

Proof. Assuming that the equilibrium throughput f e is non deadlock-free monotonic with
respect to λ, it follows that there exist an equilibrium marking me such that f e = 0. Thus,
at least one input place of every transition is unmarked at me . Let P0 = {p | me [p] = 0}
be such set of unmarked places, then P0 does not contain the support of any P-semiflow
(me does not activate any suitable configuration). Hence, P0 contains a T-coverture TC of
a problematic configuration C. If C is B-Problematic, then me is activating it; on the other
hand, if C is nB-Problematic, from Lemma 5.2, a B-Problematic configuration exists, this
last configuration is also activated by me . 

Theorem 5.6 states that ¬DF M (λ) =⇒ B, the converse is not true. Figure 10(a)
shows a TCPN system that has an equilibrium marking activating only a B-Problematic
configuration (B property); nonetheless, it is DF M (λ) —from the fixed initial marking
m0 = [3 0 0 2 0]T , it is not possible to reach any deadlock.
The relationships shown at the upper half of Figure 8 (w.r.t. λ) have their counterpart
at the lower part of the same figure, but these are related to the variation of the initial
marking (w.r.t. m0 ).

5.3. Basic properties w.r.t. initial marking and the structure

In this part, continuity, monotonicity and deadlock-free monotonicity of the equilibrium
throughput with respect to the initial marking and their relationship with the structural
properties nB and B, are considered.
Similar to Theorem 5.6, the following result, concerning to deadlock-free monotonicity
with respect to the initial marking, can be settled. Actually, its proof is omitted because it
is practically identical to the one of Theorem 5.6.

Theorem 5.7. Let hN, λ, m0 i be an MTS TCPN system with λ fixed. If the equilib-
rium throughput is non deadlock-free monotonic with respect to m0 , then there exists an
equilibrium marking me that does not activate a suitable configuration and activates a B-
Problematic one.
19
 (a) B property does not imply (b) B property does not imply
¬DF M (λ) ¬M (m0 ) nor ¬DF M (m0 )

Figure 10: Two TCPN systems with B property: (a) its equilibrium throughput is DF M (λ); (b) its
equilibrium throughput is M (m0 ) and DF M (m0 )

The reverse is not true. Figure 10(b) shows a TCPN system that has an equilibrium
marking activating only a B-Problematic configuration (B property), in fact the net has
only two B-Problematic configurations. For the fixed firing rates λ = [2 2 5]T the TCPN
system is DF M (m0 ) —actually, any firing rate vector fulfilling λ[t1 ] = λ[t2 ] does not allow
the marking system to reach a deadlock.
Now, a result from [8] is recalled. It says that if a TCPN system has the suitability
property, then its equilibrium throughput is monotonic with respect to the initial marking.
This result is rephrased using its contrapositive sentence.

Theorem 5.8 ([8]). Let hN, λ, m0 i be an MTS TCPN system with λ fixed. If the equi-
librium throughput is non monotonic with respect to m0 , then there exists an equilibrium
marking me that does not activate any suitable configuration.

In contrast with the analogous Theorem 5.5, the converse of Theorem 5.8 does not hold.
The TCPN system in Figure 10(b) works as a counterexample: the system is B-Problematic,
however for the given firing rates the equilibrium throughput is monotonic with respect to
m0 . In spite of this, in the case in which the system has the nB-Problematic property, then
it can be concluded that the equilibrium throughput is ¬M (m0 ).

Proposition 5.9. Let hN, λ, m0 i be an MTS TCPN system with λ fixed. If every reachable
equilibrium marking me that does not activate a suitable configuration, activates only an nB-
Problematic configuration, then the equilibrium throughput is non monotonic with respect to
m0 .
20
Proof. Let me be an equilibrium marking that does not activate a suitable configuration
and activates only an nB-Problematic configuration C, with equilibrium throughput αx =
ΛΠC me ; thus, from Theorem 5.7, this equilibrium throughput is such that α > 0. Let y be
a P-flow such that kyk ⊆ TC and y T me > 0. Let us construct a |T |-sized vector w with its
k-th component of the form
1
w[tk ] = P re(pj , tk )y[pj ], with (pj , tk ) ∈ C.
λ[tk ]

Premultiplying the throughput equation by w results in αwT x = y T me because of its

y T m0
construction. Since y T me = y T m0 , equation α = is obtained. With fixed λ value,
wT x
the product wT x only can be a positive constant number K (because α > 0). Since the
P-flow contains at least one negative element, say y[pj ], an initial marking m00 can be found
such that only the marking of place pj is increased, that is, in the vector m00 − m0 ≥ 0
every component is null except for the positive element associated to pj , then α0 − α =
K
y (m00 − m0 ) < 0, hence the system is non monotonic with respect to m0 .
1 T


The following result shows the equivalence between M (m0 ) and DF M (m0 ).

Theorem 5.10. Let hN, λ, m0 i be an MTS TCPN system with fixed firing rates λ. The
equilibrium throughput is deadlock-free monotonic with respect to m0 if, and only if, it is
monotonic with respect to m0 .

Proof. For the sufficiency: follows directly from Definition 3.4.

For the necessity: assuming deadlock-free monotonicity with respect to m0 , if this is
because equilibrium markings only activate suitable configurations, then the equilibrium
throughput is monotonic with respect to m0 (Theorem 5.8).
Assuming DF M (m0 ), it follows that there is no equilibrium marking activating an nB-
Problematic configuration CnB : if that were the case, the initial marking could be increased so
that the equilibrium throughput is reduced (Proposition 5.9); moreover, if m0 is increased
so that the equilibrium marking reaches the B-Problematic configuration associated with
CnB , a null equilibrium throughput is obtained, contradiction.
It remains to analyze the equilibrium throughput when an equilibrium marking me
activates only a B-Problematic configuration CB . Since the equilibrium throughput is
DF M (m0 ), the equilibrium marking must be greater than zero which is possible only when
relation between each pair of transitions in conflict, ta and tb , fulfills:

λa x[ta ]P re0 [pc , ta ]

= , where (pc , ta ), (pc , tb ) ∈ C. (7)
λb x[tb ]P re0 [pc , tb ]

This relation induces a conservative marking law within the region RCB , which leads to the
same condition (suitability property) that guarantees the monotonicity of the equilibrium
throughput with respect to m0 . 

21
 The most remarkable advantage that the aforementioned results provide is the capability
to link performance properties with structural objects of the net. As can be noted from
Figure 8, if the equilibrium throughput is non deadlock-free monotonic (either with respect
to λ or m0 ), it is because a B-Problematic configuration is activated by an equilibrium
marking; this can be exploited to avoid deadlock situations, as will be explained in the
next section. Figure 8 also shows that if an nB-Problematic configuration is activated, the
increase of the firing rates or the initial marking results in the decrease of the equilibrium
throughput; this represents information that can be used to enforce the monotonicity of
the equilibrium throughput. In this way, the appearance of undesired behaviors could be
avoided by taken into account some restrictions on the initial marking and on the firing
rates.

6. Firing rates and initial marking sets for ensuring monotonicity properties
This section analyzes how structural properties provide an insight about how λ and
m0 can be constrained to enforce monotonicity and deadlock-free monotonicity in a TCPN
system. T-covertures and its induced subnets contribute to find a set of firing rates and of
initial markings that ensure the desired properties of a system. An immediate property of
T-covertures is the following.

Property 6.1. Given a configuration C of an MTS net, its associated T-coverture TC is a

siphon.

The net in Fig. 5(b) has eight configurations (see Fig. 9):

• C1 , with T-coverture TC1 = {p4 , p5 , p6 }, is a B-Problematic configuration;

• C2 , with T-coverture TC2 = {p1 , p4 , p5 , p6 }, is an nB-Problematic configuration;

• C3 to C8 are suitable configurations.

The associated T-covertures of configurations C3 to C8 , contain the support of a P-semiflow

(those siphons cannot get emptied if initially marked). Configuration C1 has a subnet in
which the firing of t1 reduces the marking in its T-coverture (the siphon can be emptied).
For configuration C2 , its T-coverture can also be emptied because of t1 , but it also depends
on the marking of place p1 (the siphon can be controlled).
The fact that any equilibrium marking activating a suitable configuration has an as-
sociated non null equilibrium throughput, can also be obtained by reasoning using the
T-covertures; since TC contains the support of a P-semiflow initially marked, the siphon TC
cannot be emptied, thus it is not possible to obtain a null equilibrium throughput (i.e., a
deadlock).
T-covertures are structural objects that provide information about how firing rates and
the initial marking can be constrained for avoiding undesired behaviors in a TCPN system.

22
6.1. Deadlock-free monotonicity with respect to an initial marking set
From Theorems 5.6 and 5.7, when the equilibrium throughput is non deadlock-free mono-
tonic either with respect to λ or m0 , necessarily an equilibrium marking me exists that
activates a B-Problematic configuration in which its associated equilibrium throughput is
f e = 0. If it is possible to restrict the reachability of such markings (by restricting the
initial marking), then the system will be deadlock-free.
Given an MTS net N , if the T-coverture of every B-Problematic configuration of N
is contained in the T-coverture of an nB-Problematic configuration, then the invariant law
associated to its P-flow y allows to find the set of initial markings such that the CPN system
hN, m0 i is deadlock-free.

Proposition 6.2. Let N be an MTS net. The system hN, m0 i is deadlock-free if the two
following conditions hold:
1. for every B-Problematic configuration CB there exists an nB-Problematic one CnB such
that TCB ⊂ TCnB ; and
2. for a P-flow y whose support is in TCnB with y[p] < 0 where p ∈ TCnB , p 6∈ TCB , the
inequality y T m0 > 0 holds.

Proof. Let TCB be the T-coverture of an arbitrary B-Problematic configuration, which

is contained in TCnB , the T-coverture of an nB-Problematic configuration. If an initial
marking m0 meets the second condition, then for every reachable marking m of the system
y T m = y T m0 > 0. This can be rewritten as y +T m = y T m0 + y −T m > 0, where y +
and y − are non negative vectors such that y = y + − y − . Thus, at least one place in TCB
has a positive marking at m, that is, TCB cannot be emptied. Since this is true for every
B-problematic configuration, the system hN, m0 i is deadlock-free. 

Following with the net in Fig. 5(b), deadlock-freeness can be guaranteed by avoiding
every marking activating C1 with null marking in places p4 , p5 , and p6 . In this net TC1 ⊂ TC2 ,
and the support of y = [−1 0 0 0 1 1]T is contained in TC2 . Restricting the initial marking
to fulfill the inequality y T m0 > 0 (that is, the initial marking is in M0 = {m0 | m0 [p1 ] <
m0 [p5 ] + m0 [p6 ]}), every reachable marking m satisfies m[p5 ] + m[p6 ] > 0. Therefore,
from Proposition 6.2, the TCPN system is deadlock-free. Note that this is achieved without
setting any restriction on the firing rates, thus they remain as free variables for improving
the equilibrium throughput, without the risk of reaching a deadlock.

Corollary 6.3. Let hN, λ, m0 i be an MTS TCPN system with an arbitrary fixed λ. The
equilibrium throughput of the system is deadlock-free monotonic with respect to

M0 = {m0 | conditions on Proposition 6.2 are satisfied}.

For the above example, the physical interpretation of this restriction is that the input
buffer capacity of the manufacturing process cannot exceed the machine capacity and the
number of resources on the system. This guarantees that at least one resource is available
for unloading the machine, avoiding a blocking situation.
23
 2

1.5

0.5

0
4

3
4
2 3.5
3
2.5
1 2
1.5
1
0 0.5
0

Figure 11: Equilibrium throughput depending on firing rates.

6.2. Monotonicity with respect to a firing rate set

Now a set of firing rates for a given CPN system hN, m0 i is obtained, the intention is that
the timed system exhibits a monotonic equilibrium throughput with respect to such a set.
From Theorem 5.5, when an equilibrium marking activates an nB-Problematic configuration,
there exists a set of transitions for which its firing rate variation produces a decrease in
the equilibrium throughput of the system. This set can be easily determined using the
corresponding T-covertures.

Property 6.4. Let hN, m0 i be an MTS CPN system and SCnB (N ) be its set of nB-Problematic
configurations that can be activated. If for every C ∈ SCnB (N ), with associated P-flow
y = y + − y − , the firing rates of transitions in ky − k• are not increased, then the equilibrium
throughput of hN, λ, m0 i does not decrease.

When an equilibrium marking is activating an nB-Problematic configuration, the func-

tion α has negative values because of the P-flow y associated to the nB-Problematic config-
uration. The equilibrium throughput (function α) decreases by increasing the firing rate of
a transition which is restricted by a place p such that the entry y[p] is negative.
In order to illustrate Property 6.4, consider the TCPN system hN, λ, m0 i where N is
the net in Fig. 5(b) with initial marking m0 = [5 0 0 0 3 4]T , and λ = [λ1 1 λ3 1]T be its
firing rates, where λ1 and λ3 are taken as variation parameters. The equilibrium throughput
of the system α, that depends on λ1 and λ3 , is shown in Fig. 11. When an equilibrium
marking activates configuration C2 , the increase of λ1 produces a decrease of the equilibrium
24
throughput. In this particular configuration, the variation of λ must be restricted such that
λ1 does not increase.

7. Concluding remarks
This paper analyzes three behavioral properties of the equilibrium throughput in MTS
TCPN systems:monotonicity, continuity, and deadlock-free monotonicity. In particular, their
dependencies on the firing rate and initial marking variations has been deepened. Similarities
between Figures 6(a) and 6(b) were explained through the dependency of f e on λ and
m0 . They are not exactly equal due to the difference between both kind of variations;
while a change in λ implies a change of the operation modes, the change in m0 implies
a different initial condition. Behavioral properties were characterized using configurations,
these last structural objects are the actual reason for the presence of undesired behaviors in
the equilibrium throughput. Figure 8 summarizes the relationships between them and the
structural properties of a TCPN system. Based on these relationships, parameters λ and
m0 are computed to ensure the desired properties of the equilibrium throughput.
Attempting to analyze or enforce monotonicity or deadlock-free monotonicity, may result
in a difficult problem because of the potentially large number of configurations in a net, which
can be exponential. In our current research, we have study some reduction techniques that
allow to alleviate the practical complexity of such computation [15], but a major issue still
remains: the number of problematic configurations may be too large. It is our belief that this
set can be fully characterized using particular structures containing choice places and join
transitions —another set but with polynomial size. The main idea is to identify concisely
the root of problematic configurations through these nodes.

References
[1] F. Vernadat, F. Dicesare, G. Harhalakis, J. Proth, M. Silva, Practice of Petri nets in manufacturing,
Chapman and Hall, 1993.
[2] M. A. Marsan, G. Balbo, G. Conte, S. Donatelli, G. Franceschinis, Modelling with Generalized Stochas-
tic Petri Nets, 1st Edition, John Wiley & Sons, Inc., New York, NY, USA, 1994.
[3] D. Angeli, P. D. Leenheer, E. D. Sontag, A Petri net approach to the study of persistence in chemical
reaction networks, Mathematical Biosciences 210 (2) (2007) 598 – 618.
[4] M. Fanti, A. Mangini, M. Dotoli, W. Ukovich, A three-level strategy for the design and performance
evaluation of hospital departments, Systems, Man, and Cybernetics: Systems, IEEE Transactions on
43 (4) (2013) 742–756.
[5] T. Murata, Petri nets: Properties, analysis and applications, Proceedings of the IEEE 77 (4) (1989)
541–580.
[6] R. David, H. Alla, Discrete, Continuous, and Hybrid Petri Nets, 2nd Edition, Springer Publishing
Company, Incorporated, 2010.
[7] M. Silva, J. Júlvez, C. Mahulea, C. R. Vázquez, On fluidization of discrete event models: observation
and control of continuous Petri nets, Discrete Event Dynamic Systems 21 (4) (2011) 427–497.
[8] C. Mahulea, L. Recalde, M. Silva, Basic server semantics and performance monotonicity of continuous
Petri nets, Discrete Event Dynamic Systems 19 (2) (2009) 189–212.
[9] M. Navarro-Gutiérrez, A. Ramı́rez-Treviño, M. Silva, Homothecy, bifurcations, continuity and mono-
tonicity in timed continuous Petri nets under infinite server semantics, Nonlinear Analysis: Hybrid
Systems 26 (2017) 48 – 67.
25
[10] C. R. Vázquez, M. Silva, Timing and liveness in continuous Petri nets, Automatica 47 (2) (2011) 283
– 290.
[11] D. Lefebvre, E. Leclercq, Piecewise constant timed continuous PNs for the steady state estimation of
stochastic PNs, Discrete Event Dynamic Systems 22 (2) (2012) 179–196.
[12] M. Navarro-Gutiérrez, A. Ramı́rez-Treviño, M. Silva, On the duality between firing rate and initial
marking variations in TCPN, in: 2016 IEEE 21st International Conference on Emerging Technologies
and Factory Automation (ETFA), 2016, pp. 1–8.
[13] J. Júlvez, L. Recalde, M. Silva, Steady-state performance evaluation of continuous mono-t-semiflow
Petri nets, Automatica 41 (4) (2005) 605 – 616.
[14] A. Meyer, Discontinuity induced bifurcations in timed continuous Petri nets, in: WODES’12: 11th
International Workshop on Discrete Event Systems, Guadalajara, México, Vol. 11, 2012, pp. 28–33.
[15] M. Navarro-Gutierrez, A. Ramı́rez-Treviño, M. Silva, On monotonicity and continuity: computing prob-
lematic configurations in timed continuous Petri nets, in: WODES’18: 14th International Workshop
on Discrete Event Systems, Sorrento, Italy, 2018.

26