WQO Dichotomy for 3Graphs
Abstract
We investigate dataenriched models, like Petri nets with data, where executability of a transition is conditioned by a relation between data values involved. Decidability status of various decision problems in such models may depend on the structure of data domain. According to the WQO Dichotomy Conjecture, if a data domain is homogeneous then it either exhibits a well quasiorder (in which case decidability follows by standard arguments), or essentially all the decision problems are undecidable for Petri nets over that data domain.
We confirm the conjecture for data domains being 3graphs (graphs with 2colored edges). On the technical level, this results is a significant step beyond known classification results for homogeneous structures.
1 Introduction
In Petri nets with data, tokens carry values from some data domain, and executability of transitions is conditioned by a relation between data values involved. One can consider unordered data, like in [25], i.e., an infinite data domain with the equality as the only relation; or ordered data, like in [21], i.e., an infinite densely totally ordered data domain; or timed data, like in timed Petri nets [1] and timedarc Petri nets [15]. In [19] an abstract setting of Petri nets with an arbitrary fixed data domain \({\mathbb {A}}\) has been introduced, parametric in a relational structure \({\mathbb {A}}\). The setting uniformly subsumes unordered, ordered and timed data (represented by \({\mathbb {A}}= ({\mathbb {N}}, =)\), \({\mathbb {A}}= ({\mathbb {Q}}, \le )\) and \({\mathbb {A}}= ({\mathbb {Q}}, \le , +1)\), respectively).
Following [19], in order to enable finite presentation of Petri nets with data, and in particular to consider such models as input to algorithms, we restrict to relational structures \({\mathbb {A}}\) that are homogeneous [23] and effective (the formal definitions are given in Sect. 2). Certain standard decision problems (like the termination problem, the boundedness problem, or the coverability problem, jointly called from now on standard problems) are all decidable for Petri nets with ordered data [21] (and in consequence also for Petri nets with unordered data), as the model fits into the framework of wellstructured transition systems of [11]. Most importantly, the structure \({\mathbb {A}}= ({\mathbb {Q}}, \le )\) of ordered data admits well quasiorder (wqo) in the following sense: for any wqo X, the set of finite induced substructures of \(({\mathbb {Q}}, \le )\) (i.e., finite total orders) labeled by elements of X, ordered naturally by embedding, is a wqo (this is exactly Higman’s lemma). Moreover, essentially the same argument can be used for any other homogeneous effective data domain which admits wqo (see [19] for details). On the other hand, for certain homogeneous effective data domains \({\mathbb {A}}\) the standard problems become all undecidable. In the quest for understanding the decidability borderline, the following hypothesis has been formulated in [19]:
Conjecture 1
(Wqo Dichotomy Coinjecture [19]). For an effective homogeneous structure \({\mathbb {A}}\), either \({\mathbb {A}}\) admits wqo (in which case the standard problems are decidable for Petri nets with data \({\mathbb {A}}\)), or all the standard problems are undecidable for Petri nets with data \({\mathbb {A}}\).
According to [19], the conjecture could have been equivalently stated for another dataenriched models, e.g., for finite automata with one register [2]. In this paper we consider, for the sake of presentation, only Petri nets with data. Wqo Dichotomy Conjecture holds in special cases when data domains \({\mathbb {A}}\) are undirected or directed graphs, due to the known classifications of homogeneous graphs [6, 18].
Contributions. We confirm the Wqo Dichotomy Conjecture for data domains \({\mathbb {A}}\) being strongly^{1} homogeneous 3graphs. A 3graph is a logical structure with three irreflexive symmetric binary relations such that every pair of elements of \({\mathbb {A}}\) belongs to exactly one of the relations (essentially, a clique with 3colored edges).
Our main technical contribution is a complex analysis of possible shapes of strongly homogeneous 3graphs, constituting the heart of the proof. We believe that this is a significant step towards full classification of homogeneous 3graphs. The classification of homogeneous structures is a wellknown challenge in model theory, and has been only solved in some cases by now: for undirected graphs [18], directed graphs (the proof of Cherlin spans a book [6]), multipartite graphs [16], and few others (the survey [23] is an excellent overview of homogeneous structures). Although the full classification of homogeneous 3graphs was not our primary objective, we believe that our analysis significantly improves our understanding of these structures and can be helpful for classification.
Our result does not fully settle the status of the Wqo Dichotomy Conjecture. Dropping the (mild) strong homogeneity assumption, as well as extending the proof to arbitrarily many symmetric binary relations, is left for future work.
Related Research. Net models similar to Petri nets with data have been continuously proposed since the 80s, including, among the others, highlevel Petri nets [13], colored Petri nets [17], unordered and ordered data nets [21], \(\nu \)Petri nets [25], and constraint multiset rewriting [5, 8, 9]. Petri nets with data can be also considered as a reinterpretation of the classical definition of Petri nets in sets with atoms [3, 4], where one allows for orbitfinite sets of places and transitions instead of just finite ones. The decidability and complexity of standard problems for Petri nets over various data domains has attracted a lot of attention recently, see for instance [14, 21, 22, 24, 25].
Wqos are important for their wide applicability in many areas. Studies of wqos similar to ours, in case of graphs, have been conducted by Ding [10] and Cherlin [7]; their framework is different though, as they concentrate on subgraph ordering while we investigate induced subgraph (or substructure) ordering.
2 Petri Nets with Homogeneous Data

Equality data domain: natural numbers with equality \({\mathbb {A}}_{=}= ({\mathbb {N}}, =)\). Note that any other countably infinite set could be used instead of natural numbers, as the only available relation is equality.

Total order data domain: rational numbers with the standard order \({\mathbb {A}}_{\le }= ({\mathbb {Q}}, \le )\). Again, any other countably infinite dense total order without extremal elements could be used instead.

Nested equality data domain: \({\mathbb {A}}_{1}= ({\mathbb {N}}^2, =_1, =)\) where \(=_1\) is equality on the first component: \((n, m) =_1 (n', m')\) if \(n = n'\) and \(m\ne m'\). Essentially, \({\mathbb {A}}\) is an equivalence relation with infinitely many infinite equivalence classes.
Note that two latter structures essentially extend the first one: in each case the equality is either present explicitly, or is definable. From now on, we always assume a fixed countably infinite relational structure \({\mathbb {A}}\) with equality over a finite vocabulary (signature) \(\varSigma \).

arcs are labelled by pairwise disjoint finite nonempty sets of variables;

transitions are labelled by firstorder formulas over the vocabulary \(\varSigma \) of \({\mathbb {A}}\), such that free variables of the formula labeling a transition t belong to the union of labels of the arcs incident to t.
Example 1
For illustration consider a Petri net with equality data \({\mathbb {A}}_{=}\), with two places \(p_1, p_2\) and two transitions \(t_1, t_2\) depicted on Fig. 1. Transition \(t_1\) outputs two tokens with arbitrary but distinct data values onto place \(p_1\). Transition \(t_2\) inputs two tokens with the same data value, say a, one from \(p_1\) and one from \(p_2\), and outputs 3 tokens: two tokens with arbitrary but equal data values, say b, one onto \(p_1\) and the other onto \(p_2\); and one token with a data value \(c \ne a\) onto \(p_2\). Note that the transition \(t_2\) does not specify whether \(b=a\), or \(b = c\), or \(b \ne a,c\), and therefore all three options are allowed. Variables \(y_1, y_2\) can be considered as input variables of \(t_2\), while variables \(z_1, z_2, z_3\) can be considered as output ones; analogously, \(t_1\) has no input variables, and two output ones \(x_1, x_2\).
As usual, for a net N and its configuration C, a run of (N, C) is a maximal, finite or infinite, sequence of steps starting in C.
Remark 1
As for classical Petri nets, an essentially equivalent definition can be given in terms of vector addition systems (such a variant has been used in [14] for equality data). Petri nets with equality data are equivalent to (even if defined differently than) unordered data Petri nets of [21], and Petri nets with total ordered data are equivalent to ordered data Petri nets of [21].
Effective Homogeneous Structures. For two relational \(\varSigma \)structures \(\mathcal{{A}}\) and \(\mathcal{{B}}\) we say that \(\mathcal{{A}}\) embeds in \(\mathcal{{B}}\), written \(\mathcal{{A}} \unlhd \mathcal{{B}}\), if \(\mathcal{{A}}\) is isomorphic to an induced substructure of \(\mathcal{{B}}\), i.e., to a structure obtained by restricting \(\mathcal{{B}}\) to a subset of its domain. This is witnessed by an injective function^{2} \(h : \mathcal{{A}} \rightarrow \mathcal{{B}}\), which we call embedding. We write \({\textsc {Age}}({\mathbb {A}}) = \left\{ \, \mathcal{{A}} \text { a finite structure} \,  \, \mathcal{{A}} \unlhd {\mathbb {A}}\,\right\} \) for the class of all finite structures that embed into \({\mathbb {A}}\), and call it the age of\({\mathbb {A}}\).
Homogeneous structures are defined through their automorphisms: \({\mathbb {A}}\) is homogeneous if every isomorphism of two its finite induced substructures extends to an automorphism of \({\mathbb {A}}\). In the sequel we will also need an equivalent definition using amalgamation. An amalgamation instance consists of three structures \(\mathcal{{A}}, \mathcal{{B}}_1, \mathcal{{B}}_2 \in {\textsc {Age}}({\mathbb {A}})\) and two embeddings \(h_1 : \mathcal{{A}}\rightarrow \mathcal{{B}}_1\) and \(h_2 : \mathcal{{A}} \rightarrow \mathcal{{B}}_2\). A solution of such instance is a structure \(\mathcal{{C}} \in {\textsc {Age}}({\mathbb {A}})\) and two embeddings \(g_1 : \mathcal{{B}}_1 \rightarrow \mathcal{{C}}\) and \(g_2 : \mathcal{{B}}_2 \rightarrow \mathcal{{C}}\) such that \(g_1 \circ h_1 = g_2 \circ h_2\) (we refer the reader to [12] for further details). Intuitively, \(\mathcal{{C}}\) represents ‘gluing’ of \(\mathcal{{B}}_1\) and \(\mathcal{{B}}_2\) along the partial bijection \(h_2 \circ ({h_1}^{1})\). In this paper we will restrict ourselves to singleton amalgamation instances, where only one element of \(\mathcal{{B}}_1\) is outside of \(h_1(\mathcal{{A}})\), and likewise for \(\mathcal{{B}}_2\).
An example singleton amalgamation instance is shown on the right, where the graph \(\mathcal{{A}}\) consists of the single edge connecting two middle black nodes, \(\mathcal{{B}}_1\) is the left triangle, and \(\mathcal{{B}}_2\) the right one. The dashed line represents an edge that may (but does not have to) appear in a solution. \({\mathbb {A}}\) is homogeneous if, and only if every amalgamation instance has a solution; in such case we say that \({\textsc {Age}}({\mathbb {A}})\) has the amalgamation property. See [23] for further details.
A solution \(\mathcal{{C}}\) necessarily satisfies \(g_1(h_1(\mathcal{{A}})) = g_2(h_2(\mathcal{{A}})) \subseteq g_1(\mathcal{{B}}_1) \cap g_2(\mathcal{{B}}_2)\); a solution is strong if \(g_1(h_1(\mathcal{{A}})) = g_1(\mathcal{{B}}_1) \cap g_2(\mathcal{{B}}_2)\). Intuitively, this forbids additional gluing of \(\mathcal{{B}}_1\) and \(\mathcal{{B}}_2\) not specified by the partial bijection \(h_2 \circ ({h_1}^{1})\). If every amalgamation instance has a strong solution we call \({\mathbb {A}}\) strongly homogeneous. This is a mild restriction, as homogeneous structures are typically strongly homogeneous.
The equality, nested equality, and total order data domains are strongly homogeneous structures. For instance, in the latter case finite induced substructures are just finite total orders, which satisfy the strong amalgamation property. Many other natural classes of structures have the amalgamation property: finite graphs, finite directed graphs, finite partial orders, finite tournaments, etc. Each of these classes is the age of a strongly homogeneous relational structure, namely the universal graph (called also random graph), the universal directed graph, the universal partial order, the universal tournament, respectively. Examples of homogeneous structures abound [23].
Homogeneous structures admit quantifier elimination: every firstorder formula is equivalent to (i.e., defines the same set as) a quantifierfree one [23]. Thus it is safe to assume that formulas labeling transitions are quantifierfree.
Admitting wqo . A well quasiorder (wqo) is a wellfounded quasiorder with no infinite antichains. For instance, finite multisets \(\mathcal{M}(P)\) over a finite set P, ordered by multiset inclusion \(\sqsubseteq \), are a wqo. Another example is the embedding quasiorder \(\unlhd \) in \({\textsc {Age}}({\mathbb {A}}_{\le })\) (= all finite total orders) isomorphic to the ordering of natural numbers. Finally, the embedding quasiorder in \({\textsc {Age}}({\mathbb {A}})\) can be lifted from finite structures to finite structures labeled by elements of some ordered set \((X, \le )\): for two such labeled structures \(a : \mathcal{{A}} \rightarrow X\) and \(b: \mathcal{{B}} \rightarrow X\) we define \( a \unlhd _{X} b \) if some embedding \(h : \mathcal{{A}} \rightarrow \mathcal{{B}}\) satisfies \(a(x) \le b(h(x))\) for every \(x\in \mathcal{{A}}\). We say that \({\mathbb {A}}\) admits wqo when for every wqo \((X, \le )\), the lifted embedding order \(\unlhd _{X}\) is a wqo too. For instance, \({\mathbb {A}}_{\le }\) admits wqo by Higman’s lemma. The Wqo Dichotomy Conjecture for homogeneous undirected (and also directed) graphs is easily shown by inspection of the classifications thereof [6, 18]:
Theorem 1
A homogeneous graph \({\mathbb {A}}\) either admits wqo, or all standard problems are undecidable for Petri nets with data \({\mathbb {A}}\).
Standard Decision Problems. A Petri net with data N can be finitely represented by finite sets P, T, A and appropriate labelings with variables and formulas. Due to the homogeneity of \({\mathbb {A}}\), a configuration C can be represented (up to automorphism of \({\mathbb {A}}\)) by a structure \(\mathcal{{A}} \in {\textsc {Age}}(A)\) labeled by \(\mathcal{M}(P)\). We can thus consider the classical decision problems that input Petri nets with data \({\mathbb {A}}\), like the termination problem: does a given (N, C) have only finite runs? The data domain is considered as a parameter, and hence itself does not constitute part of input. Another classical problem is the place nonemptiness problem (markability): given (N, C) and a place p of N, does (N, C) admit a run that puts at least one token on place p? One can also define the appropriate variants of the coverability problem (equivalent to the place nonemptiness problem), the boundedness problem, the evitability problem, etc. (see [19] for details). All the decision problems mentioned above we jointly call standard problems.
A \(\varSigma \)structure \({\mathbb {A}}\) is called effective if the following age problem for \({\mathbb {A}}\) is decidable: given a finite \(\varSigma \)structure \(\mathcal{{A}}\), decide whether \(\mathcal{{A}} \unlhd {\mathbb {A}}\). If \({\mathbb {A}}\) admits wqo then application of the framework of wellstructured transition systems [11] to the lifted embedding order \(\unlhd _{\mathcal{M}(P)}\) yields:
Theorem 2
([19]). If an effective homogeneous structure \({\mathbb {A}}\) admits wqo then all the standard problems are decidable for Petri nets with data \({\mathbb {A}}\).
3 Results
A 3graph \({\mathbb {G}}= (V, C_1, C_2, C_3)\) consists of a set V and three irreflexive symmetric binary relations \(C_1, C_2, C_3 \subseteq V^2\) such that every pair of distinct elements of V belongs to exactly one of the three relations. In the sequel we treat a 3graph as a clique with 3colored edges. Any graph, including \({\mathbb {A}}_{=}\) and \({\mathbb {A}}_{1}\), can be seen as a 3graph. Our main result confirms the Wqo Dichotomy Conjecture for strongly homogeneous 3graphs:
Theorem 3
An effective strongly homogeneous 3graph \({\mathbb {G}}\) either admits wqo, or all standard problems are undecidable for Petri nets with data \({\mathbb {G}}\).
The core technical result of the paper is Theorem 4 below. A path is a finite graph with nodes \(\{v_1, \ldots , v_n\}\) whose only edges are pairs \(\{v_i, v_{i+1}\}\). The nodes \(v_1, v_n\) are ends of the path, and n is its length.
Theorem 4
A strongly homogeneous 3graph \({\mathbb {G}}\) either admits wqo, or for some \(i, j \in \{1, 2, 3\}\) (not necessarily distinct) the graph \((V, C_i \cup C_j)\) contains arbitrarily long paths as induced subgraphs.
In the rest of the paper we concentrate solely on (parts of) the proof of Theorem 4. The omitted parts, and well as the proof that Theorem 4 implies Theorem 3, are to be found in the full version of this paper [20].
Example 2
For a quasiorder \((X, \le )\), the multiset inclusion is defined as follows for \(m, m' \in \mathcal{M}(X)\): \(m'\) is included in m if \(m'\) is obtained from m by a sequence of operations, where each operation either removes some element, or replaces some element by a smaller one wrt. \(\le \). The structure \({\mathbb {A}}_{=}= ({\mathbb {N}}, =)\) admits wqo. Indeed, \({\textsc {Age}}({\mathbb {A}}_{=})\) contains just finite pure sets, thus \(\unlhd _{X}\) is quasiorderisomorphic to the multiset inclusion on \(\mathcal{M}(X)\), and is therefore a wqo whenever the underlying quasiorder \((X, \le )\) is. Similarly, \({\mathbb {A}}_{1}= ({\mathbb {N}}^2, =_1, =)\) also admits wqo, as \(\unlhd _{X}\) is quasiorderisomorphic to the multiset inclusion on \(\mathcal{M}(\mathcal{M}(X))\).
On the other hand, consider a 3graph \(({\mathbb {N}}^2, =_1, =_2,\) \(\ne _{12})\) where \(=_2\) is symmetric to \(=_1\) and \((n, m) \ne _{12} (n', m')\) if \(n \ne n'\) and \(m \ne m'\). It refines \({\mathbb {A}}_{1}\) and does not admit wqo. Indeed, in agreement with Theorem 4, the graph \(({\mathbb {N}}^2, =_1 \cup =_2)\) contains arbitrarily long paths of the shape presented on the right, where the two colors depict \(=_1\) and \(=_2\), respectively, and lack of color corresponds to \(\ne _{12}\). Note that \(({\mathbb {N}}^2, =_1, =_2, \ne _{12})\) is homogeneous but not strongly so.
4 Proof of Theorem 4
From now on we consider a fixed 3graph \({\mathbb {G}}= (V, C_1, C_2, C_3)\) as data domain, assuming \({\mathbb {G}}\) to be countably infinite and strongly homogeneous. We treat \({\mathbb {G}}\) as a clique with 3colored edges: we call \(C_1, C_2\) and \(C_3\) colors and put Open image in new window . To denote individual colors from this set, we will use variables Open image in new window and Open image in new window . A path in the graph Open image in new window we call Open image in new window ( Open image in new window ); for simplicity, we will write Open image in new window instead of Open image in new window path. Likewise we speak of Open image in new window cliques, Open image in new window cliques, Open image in new window cycles, etc. A triangle Open image in new window is a 3clique with edges colored by Open image in new window . (Note that Open image in new window ).
Sketch of the Proof. The Lemma 1 below states that any 3graph \({\mathbb {G}}\) has to meet one of the four listed cases. It splits the proof into four separate paths:
We present in detail only one of the three nontrivial paths – one corresponding to case (C). Cases (A) and (B) are treated in the full version [20]. Case (A) constitutes the most difficult part of the proof and involves a complex and delicate analysis of consequences of the amalgamation property. It consists of four step that deduce extension of the assumed induced substructures by individual vertices, individual edges, paths of length 2, resp., culminating in derivation of arbitrarily long paths. Thus in case (A) only the second condition of Theorem 4 is possible, while in the other two cases both conditions of Theorem 4 may hold true.
Lemma 1
 (A)for some color Open image in new window , \({\mathbb {G}}\) contains the following induced substructures:
 (B)
for some colors Open image in new window , Open image in new window is a union of disjoint cliques,
 (C)
for some color Open image in new window , Open image in new window is a union of finitely many disjoint infinite cliques,
 (D)
for some colors Open image in new window , Open image in new window contains arbitrarily long paths.
Proof
By Ramsey theorem, \({\mathbb {G}}\) contains an arbitrarily large monochromatic cliques. Let us state a bit stronger requirement:
Condition \(\spadesuit \): For some Open image in new window , \({\mathbb {G}}\) contains arbitrarily large Open image in new window cliques and a triangle Open image in new window with exactly two Open image in new window edges ( Open image in new window ).
Consider two cases, depending on whether the condition \(\spadesuit \) is satisfied or not.
Case \(1^\circ \). Assume that \({\mathbb {G}}\) contains both arbitrarily large Open image in new window cliques and a triangle Open image in new window for some Open image in new window . Let Open image in new window be the third, remaining color. Our goal will be to show that either (A) or (B) holds.
If the graph Open image in new window is a disjoint sum of cliques, we immediately obtain (B). Suppose the contrary. We get that \({\mathbb {G}}\) has to contain one of the three possible counterexamples for transitivity of relation Open image in new window :
If it contains the triangle Open image in new window or Open image in new window , case (A) holds.
Suppose we got Open image in new window . Let us check this time whether colors Open image in new window and Open image in new window form a union of disjoint cliques. Again, if it is so, we easily get (B), so we assume the contrary. Similarly, we necessarily obtain one of the following triangles:

for Open image in new window , because together with subgraphs resulting from assumption \(\spadesuit \) (i.e. with triangle Open image in new window and the Open image in new window cliques) we get all graphs required by (A).

for Open image in new window paired with the triangle Open image in new window we just obtained, using color Open image in new window appearing in those triangles in place of Open image in new window in condition (A).
It only remains to consider the situation when we got Open image in new window . We use it together with previously obtained triangle Open image in new window to build the following instance of singleton amalgamation:
Depending on the color of the dashed edge, in the solution we get one of the following triangles:
and each one alone completes the requirements of (A). This closes case \(1^\circ \).
Case \(2^\circ \). Suppose condition \(\spadesuit \) is false. Remind that \({\mathbb {G}}\) contains arbitrarily large Open image in new window cliques for some Open image in new window . Since \(\spadesuit \) does not hold, the graph does not contain a triangle Open image in new window – in other words, the color Open image in new window appears only within cliques. We conclude that Open image in new window is a union of disjoint cliques. Clearly at least one of such cliques has to be infinite. By homogeneity we get that all the cliques in Open image in new window have to be infinite. Now our target is to show that either (C) or (D) holds.
The case (C) is fulfilled when there are only finitely many Open image in new window cliques. Let us assume the contrary. In each of the Open image in new window cliques we chose one vertex. Edges between the chosen vertices form an infinite Open image in new window clique K. Using Ramsey theorem again, we conclude that in K one of the colors Open image in new window forms arbitrarily large monochromatic cliques. W.l.o.g. suppose that this is color Open image in new window .
If the graph \({\mathbb {G}}\) contained Open image in new window for some Open image in new window , then the assumptions of \(\spadesuit \) would be met, leading to a contradiction. Therefore we conclude that Open image in new window is a union of disjoint infinite Open image in new window cliques.
When there are only finitely many Open image in new window cliques, condition (C) is fulfilled. Otherwise we know that \({\mathbb {G}}\) is a union of infinitely many Open image in new window cliques for both Open image in new window and Open image in new window . Using homogenity, it is easy to show that then every pair of differently colored cliques has exactly one common vertex, so the graph \({\mathbb {G}}\) takes the form as depicted in Example 2. A graph of such form contains arbitrarily long Open image in new window path, so the requirements of (D) are met. \(\square \)
4.1 Case (C)
Let Open image in new window be the color that satisfies condition (C), and Open image in new window , Open image in new window — the remaining two colors. In this section we often treat \({\mathbb {G}}\) as the kpartite graph Open image in new window (for some \(k\in \mathbb {N}\)): k cliques of color Open image in new window allow to distinguish k groups of vertices \(V_1 \cup V_2 \cup \dots \cup V_k = V\) (from now on we will refer to them as layers). The remaining two colors can be interpreted as existence ( Open image in new window ) and nonexistence ( Open image in new window ) of edges between these groups.
Remark \(\bigstar \): We observe that the special color Open image in new window between vertices within each layer \(V_i\) ensures that the automorphisms of \({\mathbb {G}}\) will not ‘mix’ those layers: when two vertices u, v belong to a common layer \(V_i\), then their images f(u), f(v) will also belong to some common layer \(V_j\), no matter what automorphism \(f \in \mathrm{Aut}({\mathbb {G}})\) we choose. Obviously, the automorphisms can switch positions of whole layers, e.g. move all vertices from \(V_i\) to some \(V_j\) and vice versa—in this respect the layers are undistinguishable.
Lemma 2
For every \(i, j \in \{1, 2, \dots , k\}\) and Open image in new window ( Open image in new window ) the bipartite graph Open image in new window (with two distinguishable sides \(V_i, V_j\)) is homogeneous.
The vertex sets \(V_i\) and \(V_j\) are used here as unary relations that allow to tell the two layers of \({\mathbb {G}}_{i,j}\) (sides of \({\mathbb {G}}_{i,j}\)) apart. An example is shown on the right, with three layers \(V_1, V_2\) and \(V_3\), and three bipartite graphs \({\mathbb {G}}_{1,2}\), \({\mathbb {G}}_{2,3}\) and \({\mathbb {G}}_{1,3}\).
Proof
Things get more complicated when f operates only on some single layer of \({\mathbb {G}}_{i,j}\). W.l.o.g. suppose that it ‘touches’ only \(V_i\), so \(V(G_1) \cap V_j = \emptyset \). Now the above construction will not work out of the box—if we were unlucky, the automorphism of \({\mathbb {G}}\) we get by homogeneity moves the whole layer \(V_j\) to some \(V_n\) located ‘outside’ the graph \({\mathbb {G}}_{i,j}\) (\(n \notin \{i, j\}\)).

\(V(S_n) \cap V_m = 1\) for \(m \ne i\) (and this single vertex will be denoted as \(v_m^{(n)}\)),

\(V(S_n) \cap V_i = N\) (denote these vertices as \(s^{(n)}_1, s^{(n)}_2, s^{(n)}_3, \dots , s^{(n)}_N\)).
Furthermore, we define the type of graph \(S_n\) to be the sequence of types arising between \(V_i\) and other layers plus the list of edgecolors between all pairs of vertices \(v_\bullet ^{(n)}\) (enumerated in some consistent way). As there are only finitely many such types, by pigeonhole principle there exists a pair of graphs \(S_a\) and \(S_b\) with the same type.
Above, the colored triangles represent the types of connections. The order of those types may get permuted when applying \(h'\), but still—in line with the remark \(\bigstar \) — for each \(\beta \in \{1,2,\dots , k\} \setminus \{i\}\) the vertex \(h'\!\left( v_\beta ^{(a)}\right) \) must stay in the same layer as \(h'\!\left( v_\beta ^{(b)}\right) \), furthermore their type of connection with layer \(V_i\) is preserved.
We are going to apply to graphs \({\mathbb {G}}_{i,j}\) the following classification result:
Theorem 5
([16]). A countably infinite homogeneous bipartite graph (with distinguishable sides) is either empty, or full, or a perfect matching, or the complement of a perfect matching, or a universal graph.
 1.
all edges of \({\mathbb {G}}_{i, j}\) have the same color Open image in new window , i.e. \({\mathbb {G}}_{i,j}\) is a full or empty bipartite graph,
 2.
one of the colors Open image in new window forms a perfect matching in \({\mathbb {G}}_{i,j}\), the second one ( Open image in new window ) is then the complement of this matching.
Graphs of type 2. may be seen as bijections between their sets of vertices (layers). Lemma 3 states that those bijections have to agree with each other.
Lemma 3
Proof
 I.
\(v \notin \psi (\{x, x'\})\),
 II.
if \({\mathbb {G}}_{j,k}\) is a graph of type 2. defining a bijection \(\phi :V_k \rightarrow V_j\), then also \(v \notin \phi (\{y, y'\})\).

in \({\mathbb {G}}_{i,j}\) by definition of isomorphism we need the edges \(\{x, v\}\) and \(\{g(x), g(v)\}\) to be equally colored, and, in fact, they are. We get this thanks to the condition I.: x is connected with all vertices from \(V_j \setminus \{\psi (x)\}\) by Open image in new window edges, Open image in new window . We similarly handle \(x'\).

in turn in \({\mathbb {G}}_{j,k}\) — if it is a graph of type 1, the needed equality of colors of edges \(\{y, v\}\) and \(\{g(y), g(v)\}\) trivially holds. If it is a graph of type 2, the equality of colors is derived similarly as in \({\mathbb {G}}_{i,j}\), using the condition II.
Presence of the vertex v ensures that layer \(V_j\) is preserved by the full automorphism \(g' \in \mathrm{Aut}({\mathbb {G}})\) we get by homogeneity.
Since \({\mathbb {G}}_{i,j}\) is of type 2, the vertex \(\psi (x')\) is the only possible choice for the image of \(\psi (x)\) under \(g'\) — this is the only vertex \(x'\) is connected to by an appropriately colored edge. Because \(g'\) is an automorphism, we get that Open image in new window , which leads us to the contradiction. \(\square \)
From the lemma we have just proved one easily derives the following corollary:
Corollary 1
In Lemma 5 below, which is the last step of the proof of case (C), we will apply the following fact:
Lemma 4
Consider a homogeneous 3graph \({\mathbb {G}}\) and a partition of its vertex set \( V \ = \ \bigcup _{n \in {\mathbb {N}}} U_n \) into sets \(U_\bullet \) of equal finite cardinality. Suppose further that for every \(n\in {\mathbb {N}}\), there is an automorphism \(\pi _n\) of \({\mathbb {G}}\) that swaps \(U_0\) with \(U_n\) and is identity elsewhere. Then \({\mathbb {G}}\) admits wqo.
Proof
Let \(G_n\) denote the induced substructure of \({\mathbb {G}}'\) on vertex set \(U_n\). By the assumptions, for every \(n,m \in {\mathbb {N}}\) there is a swap of \(U_n\) and \(U_m\) that, extended with identity elsewhere, is an automorphism of \({\mathbb {G}}'\). In consequence, all structures \(G_\bullet \) are isomorphic, and the embedding order \(\unlhd \) of induced substructures of \({\mathbb {G}}'\) is isomorphic to finite multisets over \({\textsc {Age}}(G_0)\), ordered by multiset inclusion. Thus \(({\textsc {Age}}({\mathbb {G}}'), \unlhd )\) is isomorphic to the multiset inclusion in \(\mathcal{M}({\textsc {Age}}(G_0))\), which is a wqo as \(U_0\) is finite. For any wqo \((X, \le )\), analogous isomorphism holds between the lifted embedding order \(({\textsc {Age}}({\mathbb {G}}'), \unlhd _{X})\) and the multiset inclusion in multisets over induced substructures of \(G_0\) labeled by elements of X, and again the latter order is a wqo. Thus \({\mathbb {G}}'\) admits wqo. \(\square \)
Lemma 5
The 3graph \({\mathbb {G}}\) admits wqo.
Proof
 (a)
every layer \(V_i\) shares with every set \(U_n\) exactly one vertex: \(U_n \cap V_i = \{ v^{(n)}_i \}\),
 (b)
if \(f_{i,j}\) is the bijection determined by \({\mathbb {G}}_{i,j}\) (a graph of type 2.), then \(f_{i,j}(v^{(n)}_i) \in U_n\), so all the bijections preserve every set \(U_\bullet \).
Intuitively, \({\mathbb {G}}\) can by cut into thin ‘slices’ perpendicular to the layers \(V_\bullet \). By thin we mean that the slices have exactly one vertex in each layer. The cut is made along the bijections dictated by the graphs of type 2. as in the picture bellow:
By Lemma 4 we deduce that \({\mathbb {G}}\) admits wqo, which completes the proof. \(\square \)
Footnotes
References
 1.Abdulla, P.A., Nylén, A.: Timed Petri nets and BQOs. In: Colom, J.M., Koutny, M. (eds.) ICATPN 2001. LNCS, vol. 2075, pp. 53–70. Springer, Heidelberg (2001). https://doi.org/10.1007/3540457402_5CrossRefMATHGoogle Scholar
 2.Bojańczyk, M., Braud, L., Klin, B., Lasota, S.: Towards nominal computation. Proc. POPL 2012, 401–412 (2012)MATHGoogle Scholar
 3.Bojańczyk, M., Klin, B., Lasota, S.: Automata theory in nominal sets. Logical Methods Comput. Sci. 10(3:4) (2014). Paper 4Google Scholar
 4.Bojańczyk, M., Klin, B., Lasota, S., Toruńczyk, S.: Turing machines with atoms. LICS 2013, 183–192 (2013)MathSciNetMATHGoogle Scholar
 5.Cervesato, I., Durgin, N.A., Lincoln, P., Mitchell, J.C., Scedrov, A.: A metanotation for protocol analysis. In: Proceedings of CSFW 1999, pp. 55–69 (1999)Google Scholar
 6.Cherlin, G.: The classification of countable homogeneous directed graphs and countable homogeneous ntournaments. Mem. Am. Math. Soc. 131(621), xiv+161 (1998)MathSciNetMATHGoogle Scholar
 7.Cherlin, G.: Forbidden substructures and combinatorial dichotomies: WQO and universality. Discrete Math. 311(15), 1543–1584 (2011)MathSciNetCrossRefGoogle Scholar
 8.Delzanno, G.: An overview of MSR(C): a CLPbased framework for the symbolic verification of parameterized concurrent systems. Electr. Notes Theor. Comput. Sci. 76, 65–82 (2002)CrossRefGoogle Scholar
 9.Delzanno, G.: Constraint multiset rewriting. Technical report DISITR0508, DISI, Universitá di Genova (2005)Google Scholar
 10.Ding, G.: Subgraphs and wellquasiordering. J. Graph Theor. 16(5), 489–502 (1992)MathSciNetCrossRefGoogle Scholar
 11.Finkel, A., Schnoebelen, P.: Wellstructured transition systems everywhere! Theor. Comput. Sci. 256(1–2), 63–92 (2001)MathSciNetCrossRefGoogle Scholar
 12.Fraïssé, R.: Theory of Relations. NorthHolland, Amsterdam (1953)MATHGoogle Scholar
 13.Genrich, H.J., Lautenbach, K.: System modelling with highlevel Petri nets. Theor. Comput. Sci. 13, 109–136 (1981)MathSciNetCrossRefGoogle Scholar
 14.Hofman, P., Lasota, S., Lazić, R., Leroux, J., Schmitz, S., Totzke, P.: Coverability trees for Petri nets with unordered data. In: Jacobs, B., Löding, C. (eds.) FoSSaCS 2016. LNCS, vol. 9634, pp. 445–461. Springer, Heidelberg (2016). https://doi.org/10.1007/9783662496305_26CrossRefGoogle Scholar
 15.Jacobsen, L., Jacobsen, M., Møller, M.H., Srba, J.: Verification of timedarc Petri nets. In: Černá, I., Gyimóthy, T., Hromkovič, J., Jefferey, K., Králović, R., Vukolić, M., Wolf, S. (eds.) SOFSEM 2011. LNCS, vol. 6543, pp. 46–72. Springer, Heidelberg (2011). https://doi.org/10.1007/9783642183812_4CrossRefMATHGoogle Scholar
 16.Jenkinson, T., Truss, J.K., Seidel, D.: Countable homogeneous multipartite graphs. Eur. J. Comb. 33(1), 82–109 (2012)MathSciNetCrossRefGoogle Scholar
 17.Jensen, K.: Coloured Petri nets and the invariantmethod. Theor. Comput. Sci. 14, 317–336 (1981)MathSciNetCrossRefGoogle Scholar
 18.Lachlan, A.H., Woodrow, R.E.: Countable ultrahomogeneous undirected graphs. Trans. Amer. Math. Soc. 262(1), 51–94 (1980)MathSciNetCrossRefGoogle Scholar
 19.Lasota, S.: Decidability border for Petri nets with data: WQO dichotomy conjecture. In: Kordon, F., Moldt, D. (eds.) PETRI NETS 2016. LNCS, vol. 9698, pp. 20–36. Springer, Cham (2016). https://doi.org/10.1007/9783319390864_3CrossRefMATHGoogle Scholar
 20.Lasota, S., Piórkowski, R.: WQO dichotomy for 3graphs. CoRR, arXiv:1802.07612 (2018)
 21.Lazić, R., Newcomb, T., Ouaknine, J., Roscoe, A.W., Worrell, J.: Nets with tokens which carry data. In: Kleijn, J., Yakovlev, A. (eds.) ICATPN 2007. LNCS, vol. 4546, pp. 301–320. Springer, Heidelberg (2007). https://doi.org/10.1007/9783540730941_19CrossRefMATHGoogle Scholar
 22.Lazic, R., Schmitz, S.: The complexity of coverability in \(\nu \)Petri nets. In: Proceedings of LICS 2016, pp. 467–476 (2016)Google Scholar
 23.Macpherson, D.: A survey of homogeneous structures. Discrete Math. 311(15), 1599–1634 (2011)MathSciNetCrossRefGoogle Scholar
 24.RosaVelardo, F.: Ordinal recursive complexity of unordered data nets. Inf. Comput. 254, 41–58 (2017)MathSciNetCrossRefGoogle Scholar
 25.RosaVelardo, F., de FrutosEscrig, D.: Decidability and complexity of Petri nets with unordered data. Theor. Comput. Sci. 412(34), 4439–4451 (2011)MathSciNetCrossRefGoogle Scholar
Copyright information
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made. The images or other third party material in this book are included in the book's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the book's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.