D-26111 Oldenburg, Germany (22email: evgeny.erofeev@informatik.uni-oldenburg.de)
The Complexity of Boolean State Separation (Technical Report)
Abstract
For a Boolean type of nets , a transition system is synthesizeable into a -net if and only if distinct states of correspond to distinct markings of , and prevents a transition firing if there is no related transition in . The former property is called -state separation property (-SSP) while the latter – -event/state separation property (-ESSP). is embeddable into the reachability graph of a -net if and only if has the -SSP. This paper presents a complete characterization of the computational complexity of -SSP for all Boolean Petri net types.
Keywords:
Boolean Petri nets, Boolean state separation, complexity characterization1 Introduction
Providing a powerful mechanism for the modeling of conflicts, dependencies and parallelism, Petri nets are widely used for studying and simulating concurrent and distributed systems. In system analysis, one aims to check behavioral properties of system models, and many of these properties are decidable [7] for Petri nets and their reachability graphs, which represent systems’ behaviors. The task of system synthesis is opposite: a (formal) specification of the system’s behavior is given, and the goal is then to decide whether this behavior can be implemented by a Petri net. In case of a positive decision, such a net should be constructed.
Boolean Petri nets form a simple yet rich and powerful family of Petri nets [3, 4, 8, 10, 12, 13, 16], applied in asynchronous circuits design [24, 26], concurrent constraint programs [9] and analysis of biological systems models [6]. In Boolean nets, each place contains at most one token, for any reachable marking. Hence, a place can be interpreted as a Boolean condition that is true if marked and false otherwise. A place and a transition of such a net are related by one of the Boolean interactions that define in which way and influence each other. The interaction inp (out) defines that must be true (false) before and false (true) after ’s firing; free (used) implies that ’s firing proves that is false (true); nop means that and do not affect each other at all; res (set) implies that may initially be both false or true but after ’s firing it is false (true); swap means that inverts ’s current Boolean value. Boolean Petri nets are classified by the sets of interactions that can be applied. A set of Boolean interactions is called a type of net, and a net is of type (a -net) if it applies at most the interactions of . For a type , the -synthesis problem consists in deciding whether a specification given in the form of a labeled transition system (TS) is isomorphic to the reachability graph of some -net , and in constructing if it exists. The complexity of -synthesis has been studied in different settings [17, 20, 22], and varies substantially from polynomial [16] to NP-complete [2].
In order to perform synthesis, that is, to implement the behavior specified by the given TS with a -net, two general problems have to be resolved: The -net has to distinguish the global states of the TS, and the -net has to prevent actions at states where they are not permitted by the TS. In the literature [3], the former requirement is usually referred to as -state separation property (-SSP), while the latter – -event/state separation property (-ESSP). Both -SSP and -ESSP define decision problems that ask whether a given TS fulfills the respective property. The present work focuses exclusively on the computational complexity of -SSP. The interest to state separation is motivated in several ways. First, many synthesis approaches are very sensitive to the size of the input’s state space. This raises the question if some initial, so-called pre-synthesis procedures [5, 25] can be employed as a quick-fail mechanism, i.e., techniques with a small computational overhead that would gain some helpful information for the main synthesis, or reject the input if exact (up to isomorphism) synthesis is not possible. Since -synthesis allows a positive decision if and only if -SSP and -ESSP do [3], an efficient decision procedure for -SSP could serve as a quick-fail pre-process check. Second, if exact synthesis is not possible for the given TS, one may want to have a simulating model, i.e., a -net that over-approximates [14, 15] the specified behavior with some possible supplement. Formally, the TS then has to be injectively embeddable into the reachability graph of a -net. It is well known from the literature [3] that a TS can be embedded into the reachability graph of a -net if and only if it has the -SSP. Finally, in comparison to -ESSP, so far the complexity of -SSP is known to be as hard [1, 16, 22, 23] or actually less hard [18, 19]. On the contrary, in this paper, for some types of nets, deciding the -SSP is proven harder (NP-complete) than deciding the -ESSP (polynomial), e.g., for . From the contribution perspective, the -SSP has been previously considered only in the broader context of -synthesis, and only for selected types [16, 19, 21, 22]. In this paper, we completely characterize the complexity of -SSP for all 256 Boolean types of nets and discover 150 new hard types, cf. §1 - §3, §6, §9 of Figure 1, and 4 new tractable types, cf. Figure 1 §10, and comprise the known results for the other 102 types as well, cf. Figure 1 §4, §5, §7, §8. In particular, our characterization categorizes Boolean types with regard to their behavioral capabilities, resulting from the Boolean interactions involved. This reveals the internal organization of the entire class of Boolean nets, suggesting a general approach for reasoning about its subclasses.
This paper is organized as follows. In Section 2, all the necessary notions and definitions will be given. Section 3 presents NP-completeness results for the types with nop-interaction. -SSP for types without nop is investigated in Section 4. Concluding remarks are given in Section 5. One proof is shifted to the appendix.
Type of net | Complexity | Quantity | |
---|---|---|---|
1 | with | NP-complete | 16 |
2 |
with ,
with |
NP-complete | 32 |
3 |
with ,
with , with |
NP-complete | 32 |
4 |
with ,
with |
polynomial | 16 |
5 | and | NP-complete | 2 |
6 |
and ,
and with |
NP-complete | 10 |
7 |
with ,
with |
polynomial | 20 |
8 | polynomial | 64 | |
9 | with and | NP-complete | 60 |
10 | with | polynomial | 4 |
2 Preliminaries
In this section, we introduce necessary notions and definitions, supported by illustrations and examples, and some basic results that are used throughout the paper.
Transition Systems. A (finite, deterministic) transition system (TS, for short) is a directed labeled graph with the set of nodes (called states), the set of labels (called events) and partial transition function . If is defined, we say that occurs at state , denoted by . By , we denote . This notation extends to paths, i.e., denotes for all . A TS is loop-free, if implies . A loop-free TS is bi-directed if implies . We say is a simple bi-directed path if for all with . An initialized TS is a TS with a distinct initial state , where every state is reachable from by a directed labeled path.
Boolean Types of Nets [3]. The following notion of Boolean types of nets allows to capture all Boolean Petri nets in a uniform way. A Boolean type of net is a TS such that is a subset of the Boolean interactions: . Each interaction is a binary partial function as defined in Figure 4. For all and all , the transition function of is defined by . Notice that a type is completely determined by . Hence we often identify with , cf. Figure 1. Moreover, since captures all meaningful Boolean interactions [22, p. 617] and is defined by , there are 256 Boolean types of nets at all. For a Boolean type of net , we say that its state is inside and is outside. An interaction exits if , enters if , saves 1 if and saves 0 if . Accordingly, we group interactions together by , , , and .
For a net of type (-net), the interactions of determine relations between places and transitions of the net. For instance, if a place and a transition are related via inp, then has to be marked (true) to allow to fire, and becomes unmarked (false) after the firing (cf. Figure 4). Since we are only concerned with state separation, we omit the formal definition of -nets and rather refer to, e.g., [3] for a comprehensive introduction to the topic.
-Regions. The following notion of -regions is the key concept for state separation. A -region of TS consists of the mappings support and signature , such that for every edge of , the edge belongs to the type ; we also say allows ( and thus the region) . If is a path in , then is a path in . We say is the image of (under ). For region and path , event with is called state changing on the path , if for in . Notice that is implicitly defined by and : Since is reachable, for every state , there is a path such that . Thus, since is deterministic, we inductively obtain by for all and . Hence, we can compute and thus purely from and , cf. Figure 4. If , then a -region of a TS is called normalized if for as many events as allows: for all , and if then there is such that .
-State Separation Property. A pair of distinct states of defines a state separation atom (SSP atom). A -region solves if . If exists, then is called -solvable. If and, for all , the atom is -solvable, then is called -solvable. A TS has the -state separation property (-SSP) if all of its SSP atoms are -solvable. A set of -regions of is called -separative if for each SSP atom of there is a -region in that solves it. By the next lemma, if , then has the -SSP if and only if it has a -separative set of normalized -regions:
Lemma 1.
Let be a TS and be a nop-equipped Boolean type of nets. There is a -separative set of if and only if there is a -separative set of normalized -regions of .
Proof.
The if-direction is trivial. Only-if: Let be a non-normalized -region, i.e., there is such that implies . Since is nop-equipped, for all . Thus, a -region can be constructed from , where is equal to except for . Since is finite, a normalized region can be obtained from by inductive application of this procedure. ∎
By the following lemma, -SSP and -SSP are equivalent if and are isomorphic:
Lemma 2 (Without proof).
If and are isomorphic types of nets then a TS has the -SSP if and only if has the -SSP.
In this paper, we consider the -SSP also as decision problem that asks whether a given TS has the -SSP. The decision problem -SSP is in NP: By definition, has at most SSP atoms. Hence, a Turing-machine can (non-deterministically) guess a -separative set such that and (deterministically) check in polynomial time its validity if it exists.
In what follows, some of our NP-completeness results base on polynomial-time reductions of the following decision problem, which is known to be NP-complete [11]:
Cubic Monotone 1-in-3 3Sat. (CM 1-in-3 3Sat) The input is a Boolean formula of negation-free three-clauses , where , with set of variables ; every variable occurs in exactly three clauses, implying . The question to decide is whether there is a (one-in-three model) satisfying for all .
Example 1 (CM 1-in-3 3Sat).
The instance of CM 1-in-3 3Sat with set of variables and clauses , , , , and has the one-in-three model .
3 Deciding the State Separation Property for nop-equipped Types
In this section, we investigate the computational complexity of nop-equipped Boolean types of nets. For technical reasons, we separately consider the types that include neither res nor set (§5 - §7 in Figure 1) and the ones that have at least one of them (§1 - §4 in Figure 1).
First of all, the fact that -SSP is polynomial for the types of §7 in Figure 1 is implied by the results of [16][22, p. 619]. Moreover, for the types of Figure 1 §5 the NP-completeness of -SSP has been shown in [23] (), and in [19] (, there referred to as -bounded P/T-nets). Thus, in order to complete the complexity characterization for the nop-equipped types that neither contain res nor set, it only remains to ascertain the complexity of -SSP for the types Figure 1 §6. The following Subsection 3.1 proves that -SSP is NP-complete for these types.
Then, we proceed with the types of §1 - §4 in Figure 1. The fact that -SSP is polynomial for the types of §4 follows from [22, p. 619]. The NP-completeness of -SSP for the remaining types (§1 - §3) will be demonstrated in Subsection 3.2.
3.1 Complexity of -SSP for nop-equipped Types without res and set
The following theorem summarizes the complexity for the types of §5 - §7 in Figure 1.
Theorem 3.1.
Let be a nop-equipped Boolean type of nets such that . The -SSP is NP-complete if and , otherwise it is polynomial.
As just discussed in Section 3, to complete the proof of Theorem 3.1 it remains to characterize the complexity of -SSP for the types of Figure 1 §6. Since -SSP is NP-complete if [23], by Lemma 1, the -SSP is also NP-complete if and .
Thus, in what follows, we restrict ourselves to the types and , where , and argue that their -SSP are NP-complete. To do so, we let and show the hardness of -SSP by a reduction of CM 1-in-3 3Sat. By Lemma 1, this also implies the hardness of -SSP, where . Furthermore, by Lemma 2, the latter shows the NP-completeness of -SSP if and . The following paragraph introduces the intuition of our reduction approach.
The Roadmap of Reduction. Let and be an input of CM 1-in-3 3Sat with variables and clauses for all . To show the NP-completeness of -SSP, we reduce a given input to a TS (i.e., an input of -SSP) as follows: For every clause , the TS has a directed labeled path that represents by using its variables as events:
We ensure by construction that has an SSP atom such that if is a -region solving , then and for all . Thus, for all , the path is a path from to in . First, this obviously implies that there is an event such that . Second, it is easy to see that there is no path in on which inp occurs twice, cf. Figure 4. The following figure sketches all possibilities of , i.e., and , and , and and , respectively:
Hence, the event is unique. Since this is simultaneously true for all paths , the set selects exactly one variable per clause and thus defines a one-in-three model of . Altogether, this approach shows that if has the -SSP, which implies that is -solvable, then has a one-in-three model.
Conversely, our construction ensures that if has a one-in-three model, then and the other separation atoms of are -solvable, that is, has the -SSP.
The Reduction of for . In the following, we introduce the announced TS , cf. Figure 5. The initial state of is . First of all, the TS has the following path that provides the announced SSP atom :
Moreover, for every , the TS has the following path that uses the variables of as events and provides the subpath :
Finally, the TS has, for all , the following path :
Notice that the paths and have the same “final”-state . Obviously, the size of is polynomial in the size of . The following Lemma 3 and Lemma 4 prove the validity of our reduction and thus complete the proof of Theorem 3.1.
Lemma 3.
Let . If has the -SSP, then has a one-in-three model.
Proof.
Let be a -region that solves , that is, . By and , this implies , and . If is a path in , then, by , we get for all , for all and for all . This implies , as well as and for all . Furthermore, by and , we get for all . Hence, for all , the image of is a path from to in . Hence, as just discussed above, selects exactly one variable per clause and defines a one-in-three model of . ∎
Lemma 4.
Let . If has a one-in-three model, then has the -SSP.
Proof.
Let be a one-in-three model of .
The following region solves for all and all , where : ; for all , if , then , otherwise . Note Section 0.A, Figure 10 for an example of .
Let be arbitrary but fixed. The following region solves for all and all : ; for all , if , then , otherwise . Note Section 0.A, Figure 10 for an example of .
The following region solves and and for all : ; for all , if , then , otherwise . Note Section 0.A, Figure 12 for an example of .
The following region uses the one-in-three model of and solves as well as for all : ; for all , if , then , otherwise . Note Figure 5 for an example of .
Let be arbitrary but fixed. The following region solves for all and all , where : ; for all , if , then , otherwise . Note Section 0.A, Figure 12 for an example of .
By the arbitrariness of for and , it remains to show that are pairwise separable for all . Let be arbitrary but fixed. We present only a region that solves for all . It is then easy to see that the remaining atoms are similarly solvable. Let select the other two clauses of that contain , that is, . is defined as follows: ; for all , if , then , otherwise . Note Section 0.A, Figure 13 for an example of .
Similarly, one gets regions where or has inp-signature. These regions solve the remaining atoms of . Since was arbitrary, this completes the proof. ∎
3.2 Complexity of -SSP for nop-equipped Types with res or set.
The next theorem states that -SSP is NP-complete for the nop-equipped types that have not yet been considered, cf Figure 1 §1 - §3. Moreover, it summarizes the complexity of -SSP for all types of Figure 1 §1 - §4:
Theorem 3.2.
Let and be Boolean type of nets and and .
-
1.
The -SSP is NP-complete if , otherwise it is polynomial.
-
2.
The -SSP is NP-complete if , otherwise it is polynomial.
In this section we complete the proof of Theorem 3.2 as follows. Firstly, we let and, by a reduction of -SSP, we show that the -SSP is NP-complete if and or if . By Lemma 2, the former also implies the NP-completeness of -SSP if and . Altogether, this proves the claim for all the types listed in §1 and §3 of Figure 1.
Secondly, we let and reduce -SSP to -SSP, where and . Again by Lemma 2, this also implies the NP-completeness of -SSP if and . Hence, this proves the claim for all the types listed in §2 of Figure 1 and thus completes the proof of Theorem 3.2.
For the announced reductions, we use the following extensions of a TS , cf. Figure 6. Let be a loop-free TS, and let be the set containing for every event the unambiguous and fresh event that is associated with . The backward-extension of extends by and additional backward edges: for all and all , if , then and . The oneway loop-extension of a TS extends by some additional loops: for all and all , we define and, for all and all , if , then . Finally, the loop-extension of is an extension of , where for all and all , we define and, for all and all , if , then .
Depending on the considered type , we let or and reduce a loop-free TS either to its backward-, oneway loop- or loop-extension and show that allows a -region of if and only if it allows a -region of the extension:
Lemma 5.
Let , , a loop-free TS, and , and the backward-, oneway loop- and loop-extension of , respectively.
-
1.
If and , then allows a -region of if and only if it allows a normalized -region of .
-
2.
If , then allows a -region of if and only if it allows a normalized -region of .
-
3.
If and , then allows a -region of if and only if it allows a normalized -region of .
Proof.
(1): Only-if: Let be a -region of . Recall that , , imply , and , and for all edges of , respectively. Thus, it is easy to see that induces a normalized -region of as follows, cf. Figure 6: For all and its associated event , if , then ; if , then and ; if , then and .
If: Let be a normalized -region of , and let and and be arbitrary but fixed. First of all, we argue that if , then and : By definition of , we have that and are present. If and , then and imply and . By and , this immediately implies and . Similarly, if and , then and , which implies and . Consequently, since was arbitrary, the claim follows. Note that this implies in return that if , then . Since both edges were arbitrary, it is easy to see that the following -region of is well defined: for all , if there is such that , then if and , else ; otherwise, .
(2): Only-if: Recall that and are present in . Consequently, if is a -region of , then the region , similarly defined to the one for the Only-if-direction of (1), but replacing out by set, is a -region of , cf. Figure 6.
If: If is a normalized -region of , then it holds for all . This is due to the fact that if , then for all and all . Thus, is obtained from by the same arguments as the ones presented for the If-direction of (1).
(3): Only-if: Let be a -region of . Recall that and imply and for all edges of , respectively. Moreover, , and are present in . Thus, we get a normalized -region of as follows, cf. Figure 6: For all and its associated event , if , then ; if , then and .
If: Let be a normalized -region of , and let and and be arbitrary but fixed. We argue that implies and : By definition of and , we get . Thus, . Thus, if , then , which implies and . Moreover, by , this also implies . Finally, by , , and , we get and . Consequently, the following definition of yields a well-defined -region of : for all , if , then , otherwise . ∎
Notice that the TS of Section 3.1 is loop-free. Furthermore, in [23], it has been shown that -SSP is NP-complete even if is a simple directed path. Moreover, the introduced extensions of are constructible in polynomial time. Thus, by Lemma 1 and Lemma 2, the following corollary, which is easily implied by Lemma 5, completes the proof of Theorem 3.2.
Corollary 1 (Without Proof).
Let and , and let be a loop-free TS and , and its backward-, oneway loop- and loop-extension, respectively.
-
1.
If and , then has the -SSP if and only if has the -SSP.
-
2.
If , then has the -SSP if and only if has the -SSP.
-
3.
If and , then has the -SSP if and only if has the -SSP.
4 Deciding the State Separation Property for nop-free Types
The following theorem summarizes the complexity of -SSP for nop-free Boolean types:
Theorem 4.1.
Let be a nop-free type of nets and a TS.
-
1.
If or and , then deciding if has the -SSP is polynomial.
-
2.
If and , then deciding if has the -SSP is NP-complete.
The tractability of -SSP for nop-free types that are also swap-free has been shown in the broader context of -synthesis in [21]. Thus, restricted to Theorem 4.1.1, it remains to argue that -SSP is polynomial if and . The following lemma states that separable inputs of -SSP are trivial for these types and thus proves its tractability:
Lemma 6.
Let , where , and be a TS. If has the -SSP, then it has at most two states.
Proof.
If there is , then has no -regions. Hence, if such has more than one state, then it does not have the -SSP.
Let’s consider the case when is loop-free, that is, implies . First of all, note that there is at most one outgoing edge at , since if , and , then and are not separable. This can be seen as follows: If is a -region that solves , then and or and . If and , then contradicts , and contradicts . Similarly, and yields a contradiction. Thus, a separating region does not exist, which proves the claim. Secondly, if , then , since otherwise, and are not separable. This can be seen as follows: If is a -region that solves , then and or and . If and , then contradicts , and contradicts . Similarly, and yields a contradiction. This implies again that a separating region does not exist, which proves the claim and, moreover, proves the lemma. ∎
To complete the proof of Theorem 4.1, it remains to prove the NP-completeness of -SSP for the types listed in §9 of Figure 1, which are exactly covered by Theorem 4.1.2. Thus, in the remainder of this section, if not stated explicitly otherwise, we let be a nop-free type such that and . Moreover, we reduce CM 1-in-3 3Sat to -SSP again.
Basic Ideas of the Reduction. Similar to our previous approach we build a TS that has for every clause , where , a (bi-directed) path on which the elements of occur as events. Together, the corresponding paths are meant to represent . However, the current types are more diverse than , since they also allow swap and other interactions. Simultaneously, they are more restricted than , since they lack of nop. One of the main obstacles that occur is that if , then the current types basically allow and as well as and for a -region that solves . It turns out that this requires a second representation of . To do so, we use a copy that originates from by simply renaming its variables. That is, originates from by replacing every variable of by a unique and fresh variable .
Example 2 (Renaming of ).
The instance that originates from of Example 1 is defined by and , , , , and .
It is immediately clear that is one-in-three satisfiable if and only if is one-in-three satisfiable. The TS additionally has for every clause , where , of also a (bi-directed) path on which the elements of occur as events. Moreover, by the construction, the TS has a SSP atom such that if a -region solves , then either the signatures of the variable events of define a one-in-three model of or the signatures of the variable events of define a one-in-three model of . Obviously, both cases imply the one-in-three satisfiability of .
Conversely, the construction ensures, if has a one-in-three model then has the -SSP.
Similar to our approach for Theorem 3.1, the TS is a composition of several gadgets. The next Lemma 7 introduces some basic properties of -regions in bi-directed TS for nop-free types that we use to prove the functionailty of ’s gadgets. After that, Lemma 8 introduces TS that are the (isomorphic) prototypes of the gadgets of and additionally proves the essential parts of their intended functionality.
Lemma 7.
Let be a nop-free Boolean type of nets and a bi-directed TS; let be an edge of , and be two simple paths of that both apply the same sequence of events, and be a -region of .
-
1.
If , then for every edge .
-
2.
If , then for all .
-
3.
If , then is even.
Proof.
(1): is bi-directed and and imply for all .
(2): By definition of , if , then, by and , we get . Clearly, by , this implies . Thus, the claim follows easily by induction on .
(3): Since , the image of is a path of that starts and terminates at the same state. Consequently, the number of changes between and on is even. Since is bi-directed, if and only if for all . Thus, the number of events of with a swap-signature must be even. Hence, the claim. ∎
Lemma 8 (Basic Components of ).
Let be a nop-free Boolean type and a bi-directed TS with the following paths , , and , and let be a -region of :
-
1.
If solves , then either and or and ;
-
2.
If and or and , then ;
-
3.
If and for all , then there is exactly one such that .
Proof.
Since is bi-directed, we have for all .
(1): solves , thus . If or , then , a contradiction. Hence the claim, cf. Figure 8.
Let be an instance of CM 1-in-3 3Sat with the set of variables and its renamed copy with event set . In the following, we introduce the construction of .
Firstly, for every , the TS has the following gadgets , , and with starting states , , and , respectively, providing the atom :
Secondly, for every , the TS has the following gadgets and that use the elements of and as events, respectively; their starting states are , , and :
Finally, the gadgets are connected via their starting states to finally build as follows:
The following Lemma 9 and Lemma 10 prove the validity of our polynomial-time reduction and, hence, complete the proof of Theorem 4.1.
Lemma 9.
If has the -SSP, then is one-in-three satisfiable.
Proof.
Let , , and be the paths defined in Lemma 8. First of all, we observe that and and and for all and . Let be a -region that solves (which exists, since has the -SSP) and let be arbitrary but fixed. By Lemma 8.1, we have either and or and . This implies by Lemma 8.2. If and , then by Lemma 8.3, this implies that there is exactly one event such that . Consequently, since was arbitrary, if and , then selects exactly one variable of every clause for all . Thus, is a one in three-model of . Otherwise, if and , then we similarly obtain that defines a one-in-three-model of , which also implies the one-in-three satisfiability of . ∎
Lemma 10.
If is one-in-three satisfiable, then has the -SSP.
5 Conclusion
In this paper, we present the overall characterization of the computational complexity of the problem -SSP for all 256 Boolean types of nets . Our presentation includes 154 new complexity results (Figure 1: §1 - §3, §6, §9, §10) and 102 known results (Figure 1: §4 [22], §5 [19, 23], §7 [16, 22], §8 [21]) and classifies them in the overall context of boolean state separation. Besides the new 150 hardness- and 4 tractability-results, this classification is one of the main contributions of this paper. First of all, it becomes apparent that the distinction between nop-free and nop-equipped types is meaningful: Within the class of nop-free types, -SSP turns out to be NP-complete if and only if and . Within the class of nop-equipped types, a differentiation between types that satisfy and the ones with is useful: -SSP for the former ones is NP-complete if and only if and . In particular, -SSP becomes polynomial for all these types. On the other hand, for the latter ones, which include such that , -SSP is NP-complete as long as there is also an interaction in that opposes , that is, if and if .
Moreover, our proofs discover that, up to isomorphism, there are essentially four hard kernels that indicate the NP-completeness of -SSP, namely and for the nop-equipped types and and for the nop-free types. That means, for a nop-equipped type , the hardness of -SSP can either be shown by a reduction of -SSP or -SSP (or their isomorphic types), which is basically done in the proof of Lemma 5, or -SSP is polynomial othwerwise. Similarly, one finds out that for the nop-free types in question, the hardness of -SSP can be shown by a reduction of -SSP or -SSP. Due to the space limitation, a reduction that covers all hard nop-free types on one blow is given instead of explicit proof.
For future work, it remains to completely characterize the computational complexity of deciding if a TS is isomorphic to the reachability graph of a Boolean Petri net, instead of only being embeddable by an injective simulation map.
References
- [1] Badouel, E., Bernardinello, L., Darondeau, P.: Polynomial algorithms for the synthesis of bounded nets. In: TAPSOFT. Lecture Notes in Computer Science, vol. 915, pp. 364–378. Springer (1995). https://doi.org/10.1007/3-540-59293-8_207
- [2] Badouel, E., Bernardinello, L., Darondeau, P.: The synthesis problem for elementary net systems is NP-complete. Theor. Comput. Sci. 186(1-2), 107–134 (1997). https://doi.org/10.1016/S0304-3975(96)00219-8
- [3] Badouel, E., Bernardinello, L., Darondeau, P.: Petri Net Synthesis. Texts in Theoretical Computer Science. An EATCS Series, Springer (2015). https://doi.org/10.1007/978-3-662-47967-4
- [4] Badouel, E., Darondeau, P.: Trace nets and process automata. Acta Inf. 32(7), 647–679 (1995). https://doi.org/10.1007/BF01186645
- [5] Best, E., Devillers, R.R.: Pre-synthesis of Petri nets based on prime cycles and distance paths. Sci. Comput. Program. 157, 41–55 (2018). https://doi.org/10.1016/j.scico.2017.07.005, https://doi.org/10.1016/j.scico.2017.07.005
- [6] Chatain, T., Haar, S., Kolcák, J., Paulevé, L., Thakkar, A.: Concurrency in boolean networks. Nat. Comput. 19(1), 91–109 (2020). https://doi.org/10.1007/s11047-019-09748-4, https://doi.org/10.1007/s11047-019-09748-4
- [7] Esparza, J., Nielsen, M.: Decidability issues for Petri nets - a survey. Bulletin of the EATCS 52, 244–262 (1994)
- [8] Kleijn, J., Koutny, M., Pietkiewicz-Koutny, M., Rozenberg, G.: Step semantics of boolean nets. Acta Inf. 50(1), 15–39 (2013). https://doi.org/10.1007/s00236-012-0170-2
- [9] Montanari, U., Rossi, F.: Contextual occurrence nets and concurrent constraint programming. In: Schneider, H.J., Ehrig, H. (eds.) Graph Transformations in Computer Science. pp. 280–295. Springer Berlin Heidelberg, Berlin, Heidelberg (1994)
- [10] Montanari, U., Rossi, F.: Contextual nets. Acta Inf. 32(6), 545–596 (1995). https://doi.org/10.1007/BF01178907
- [11] Moore, C., Robson, J.M.: Hard tiling problems with simple tiles. Discrete & Computational Geometry 26(4), 573–590 (2001). https://doi.org/10.1007/s00454-001-0047-6
- [12] Pietkiewicz-Koutny, M.: Transition systems of elementary net systems with inhibitor arcs. In: ICATPN. Lecture Notes in Computer Science, vol. 1248, pp. 310–327. Springer (1997). https://doi.org/10.1007/3-540-63139-9_43
- [13] Rozenberg, G., Engelfriet, J.: Elementary net systems. In: Petri Nets. Lecture Notes in Computer Science, vol. 1491, pp. 12–121. Springer (1996). https://doi.org/10.1007/3-540-65306-6_14
- [14] Schlachter, U.: Over-approximative Petri net synthesis for restricted subclasses of nets. In: Klein, S.T., Martín-Vide, C., Shapira, D. (eds.) Language and Automata Theory and Applications - 12th International Conference, LATA 2018, Ramat Gan, Israel, April 9-11, 2018, Proceedings. Lecture Notes in Computer Science, vol. 10792, pp. 296–307. Springer (2018). https://doi.org/10.1007/978-3-319-77313-1_23, https://doi.org/10.1007/978-3-319-77313-1_23
- [15] Schlachter, U., Wimmel, H.: Optimal label splitting for embedding an LTS into an arbitrary Petri net reachability graph is NP-complete. CoRR abs/2002.04841 (2020), https://arxiv.org/abs/2002.04841
- [16] Schmitt, V.: Flip-flop nets. In: STACS. Lecture Notes in Computer Science, vol. 1046, pp. 517–528. Springer (1996). https://doi.org/10.1007/3-540-60922-9_42
- [17] Tredup, R.: The complexity of synthesizing nop-equipped boolean nets from g-bounded inputs (technical report) (2019)
- [18] Tredup, R.: Fixed parameter tractability and polynomial time results for the synthesis of b-bounded petri nets. In: Donatelli, S., Haar, S. (eds.) Application and Theory of Petri Nets and Concurrency - 40th International Conference, PETRI NETS 2019, Aachen, Germany, June 23-28, 2019, Proceedings. Lecture Notes in Computer Science, vol. 11522, pp. 148–168. Springer (2019). https://doi.org/10.1007/978-3-030-21571-2_10, https://doi.org/10.1007/978-3-030-21571-2_10
- [19] Tredup, R.: Hardness results for the synthesis of b-bounded Petri nets. In: Donatelli, S., Haar, S. (eds.) Application and Theory of Petri Nets and Concurrency - 40th International Conference, PETRI NETS 2019, Aachen, Germany, June 23-28, 2019, Proceedings. Lecture Notes in Computer Science, vol. 11522, pp. 127–147. Springer (2019). https://doi.org/10.1007/978-3-030-21571-2_9, https://doi.org/10.1007/978-3-030-21571-2_9
- [20] Tredup, R.: Parameterized complexity of synthesizing b-bounded (m, n)-t-systems. In: Chatzigeorgiou, A., Dondi, R., Herodotou, H., Kapoutsis, C.A., Manolopoulos, Y., Papadopoulos, G.A., Sikora, F. (eds.) SOFSEM 2020: Theory and Practice of Computer Science - 46th International Conference on Current Trends in Theory and Practice of Informatics, SOFSEM 2020, Limassol, Cyprus, January 20-24, 2020, Proceedings. Lecture Notes in Computer Science, vol. 12011, pp. 223–235. Springer (2020). https://doi.org/10.1007/978-3-030-38919-2_19, https://doi.org/10.1007/978-3-030-38919-2_19
- [21] Tredup, R., Erofeev, E.: On the complexity of synthesis of nop-free boolean petri nets. In: van der Aalst, W.M.P., Bergenthum, R., Carmona, J. (eds.) Proceedings of the International Workshop on Algorithms & Theories for the Analysis of Event Data 2020 Satellite event of the 41st International Conference on Application and Theory of Petri Nets and Concurrency Petri Nets 2020, virtual workshop, June 24, 2020. CEUR Workshop Proceedings, vol. 2625, pp. 66–84. CEUR-WS.org (2020), http://ceur-ws.org/Vol-2625/paper-05.pdf
- [22] Tredup, R., Rosenke, C.: The complexity of synthesis for 43 boolean Petri net types. In: TAMC. Lecture Notes in Computer Science, vol. 11436, pp. 615–634. Springer (2019). https://doi.org/10.1007/978-3-030-14812-6
- [23] Tredup, R., Rosenke, C., Wolf, K.: Elementary net synthesis remains NP-complete even for extremely simple inputs. In: Petri Nets. Lecture Notes in Computer Science, vol. 10877, pp. 40–59. Springer (2018). https://doi.org/10.1007/978-3-319-91268-4_3
- [24] Vogler, W., Semenov, A., Yakovlev, A.: Unfolding and finite prefix for nets with read arcs. In: Sangiorgi, D., de Simone, R. (eds.) CONCUR’98 Concurrency Theory. pp. 501–516. Springer Berlin Heidelberg, Berlin, Heidelberg (1998)
- [25] Wimmel, H.: Presynthesis of bounded choice-free or fork-attribution nets. Inf. Comput. 271, 104482 (2020). https://doi.org/10.1016/j.ic.2019.104482, https://doi.org/10.1016/j.ic.2019.104482
- [26] Yakovlev, A., Koelmans, A., Semenov, A.L., Kinniment, D.J.: Modelling, analysis and synthesis of asynchronous control circuits using Petri nets. Integr. 21(3), 143–170 (1996). https://doi.org/10.1016/S0167-9260(96)00010-7, https://doi.org/10.1016/S0167-9260(96)00010-7
Appendix 0.A Supporting Illustrations for the Proof of Lemma 4. The depicted TS originates from of Example 1.
Appendix 0.B The Proof of Lemma 10.
The Proof of Lemma 10..
Let be a one-in-three model of . In what follows, we assume and argue that has the -SSP. The assumption is possible without loss of generality: In particular, every presented region is a -region. Hence, every presented support allows a -region by simply replacing by . Thus, by Lemma 2 and and , the -SSP of implies the its -SSP also for the types in question where , which implies .
For abbreviation, let , , , , , , , and , and let , , , and be defined accordingly.
First of all, it is easy to see that, for all , the SSP atom is -solvable for all . Secondly, if is an arbitrary but fixed gadget of and and , then the atom is -solvable. This can be seen as follows: Let be arbitrary but fixed. (This event connects a certain gadget, say , with the others.) On the one hand, there is a region , such that and, for all , if , then , otherwise .
On the other hand, there is a region , such that and, for all , if and , then , otherwise . Notice that for the unique gadget of whose starting state is adjacent to holds, for all , that if , otherwise . Thus, if and , then is either solved by or by .
Justified by these observations, it only remains to consider separation atoms where and belong to the same gadget of .
Let and be arbitrary but fixed and let’s argue for the solvability of for all . The following -region solves for all : ; for all , if , then , otherwise .
The following -region solves for all : ; for all , if , then , otherwise .
The following -region solves for all : ; for all , if , then , otherwise .
Restricted to , it remains to solve the atom . The following region uses the existence of the one-in-three model and solves this atom: ; for all , if , then , otherwise .
Note that the regions , also show that if , then is -solvable. Altogether, by the arbitrariness of and and the symmetry of a gadget and its counterpart , this shows, for all , that if , then is -solvable.
Next, we let be arbitrary but fixed and argue that is -solvable if . The following -region solves for all relevant : ; for all , if , then , otherwise . Similarly, one proves that is -solvable for all relevant .
Let’s consider the -solvability of for all relevant states . In the following, we argue that there is a -region that, for all , maps to swap if and to res otherwise, such that and for all . Since , we have to deal with at most five further occurrences of in : the one in and at most four further occurrences in say and for some . Thus, to define completely, we would have to do lots of tedious and mostly meaningless case analyses. To avoid the latter, in the following, we argue informally for the existence of : The essential observation is that, by construction, if , then there is a state and an event such that . Notice that there might be events such that and as to be seen for and . Anyway, to obtain , we let and select exactly one arbitrary but fixed v-event and its corresponding w-event for every and map it to swap just like and . The other events get a free-signature. One finds out that the resulting region is well-defined and behaves as announced. Similarly, one proves the -solvability for all .
Restricted to , it remains to argue that and are pairwise separable. Fortunately, the corresponding SSP atoms are already solved by regions defined above. More exactly, the region solves , , and ; moreover, the region solves and . Hence, is -solvable for all .
Finally, it is easy to see that, on the one hand, the SSP atoms corresponding to are similarly solvable as the ones of and, on the other hand, the SSP atoms of and are similarly solvable by symmetry. Hence, since was arbitrary, this proves that if , then is -solvable for all . ∎