Given a graph G, we study the 2-edge-connected subgraph polytope \mathrm{TECSP}(G), which is given by the convex hull of the incidence vectors of all 2-edge-connected subgraphs of G. We describe the lattice points of this polytope by linear inequalities which provides an ILP-algorithm for finding a 2-edge-connected subgraph of maximum weight. Furthermore, we characterize when these inequalities define facets of \mathrm{TECSP}(G). We also consider further types of supporting hyperplanes of \mathrm{TECSP}(G) and study when they are facet-defining. Finally, we investigate the efficiency of our considered inequalities practically on some classes of graphs.
A graph G is 2-edge-connected, if and only if one of the following equivalent conditions is true:
for every pair of distinct vertices u,v \in V(G), there exist at least two edge-disjoint paths connecting u and v;
G is connected and does not contain any bridges.
Let G be 2-edge-connected and C \in \mathrm{CP}(G) be a coparallel class. Then the connected components of G-C are 2-edge-connected.
Let H \subseteq G be a 2-edge-connected subgraph of a graph G. Then for every coparallel class C \in \mathrm{CP}(G) of G we either have C \subseteq E(H) or C \cap E(H) = \emptyset. In particular, for a coparallel class C \in \mathrm{CP}(G) of a 2-edge-connected graph G, every other coparallel class C' \in \mathrm{CP}(G)\setminus \{C\} is completely contained in a single connected component of G-C.
Let G be 2-edge-connected, C \in \mathrm{CP}(G) be a coparallel class of G, and G' be a connected component of G-C that contains edges. If a coparallel class C' of G' is not a coparallel class of G, then C'=C_1 \dot{\cup} C_2, where C_1 and C_2 are coparallel classes of G.
Let G be a graph and C \in \mathrm{CP}(G) be a coparallel class of G. For all x \in \mathrm{TECSP}(G), and all e,f \in C we have x_e=x_f .
For a graph G, we have
Suppose that e \in E(G) is a bridge of G. Then we have x_e=0 for all x \in \mathrm{TECSP}(G), since bridges are never contained in 2-edge-connected subgraphs. Moreover, bridges do not have any impact on the number of coparallel classes of a graph G. In fact, we can delete all bridges and assume that G does not contain any bridges. Now, by Observation 2.5 and the fact that \mathrm{CP}(G) forms a partition of E(G), we have \dim(\mathrm{TECSP}(G)) \le \vert \mathrm{CP}(G)\vert. Assuming G does not contain any bridges, all connected components G_1,\ldots,G_m of G are 2-edge-connected. Let C \in \mathrm{CP}(G) be a coparallel class of G. Then C \subseteq E(G_i) for some i \in [m] due to Observation 2.3. Let H_1,\ldots,H_k denote the connected components of G_i-C. Due to Observation 2.2, \chi^{H_j} \in \mathrm{TECSP}(G) for all j \in [k]. Since \textbf{0} \in \mathrm{TECSP}(G), \mathrm{aff}(\mathrm{TECSP}(G)) is a linear subspace of \mathbb{R}^{E(G)} and \chi^C=\chi^{G_i}-\sum_{j=1}^k \chi^{H_j} \in \mathrm{aff}(\mathrm{TECSP}(G)), which shows \dim(\mathrm{aff}(\mathrm{TECSP}(G))) \ge \vert \mathrm{CP}(G) \vert.
We can also define the 2-edge-connected subgraph polytope for a multigraph G, i.e., a graph with multi-edges between common vertex pairs. In G, we may replace every multi-edge by a path of length two. Such a path consists of two coparallel edges. Denoting the obtained simple graph by G', the polytopes \mathrm{TECSP}(G) and \mathrm{TECSP}(G') are affinely isomorphic due to Observation 2.5. Therefore, we also cover multigraphs in this article. To simplify notation, we will always assume that the considered graphs are simple in the following.
Let G be a graph, let E_3 \subseteq E(G) be the edges contained in cuts of size 3 and let E' \subseteq E(G) be an inclusion-wise maximal set of edges not containing bridges or 2-cuts. Then \mathrm{ESP}(G) is given by
x_e=0 for each bridge e \in E(G),
x_{e_1}=x_{e_i} for each coparallel class C=\{e_1,\ldots,e_k\} \in \mathrm{CP}(G) and all i=2,\ldots,k,
0 \le x_e \le 1 for each e \in E(G) \setminus E_3 that is not a bridge,
x(F)-x(C\setminus F) \le \vert F \vert -1 for each minimal cut C \subseteq E', and with property (P), and each F \subseteq C, \vert F \vert odd; Hereby property (P) is that if C has a chord defined by cuts F_1,F_2 \subseteq E(G), then F_1 or F_2 has cardinality 2.
Moreover, the system above is nonredundant.
Let G be a 2-edge-connected graph, C \in \mathrm{CP}(G) and e\in C. Then the inequality
defines a facet of \mathrm{TECSP}(G) if and only if e is not contained in a 3-cut.
Suppose that e is contained in a 3-cut \delta(S)=\{e,e_1,e_2\} for some \emptyset \subsetneq S \subsetneq V(G). We consider the inequalities
and
These are both supporting hyperplanes of \mathrm{TECSP}(G) since 2-edge-connected subgraphs H of G cannot contain exactly one edge of a cut, as otherwise H would contain a bridge. In fact, they are special cases of the so-called asymmetric cut inequalities that will be studied in ?. These two inequalities define different proper faces of \mathrm{TECSP}(G), since e.g. \chi^G is not contained in the corresponding face, and they already imply x_e \ge 0. Hence, the latter does not define a facet.
Suppose now that e is not contained in a 3-cut and let \mathcal{F} \subseteq \mathrm{TECSP}(G) denote the face defined by x_e=0. By an analogous argument as in [BARAHONA198640], we get the result as follows. Let C \in \mathrm{CP}(G) be the coparallel class that contains e. \mathcal{F} is affinely isomorphic to the 2-edge-connected subgraph polytope \mathrm{TECSP}(G-C) of the graph G-C. Since e is not contained in a 3-cut, the same holds for any e' \in C, and therefore, two edges f_1,f_2 \in E(G) \setminus C are coparallel in G if and only if they are coparallel in G-C. This implies \vert \mathrm{CP}(G-C) \vert =\vert \mathrm{CP}(G) \vert-1 and the claim follows by Theorem 2.6.
Let G be a 2-edge-connected graph and e \in E(G). Then the inequality
defines a facet of \mathrm{TECSP}(G).
Let \mathcal{F} \subseteq \mathrm{TECSP}(G) denote the face defined by x_e = 1. Let \mathrm{CP}(G)=\{C_e,C_1,\ldots,C_\ell\}, where C_e is the coparallel class containing e. For j \in [\ell], we denote by G_j^e the connected component of the graph G-C_j containing e. Since G_j^e is 2-edge-connected (by Observation 2.2), \chi^G and \chi^{G_j^e}, for j=1,\ldots,\ell, are vertices of \mathcal{F}. We now show that these \vert \mathrm{CP}(G)\vert many vectors are affinely independent, which by Theorem 2.6 implies that \mathcal{F} is a facet. For this aim, we define a partial order on the coparallel classes C_1,\ldots,C_\ell as follows: for i,j \in [\ell],i\neq j, let
To show that <_e defines a partial order, we show the following claim.
which can be seen as follows. Suppose C_i <_e C_j. Then by definition C_i \nsubseteq E(G_j^e). Due to Observation 2.3, this already means that C_i \cap E(G_j^e)=\emptyset. Hence, G-C_i contains all edges of G_j^e which implies that G_j^e \subseteq G_i^e. Since at least one edge f of C_j is incident to an edge of G_j^e and since f is contained in G-C_i it follows that f \in E(G_i^e). By Observation 2.3, we conclude that C_j \subseteq E(G_i^e).
Using this claim we can easily verify that <_e defines a partial order on C_1,\ldots,C_\ell: If C_i <_e C_j, then C_j \subseteq E(G_i^e) and hence, C_j \nless_e C_i. This shows that <_e is antisymmetric. Suppose now that C_i <_e C_j and C_j <_e C_k. From the first relation we have C_i \nsubseteq E(G_j^e) by definition and from the second relation we have G_k^e \subseteq G_j^e by the claim above. Hence, C_i \nsubseteq E(G_k^e) which implies C_i <_e C_k and in particular, <_e is transitive.
The vectors \chi^{G_j^e} satisfy
for all j \in [\ell].
Let \pi \colon \mathrm{TECSP}(G) \to \mathbb{R}^{\ell} be the map defined by \pi(x)_i \coloneqq x_f, for any f \in C_i, for all i \in [\ell]. By Observation 2.5, this is well-defined. Furthermore, we have
Without loss of generality we can assume that our indices satisfy i \le j if C_i \le C_j. Writing the vectors \pi(\chi^G)-\pi(\chi^{G_1^e}),\ldots,\pi(\chi^G)-\pi(\chi^{G_\ell^e}) row-wise in a matrix, yields a lower triangular \ell \times \ell-matrix that has ones in the diagonal. Therefore, the matrix has rank \ell, which finishes the proof.
Let G be 2-edge-connected and \emptyset \subsetneq S \subsetneq V such that \vert\delta(S)\vert=2. Then the face \mathcal{F} \subseteq \mathrm{TECSP}(G) defined by the asymmetric cut inequality
is given by \mathcal{F}=\mathrm{TECSP}(G). In particular, the asymmetric cut inequality does not define a facet of \mathrm{TECSP}(G) in this case.
Let G be 2-edge-connected and \emptyset \subsetneq S \subsetneq V(G) such that \delta(S) is a minimal cut set with \vert \delta(S) \vert \ge 3. For e \in \delta(S) the asymmetric cut inequality
defines a facet of \mathrm{TECSP}(G) if and only if every bridge in G[S] and G[V(G) \backslash S] lies in a coparallel class together with an edge of \delta(S).
Suppose first that G[S] contains a bridge f \in E(G[S]) that is not contained in a coparallel class with any edge of \delta(S). Since \delta(S) is a minimal cut set, we know that G[S] and G[V(G) \setminus S] are connected. Then the graph G[S]-f has two connected components G_1,G_2. For i=1,2, let E_i be the edges of \delta(S) incident to G_i. Since f is not in a 2-cut with any edge of \delta(S), we have \vert E_1\vert,\vert E_2\vert \ge 2. Moreover, we have \delta(V(G_1))=E_1 \cup \{f\} and \delta(V(G_2))=E_2 \cup \{f\} and without loss of generality we can assume that e \in E_1. An example for the described situation is shown in ?. Then we have that
Since \vert E_1\vert,\vert E_2\vert \ge 2, both of the two latter asymmetric cut inequalities define proper faces of \mathrm{TECSP}(G) and it is clear that these two faces are different. This implies that the first asymmetric cut inequality is not a facet.
We now show that in all other cases, the asymmetric cut inequality defines a facet of \mathrm{TECSP}(G). The asymmetric cut inequality is actually a special case of inequality (d) in Theorem 2.8 and therefore defines a facet of \mathrm{ESP}(G) due to the remark directly below Theorem 3.4. Let \mathcal{F}_{\mathrm{ESP}(G)} and \mathcal{F}_{\mathrm{TECSP}(G)} denote the face defined by the asymmetric cut inequality of \mathrm{ESP}(G) and of \mathrm{TECSP}(G), respectively. We will show that \mathrm{aff}(\mathcal{F}_{\mathrm{ESP}(G)}) \subseteq \mathrm{aff}(\mathcal{F}_{\mathrm{TECSP}(G)}), which implies that \mathcal{F}_{\mathrm{TECSP}(G)} is a facet of \mathrm{TECSP}(G), since \mathcal{F}_{\mathrm{ESP}(G)} is a facet of \mathrm{ESP}(G), and both polytopes have the same dimension due to [BARAHONA198640, Theorem 4.1] and Theorem 2.6. Let H \subseteq G be a subgraph of G such that \chi^H is a vertex of \mathcal{F}_{\mathrm{ESP}(G)}. This implies that H either contains no edges of \delta(S) or H contains the edge e \in \delta(S) and exactly one other edge f \in \delta(S)\setminus \{e\}. Since the connected components of H are Eulerian, H is the edge-disjoint union of cycles C_1,\ldots,C_k \subseteq G. If E(H) \cap \delta(S)=\emptyset, it is clear that \chi^{C_i} \in \mathcal{F}_{\mathrm{TECSP}(G)} for all i \in [k]. If e \in E(H), there is exactly one cycle C_j that contains e for some j \in [k]. Then C_j also contains the edge f, because E(C_j) \cap \delta(S) has even cardinality, and again we have \chi^{C_i} \in \mathcal{F}_{\mathrm{TECSP}(G)} for all i \in [k], since all other cycles do not contain any edge of \delta(S). Since the all-zeros vector \mathbf{0} is contained in \mathcal{F}_{\mathrm{TECSP}(G)}, \mathrm{aff}(\mathcal{F}_{\mathrm{TECSP}(G)}) is a linear subspace and also \chi^H=\sum_{i=1}^k \chi^{C_i} \in \mathrm{aff}(\mathcal{F}_{\mathrm{TECSP}(G)}).
Let G be a 2-edge-connected graph with vertex set V, let \emptyset \subsetneq S \subsetneq V, such that \delta(S) is an inclusion-wise minimal cut with e_1 \in E(G[S]) and e_2 \in E(G[V\backslash S]). Then the connectivity cut inequality
defines a facet of \mathrm{TECSP}(G) if and only if one of the following two conditions holds:
e_1 or e_2 is a bridge in G[S] or G[V \setminus S], and \vert \delta(S) \vert=2;
neither e_1 nor e_2 is a bridge in G[S] resp. G[V \setminus S], and if there is a bridge f in G[W] for W \in \{S,V \setminus S\}, then \vert \delta(W,f)\vert=1.
Let \mathcal{F} \subseteq \mathrm{TECSP}(G) be the face of \mathrm{TECSP}(G) defined by the considered connectivity cut inequality. We provide the proof in separate claims.
Claim 1: If e_1 is a bridge in G[S] or e_2 is a bridge in G[V \setminus S] and \vert \delta(S) \vert =2, the considered connectivity cut inequality defines a facet of \mathrm{TECSP}(G).
Proof. Without loss of generality we may assume that e_1 is a bridge in G[S], since the condition is symmetric for e_1 and e_2. If \vert \delta(S) \vert = 2, the two edges of \delta(S) and e_1 belong to the same coparallel class. The connectivity cut inequality thus simplifies to x_{e_2} \le 1, and therefore defines a facet by Theorem 3.2. \blacksquare
Claim 2: If e_1 is a bridge in G[S] or e_2 is a bridge in G[V \setminus S] and \vert \delta(S) \vert \ge 3, the considered connectivity cut inequality does not define a facet of \mathrm{TECSP}(G).
Proof. Without loss of generality we may again assume that e_1 is a bridge in G[S]. Let H \subseteq G be a 2-edge-connected subgraph of G such that \chi^H \in \mathcal{F}. This is only possible if H contains at least one of e_1 and e_2. If e_2 \in E(H), it is clear that \chi^H \in \mathcal{F}_{e_2}, where \mathcal{F}_{e_2} \subseteq \mathrm{TECSP}(G) denotes the facet defined by x_{e_2} \le 1. Suppose that H contains the edge e_1. In G[S] the edge e_1 is a bridge, but, since H is 2-edge-connected, the edge e_1 has to be contained in a cycle in H. Therefore, E(H) intersects \delta(S) non-trivially. Since \chi^H \in \mathcal{F}, H also contains the edge e_2 and we have \mathcal{F} \subseteq \mathcal{F}_{e_2}. If \vert \delta(S) \vert \ge 3, \chi^G \in \mathcal{F}_{e_2} but \chi^G \notin \mathcal{F} and hence, \mathcal{F} \subsetneq \mathcal{F}_{e_2} which shows that \mathcal{F} is not a facet, since \mathcal{F}_{e_2} is a proper face of \mathrm{TECSP}(G) by Theorem 3.2. \blacksquare
Claim 3: If neither e_1 nor e_2 is a bridge in G[W] for W \in \{S,V \setminus S\}, and there is a bridge f of G[W], such that \vert \delta(W,f) \vert \ge 2, then the considered connectivity cut inequality does not define a facet.
Proof. Without loss of generality we may assume that f is a bridge of G[S]. Let G_1 and G_2 denote the connected component of G[S]-f, where G_1 contains e_1. This situation is depicted in ?. We have \delta(V(G_2))=\{f\} \cup \delta(S,f). Suppose H \subseteq G is a 2-edge-connected subgraph such that \chi^H \in \mathcal{F}. Then, one of the following cases occurs:
\bullet H contains e_1 and e_2.
If f \in E(H), then H contains exactly one edge of \delta(S,f).
If f \notin E(H), H contains exactly two edges of \delta(S) \setminus \delta(S,f) and, therefore, H does not contain any edge of \delta(S,f).
\bullet H contains exactly one edge of \{e_1,e_2\}. Then H neither contains an edge of \delta(S) nor the edge f.
In all cases, \chi^H attains equality in the asymmetric cut inequality
Let \mathcal{F}' \subseteq \mathrm{TECSP}(G) denote the face defined by (?). In particular, \mathcal{F} \subseteq \mathcal{F}'. Moreover, since neither e_1 nor e_2 is a bridge in G[S] resp. G[V\setminus S], there exists a cycle C \subseteq G, that contains exactly one edge of \delta(S,f), one edge of \delta(S) \setminus \delta(S,f) and the edge f but neither e_1 nor e_2. This means \chi^C \in \mathcal{F}' \setminus \mathcal{F} and, therefore, \mathcal{F} \subsetneq \mathcal{F}'. The assumption \vert \delta(S,f) \vert \ge 2 implies that \mathcal{F}' is a proper face of \mathrm{TECSP}(G), since for example \textbf{1}=\chi^G \notin \mathcal{F}'. Hence, the connectivity cut inequality does not define a facet in this case. \blacksquare
Claim 4: If neither e_1 nor e_2 is a bridge in G[S] resp. G[V\setminus S], and for all bridges f of G[W] for W \in \{S,V\setminus S\} (if there are any), we have \vert \delta(W,f) \vert=1, then the considered connectivity cut inequality defines a facet.
Proof. Let \mathcal{F} \subseteq \mathrm{TECSP}(G) denote the face defined by the considered connectivity cut inequality, and let U \subseteq \mathbb{R}^{E(G)} be the linear subspace with \mathrm{aff}(\mathcal{F})=y+U for y \in \mathcal{F}. We have to show that \dim(U)=\vert \mathrm{CP}(G) \vert-1. It is clear that \mathcal{F} \subsetneq \mathrm{TECSP}(G), since \textbf{0} \notin \mathcal{F}, and, hence, \dim(U) \ge \vert \mathrm{CP}(G) \vert -1 implies that \mathcal{F} is a facet. For this aim, we will give \vert \mathrm{CP}(G) \vert -1 many linearly independent vectors contained in U. First, we describe how such a considered cut looks like.
An example for the following can be seen in ?. Since \delta(S) is an inclusion-wise minimal cut, the graph G-\delta(S) consists of the two connected components G^{(1)}\coloneqq G[S] and G^{(2)} \coloneqq G[V\setminus S] with e_i \in G^{(i)} for i=1,2. For i=1,2, let B^{(i)} \subseteq E(G_i) denote the set of bridges in G^{(i)}; let G^{(i)}_1,\ldots,G^{(i)}_{n_i} denote the connected components of G^{(i)}-B^{(i)}, named such that e_i \in G^{(i)}_1. We remark that all of these connected components are 2-edge-connected (including the case that a connected component may just be an isolated vertex), since otherwise there would be more bridges not contained in any B^{(i)}. By contracting all connected components of G^{(i)}-B^{(i)} to single vertices, we obtain a tree T^{(i)} that has a vertex v^{(i)}_j for each connected component G^{(i)}_j and edge set E(T^{(i)})=\{\{v^{(i)}_j,v^{(i)}_{j'}\} \mid \text{there is } b \in B^{(i)} \text{ connecting } G^{(i)}_j \text{ and } G^{(i)}_{j'}\}. As we did for B^{(i)}, we can also interpret the edges of \delta(S) as edges having one end in V(T^{(1)}) and the other one in V(T^{(2)}) depending on to which components of G^{(i)}-B^{(i)} their incident vertices belong, and we will denote this set by \overline{\delta(S)}. Let i=1,2. Then, for every leaf w of T^{(i)} there has to be at least one edge e \in \overline{\delta(S)} that is incident to w, since otherwise the graph G would not be 2-edge-connected. Furthermore, for every leaf w \neq v^{(1)}_1,v^{(2)}_1 of T^{(i)}, there has to be exactly one edge e \in \overline{\delta(S)} that is incident to w, since otherwise the edges f \in B^{(i)} that correspond to the edges of E(T^{(i)}) contained in the unique path P \subseteq T^{(i)} connecting v^{(i)}_1 and w would not satisfy the condition \vert \delta(W_i,f) \vert =1, for W_1=S, and W_2=V\setminus S. This also implies that for every non-leaf vertex w' \in V(T^{(i)}), w' \neq v^{(i)}_1, there is no edge e \in \overline{\delta(S)} incident to w', since otherwise the edges f \in B^{(i)} that correspond to the edges of E(T^{(i)}) contained in the unique path P' \subseteq T^{(i)} between v^{(i)}_1 and w' would not satisfy the condition \vert \delta(W_i,f) \vert =1. Analogously, there can be no vertex w'' \in V(T^{(i)})\setminus \{v^{(i)}_1\} that has degree three or more in T^{(i)}. Hence, every vertex v_i \in V(T^{(i)}) \setminus \{v^{(i)}_1} has degree at most two in T^{(i)}, the degree one vertices of T^{(i)} are incident to exactly one edge of \overline{\delta(S)}, and the degree two vertices of T^{(i)} are not incident to any edge of \overline{\delta(S)}. The edges e \in \overline{\delta(S)} then connect v^{(1)}_1 or a leaf of T^{(1)} with v^{(2)}_1 or a leaf of T^{(2)}. In other words, there are k\coloneqq \vert \delta(S) \vert many internally disjoint paths P_1,\ldots,P_k between v^{(1)}_1 and v^{(2)}_1 in the graph \overline{G}\coloneqq (V(T^{(1)}\cup T^{(2)}),E(T^{(1)}\cup T^{(2)}) \cup \overline{\delta(S)}), and each of these paths contains exactly one edge of \overline{\delta(S)}. Furthermore, for \ell \in [k], P_\ell corresponds to a subgraph in G by re-substituting, for i=1,2, the vertices v^{(i)}_j by the subgraphs G^{(i)}_j, the edges E(T^{(i)}) by B^{(i)}, and \overline{\delta(S)} by \delta(S), and this subgraph will be denoted by G(P_\ell) \subseteq G. Note that each G(P_\ell) contains the subgraphs G^{(1)}_1 and G^{(2)}_1. We remark that the edges in E(G(P_\ell)) \cap (B^{(1)} \cup B^{(2)} \cup \delta(S)) define a coparallel class of G for all \ell \in [k], as long as k \ge 3. For k=2, the edges B^{(1)} \cup B^{(2)} \cup \delta(S) define a coparallel class.
Let C' \coloneqq E(G(P_1))\cap (B^{(1)}\cup B^{(2)} \cup \delta(S)) \in \mathrm{CP}(G) if k \ge 3, and C' \coloneqq B^{(1)}\cup B^{(2)} \cup \delta(S) \in \mathrm{CP}(G) for k=2. We can now enumerate vectors \chi^{(C)} for all C \in \mathrm{CP}(G) \setminus \{C'\}, such that \chi^{(C)} \in U and (\chi^{(C)})_{C \in \mathrm{CP}(G)\setminus \{C'\}} are linearly independent. Note that for two vectors \chi_1,\chi_2 \in \mathcal{F}, we have \chi_1-\chi_2 \in U. We distinguish several cases depending on the type of the coparallel class.
Case (1): Coparallel classes contained in G^{(i)}_j for i=1,2, and j \in [n_i]\setminus \{1\}: Let C \in \mathrm{CP}_{G^{(i)}_{j}}(G) for i \in \{1,2\}, and j \in [n_i] \setminus \{1\} (assuming E(G^{(i)}_{j}) \neq \emptyset) be a coparallel class. We will define a graph F to help us construct feasible solutions on \mathcal{F}: Let Q \in \{G(P_1),\ldots,G(P_k)\} be the subgraph that contains G^{(i)}_j, let Q' \in \{G(P_1),\ldots,G(P_k)\}\setminus\{Q\}, and let v_1,v_2 \in V(G^{(i)}_j) be the two vertices that are incident to an edge of \delta(S) \cup B^{(i)}. Now, let F be the union of Q and Q', after removing the edges E(G^{(i)}_j) and the vertices V(G^{(i)}_j)\setminus \{v_1,v_2\}. Consider the set \mathcal{H} \coloneqq \{H \subseteq G^{(i)}_j \mid F \cup H \text{ is2-edge-connected} \}. We have \chi^H+\chi^F \in \mathcal{F} for all H \in \mathcal{H}. Now, let w \notin V(G^{(i)}_j) be a new vertex, f_1\coloneqq \{v_1,w\}, and f_2 \coloneqq \{v_2,w\}, and consider the graph \widetilde{G} \coloneqq (V(G^{(i)}_{j}) \cup \{w\},E(G^{(i)}_{j})\cup \{f_1,f_2\}) (we add two edges here to avoid creating a multi-edge in case \{v_1,v_2\} \in E(G^{(i)}_j)). Note that \mathrm{CP}(\widetilde{G})=\mathrm{CP}_{G^{(i)}_j}(G)\cup\{\{f_1,f_2\}\}. The incidence vectors \{\chi^H + \chi^{\{f_1,f_2\}}\in \mathbb{R}^{E(\tilde{G})} \mid H \in \mathcal{H}\} are exactly the vertices of the face \mathcal{F}_{f_1} \subseteq \mathrm{TECSP}(\widetilde{G}) defined by x_{f_1} \le 1. By the proof of Theorem 3.2, there are \vert \mathrm{CP}(\widetilde{G})\vert-1=\vert \mathrm{CP}_{G^{(i)}_j}(G) \vert many linearly independent vectors of the form \chi^{H_1}-\chi^{H_2}=\chi^{E(H_1)\cup \{f_1,f_2\}}-\chi^{E(H_2)\cup \{f_1,f_2\}} \in \mathbb{R}^{E(\tilde{G})} for some H_1,H_2 \in \mathcal{H}. Since we have \chi^{H_1}-\chi^{H_2} \in U (now as incidence vectors in \mathbb{R}^{E(G)}) for all H_1,H_2 \in \mathcal{H}, this implies that we can find \vert \mathrm{CP}_{G^{(i)}_j}(G) \vert many linearly independent vectors in U, whose support is a subset of E(G^{(i)}_{j}).
We can do the above for any subgraph G^{(i)}_{j}, with i=1,2 and j \in [n_i]\setminus\{1\}, and since the supports of the vectors that correspond to different graphs G^{(i)}_j are disjoint, we find a linearly independent vector for any coparallel class C\in \mathrm{CP}_{G^{(i)}_{j}}(G). We additionally remark that all these vectors are contained in the linear subspace \{x \in \mathbb{R}^{E(G)} \mid x_e=0 \text{ for all } e \in \delta(S) \cup B^{(1)} \cup B^{(2)} \cup E(G^{(1)}_{1}) \cup E(G^{(1)}_{2})\}.
Case (2): Coparallel classes that contain an edge of \delta(S): For any two distinct subgraphs H_1,H_2 \in \{G(P_2),\ldots,G(P_k)\} (assuming k \ge 3), we have \chi^{G(P_1) \cup H_1},\chi^{H_1 \cup H_2} \in \mathcal{F}, which implies that \chi^{H_2}-\chi^{G(P_1)}= \chi^{H_1 \cup H_2}-\chi^{G(P_1) \cup H_1} \in U. This defines k-1 vectors contained in U and projecting these onto the coordinates \delta(S)\setminus \{e\}, with \{e\}=\delta(S) \cap E(G(P_1)), yields the standard unit basis of \mathbb{R}^{\delta(S)\setminus\{e\}}. Hence, these vectors are linearly independent and since the vectors considered in Case (1) are all contained in the linear subspace \{x \in \mathbb{R}^{E(G)} \mid x_e=0 \text{ for all } e \in \delta(S) \}, which is not true for the new vectors, the vectors of Case (1) together with the ones of Case (2) are linearly independent. For the next case, we point out that the vectors considered in Case (2) are all contained in the linear subspace \{x \in \mathbb{R}^{E(G)} \mid x_e=0 \text{ for all } e \in E(G^{(1)}_1) \cup E(G^{(2)}_1)\}. We additionally remark that this is the only case were we encounter a unique coparallel class to which we do not assign a vector in U. Furthermore, this also agrees with the case k=2, since here the edges in B^{(1)} \cup B^{(2)} \cup \delta(S) define a single coparallel class, and therefore we do not assign a vector of U to this coparallel class.
Case (3): Coparallel classes that are contained in G^{(i)}_1 for i=1,2: We will first give a vector for any coparallel class C \in \mathrm{CP}(G^{(i)}_1) for i=1,2. We consider all incidence vectors X \coloneqq \{\chi^H \mid H \subseteq G^{(i)}_1 \,\, 2\text{-edge-connected, } e_i \in E(H)\} \subseteq \mathcal{F}. With the same arguments as above we can find \vert \mathrm{CP}(G^{(i)}_1) \vert-1 many linearly independent vectors of the form \chi_1-\chi_2 \in U, with \chi_1,\chi_2 \in X. These vectors form a linearly independent family with all the previously considered vectors, since they are supported on a subset of E(G^{(i)}_1)\setminus\{e_i\}, whereas for all the so far considered vectors \chi \in U we have \chi_e=0 for all e \in E(G^{(i)}_1). This yields \vert \mathrm{CP}(G_1^{(1)})\vert+\vert \mathrm{CP}(G_1^{(2)})\vert-2 many new vectors contained in U.
Now, pick two distinct graphs H_1,H_2 \in \{G(P_1),\ldots,G(P_k)\}. Then \chi^{H_1 \cup H_2} \in \mathcal{F} and this implies \chi^{H_1 \cup H_2}-\chi^{G^{(1)}_1},\chi^{H_1 \cup H_2}-\chi^{G^{(2)}_1} \in U. It is clear that these vectors are linearly independent, since projecting these vectors onto the coordinates \{e_1,e_2\} yields the standard unit basis of \mathbb{R}^{\{e_1,e_2\}}. This also shows that these two vectors are linearly independent to all previously considered vectors since the latter are all contained in the linear subspace \{x \in \mathbb{R}^{E(G)} \mid x_{e_1}=x_{e_2}=0\}. Therefore, we constructed \vert \mathrm{CP}(G_1^{(1)})\vert+\vert \mathrm{CP}(G_1^{(2)})\vert many new linearly independent vectors in Case (3).
Note that \vert \mathrm{CP}(G^{(1)}_1) \vert \le \vert \mathrm{CP}_{G^{(1)}_1}(G)\vert as discussed in the preliminaries. Now, let C \in \mathrm{CP}(G^{(1)}_1) be a coparallel class of G^{(1)}_1 that is not a coparallel class of G. Then C=C_1 \dot{\cup} \ldots \dot{\cup} C_t is the disjoint union of coparallel classes C_1,\ldots,C_t \in \mathrm{CP}(G). If we contract the connected components of G^{(1)}_1-C to one single vertex each, we obtain a cycle with edges C. Moreover, the edges of any coparallel class C_s, s=1,\ldots,t, form a path in this cycle, which can be seen following the arguments of Observation 2.4. See ? for a visualization of the described situation. We can assume that the labeling is chosen such that the edges of C_s and C_{s+1} are next to each other in this cycle for all s \in [t], where C_{t+1} \coloneqq C_1. Furthermore, we can assume that either e_1 \in C_1 or e_1 is contained in a connected component of G^{(1)}_1-C that contains a vertex of an edge of C_1. For every s \in [t] there is an edge f_s \in B^{(1)}\cup\delta(S), such that f_s is incident to the vertex in the contracted graph between C_{s-1} and C_{s}, setting C_0 \coloneqq C_t. We remark that f_s is contained in exactly one graph of \{G(P_1),\ldots,G(P_k)\} and we will denote this graph by P_{f_s}. For s \in [1,t-1], let H_s denote the subgraph that contains all connected components of G^{(1)}_1-C that have at least one common vertex with an edge of C_1 \cup \ldots \cup C_s, all edges of C_1 \cup \ldots \cup C_s, and the edges (E(P_{f_1}) \cup E(P_{f_{s+1}}))\setminus E(G^{(1)}_1) (with induced vertex set). Furthermore, let H_t\coloneqq P_{f_1} \cup P_{f_t}. Then H_s is a 2-edge-connected subgraph and \chi^{H_s} \in \mathcal{F} for all s \in [t]. Hence, we have \chi^{(s)} \coloneqq \chi^{H_{s+1}}-\chi^{H_{s}} \in U for all s \in [t-1]. Now, choose g_s \in C_s for s \in [t]. Projecting \chi^{(s)}, s=1,\ldots,t-1, onto the coordinates \{g_2,\ldots,g_t\} yields the standard basis of \mathbb{R}^{\{g_2,\ldots,g_t\}}, and hence, \{\chi^{(s)}\}_{s \in [t-1]} is linearly independent, while the previously considered vectors are all contained in the linear subspace U_C \coloneqq \{x \in \mathbb{R}^{E(G)} \mid x_{g_1}=\ldots=x_{g_t}\}. We thus have t-1 additional vectors contained in U. This can analogously be done for every coparallel class of G^{(2)}_1 that is not a coparallel class of G. We remark that for different coparallel classes C_1,C_2 \in \mathrm{CP}(G^{(1)}_1) \cup \mathrm{CP}(G^{(2)}_2), the newly constructed vectors belonging to C_1 are contained in U_{C_2} and the other way around due to Observation 2.3, and therefore all the constructed vectors are linearly independent.
In total, we constructed \vert \mathrm{CP}(G) \vert-1 linearly independent vectors contained in U, which completes the proof of Claim 4. \blacksquare
Claims 1 – 4 establish the proof.
Let G be a 2-edge-connected graph with edge set E. Let
Then
It is clear that \{\chi^H \in \{0,1\}^E \mid H \subseteq G \text{ is } 2 \text{-edge-connected}\} \subseteq \mathcal{P}(G) \cap \mathbb{Z}^{E}. Let y \in \mathcal{P}(G)\cap \mathbb{Z}^E and let G[y] be the subgraph of G with edge set E(G[y])=\operatorname{\mathrm{supp}}(y) and induced vertex set of E(G[y]). Hence, \chi^{G[y]}=y. The claim follows if we can show that G[y] is 2-edge-connected. We can assume that y \neq \textbf{0}, since otherwise G[y] is the empty graph that is 2-edge-connected by definition. Suppose first that G[y] is not connected. Then there exist at least two connected components G_1 and G_2. By definition, G[y] does not contain any isolated vertices and therefore, E(G_1),E(G_2) \neq \emptyset. Considering the cut inducing set S=V(G_1), and choosing e_1 \in E(G_1), and e_2 \in E(G_2), we obtain a violated connectivity cut inequality, a contradiction to y \in \mathcal{P}(G). Suppose now that G[y] is connected but not 2-edge-connected. Then G[y] has a bridge e \in E(G[y]). Furthermore, there is a vertex set \emptyset \subsetneq S \subsetneq V(G) such that \delta(S) \cap E(G[y])=\{e\}. Then we have
and hence, y violates an asymmetric cut inequality, again a contradiction to y \in \mathcal{P}(G). This implies that G[y] is 2-edge-connected.
Let G be 2-edge-connected, let C \in \mathrm{CP}(G) be a coparallel class, and let G_1,\ldots,G_r be the connected components of G-C that contain edges. Let f \in C and e_j \in E(G_j) for j \in [r]. Then the coparallel class inequality
defines a facet of \mathrm{TECSP}(G).
Let \mathcal{F} \subseteq \mathrm{TECSP}(G) denote the face that is defined by the considered coparallel class inequality, and let U \subseteq \mathbb{R}^{E(G)} be the linear subspace with \mathrm{aff}(\mathcal{F})=y+U for y \in \mathcal{F}. Since \textbf{0} \notin \mathcal{F}, we know that \mathcal{F} is a proper face of \mathrm{TECSP}(G). We will construct \vert \mathrm{CP}(G) \vert -1 many linearly independent vectors contained in U, which implies that \mathcal{F} is a facet. For i \in [r], we consider the vectors X_i \coloneqq \{\chi^H \mid H \subseteq G_i, \, e_i \in E(H)\}. After projecting them onto the coordinates E(G_i), these are exactly the vertices of the face \mathcal{F}_{e_i} \subseteq \mathrm{TECSP}(G_i) of \mathrm{TECSP}(G_i) defined by x_{e_i} \le 1, . By the proof of Theorem 3.2, this implies that there are \vert \mathrm{CP}(G_i) \vert -1 many linearly independent vectors of the form \chi_1-\chi_2, with \chi_1, \chi_2 \in X_i, and we have \chi_1-\chi_2 \in U. We remark that \operatorname{\mathrm{supp}}(\chi_1-\chi_2) \subseteq E(G_i)\setminus \{e_i\}. Since we can construct this kind of vectors for any i \in [r], and the vectors corresponding to different components G_i \neq G_j have disjoint support, all the considered vectors are linearly independent. In total we have \vert \mathrm{CP}(G_1) \vert +\ldots+\vert \mathrm{CP}(G_r)\vert -r many vectors so far. Now, let C' \in \mathrm{CP}(G_i) be a coparallel class of G_i, that is not a coparallel class of G. We know by Observation 2.4 that C'=C_1 \dot{\cup} C_2, where C_1 and C_2 are coparallel classes of G_i. Let G^{(1)}_1,\ldots,G^{(1)}_{\ell_1} (G^{(2)}_1,\ldots,G^{(2)}_{\ell_2}) denote the connected components of G_i-C' that contain a vertex of an edge of C_1 (C_2, respectively). We remark that there are one or two components that contain a vertex of both an edge of C_1 and an edge of C_2; they are thus named twice. For j=1,2, we define the subgraph H_j containing (a) all connected components of G-C except G_i, (b) the components G^{(j)}_1,\ldots,G^{(j)}_{\ell_j}, and (c) all edges of C_j. Then H_1 and H_2 are 2-edge-connected subgraphs and at least one of them contains the edge e_i. Hence, \chi^{H_1} or \chi^{H_2} is an incidence vector \chi contained in \mathcal{F}, and \chi^G-\chi \in U. Let e \in C_1, and e' \in C_2, then all the so far considered vectors are contained in the linear subspace U' \coloneqq \{x \in \mathbb{R}^{E(G)} \mid x_e=x_{e'}\}, but \chi^G-\chi \notin U', which shows that \chi^G-\chi is linearly independent to all the previously considered vectors.
We can do this for every coparallel class C' \in \mathrm{CP}(G_1) \cup \ldots \cup \mathrm{CP}(G_r) that is not a coparallel class of G, and all newly constructed vectors are linearly independent to all the previously considered vectors by the same argument. Thus, there are \vert \mathrm{CP}(G) \vert-r-1 many affinely independent vectors contained in \mathcal{F}, since \vert \mathrm{CP}(G) \vert=\vert \mathrm{CP}_{G_1}(G) \vert + \ldots +\vert \mathrm{CP}_{G_r}(G)\vert+1. Furthermore, we have Y\coloneqq \{\chi^G-\chi^{G_i} \mid i \in [r]\} \subseteq U. These vectors are linearly independent, as one can see easily by projecting them onto the coordinates \{e_1,\ldots,e_r\}. All the so far considered vectors are contained in the linear subspace \{x \in \mathbb{R}^{E(G)}\mid x_{e_1}=\ldots=x_{e_r}=0\} which is not true for the vectors in Y. This shows that \dim(U) \ge \vert \mathrm{CP}(G) \vert -1, which finishes the proof.
Let n \in \mathbb{N}, n \ge 4.
Let n be even and v \in V(K_n). Then the odd star inequality
defines a facet of \mathrm{TECSP}(K_n).
Let n be odd, v \in V(K_n), and h=\{w_1,w_2\} \in E(K_n-v). Let f \coloneqq \{v,w_1\}. Then the odd star inequality
defines a facet of \mathrm{TECSP}(K_n).
Since K_n has neither 1- nor 2-cuts for n \ge 4, \mathrm{TECSP}(K_n) is full dimensional. Let {\mathcal{F} \subseteq \mathrm{TECSP}(K_n)} be the face defined by the considered odd star inequality and let {U \subseteq \mathbb{R}^{E(K_n)}} be the linear subspace with \mathrm{aff}(\mathcal{F})=y+U for y \in \mathcal{F}. We have to show that \dim(U)=\vert E(K_n) \vert -1. Since \chi^{K_n} \notin \mathcal{F}, the considered odd star inequality defines a proper face of \mathrm{TECSP}(G), and we have \dim(U) \le \vert E(K_n) \vert-1.
Case (i). Suppose n is even. In the following, we construct two different kind of vectors contained in U that span a subspace of dimension at least \vert E(K_n) \vert-1. Let {f_1 \neq f_2 \in E(K_n-v)} with f_1 \cap f_2=\{w\} for some w \in V(K_n) \setminus \{v\}. Let u_1, \ldots, u_{n-2} be the vertices V(K_n)\setminus\{v,w\}. Let F' \coloneqq \{\{u_1,u_2\},\{u_3,u_4\},\ldots,\{u_{n-3},u_{n-2}\}\} \subseteq E(K_n). The three subgraphs
are 2-edge-connected. Furthermore, we have
which shows \chi^H \in \mathcal{F}. Since H_1 and H_2 both contain one edge more of \delta(\{v\}) and E(K_n-v) than H, we also have \chi^{H_1},\chi^{H_2} \in \mathcal{F}. Hence, \chi^{\{f_1\}}-\chi^{\{f_2\}}=\chi^{H_1}-\chi^{H_2} \in U for any two adjacent edges f_1,f_2 \in E(K_n-v), which leads to the first type of vectors we are considering: Let f'_1,\ldots,f'_k, k\coloneqq \vert E(K_n) \vert-n+1, denote the edges of K_n-v, and we assume w.l.o.g. that the labeling is chosen such that f'_i and f'_{i+1} are adjacent for all i \in [k-1] (which is trivial to exist). We have S_1 \coloneqq \{\chi^{\{f'_i\}}-\chi^{\{f'_{i+1}\}} \mid i \in [k-1]\} \subseteq U. Since all vectors contained in S_1 are linearly independent, S_1 spans a linear subspace U_1 \subseteq U of dimension k-1=\vert E(K_n) \vert-n. We remark that U_1 \subseteq \{x \in \mathbb{R}^{E(K_n)} \mid x_{\{v,w\}}=0 \text{ for } w \in V(K_n) \setminus \{v\}\}.
We also have \chi^{\{\{v,w\}\}}+\chi^{\{f_1\}}=\chi^{H_1}-\chi^{H} \in U, which yields the second type of vectors we are considering. For each w \in V(K_n) \setminus \{v\} pick one vector y^{(w)} \in \mathbb{R}^{E(K_n)} of that type with y^{(w)}_{\{v,w\}}=1, i.e., y^{(w)}=\chi^{\{\{v,w\}\}}+\chi^{\{f'\}} for some f' \in E(K_n-v), and let U_2 \subseteq U be the span of the vectors S_2 \coloneqq \{y^{(w)} \mid{w \in V(K_n)\setminus \{v\}}\}. Since all the vectors of S_2 are linearly independent, we have \dim(U_2) =n-1. Since the only vector in \chi \in U_2 with \chi_{\{v,w\}}=0 for all w \in V(K_n)\setminus \{v\} is the all zeroes vector \textbf{0}, we see that U_1 \cap U_2=\{\textbf{0}\}. This implies \dim(U) \ge \dim(U_1) + \dim(U_2)=\vert E(K_n)\vert-n+n-1=\vert E(K_n)\vert-1, and this shows (i).
Case (ii). Suppose now that n is odd. As in the proof of (i), we again construct sufficiently many linearly independent vectors contained in U.
The first class of vectors is constructed as follows. The graph K \coloneqq K_n-w_1 \subseteq K_n is a complete graph on n-1 many vertices. As already mentioned above, we can consider the odd star inequality for K:
Let \mathcal{F}' \subseteq \mathrm{TECSP}(K) denote the corresponding face. Then it is easy to see that for x \in \mathcal{F}', we have \tilde{x} \in \mathcal{F}, where \tilde{x} is x lifted to \mathbb{R}^{E(K_n)} by adding zeroes at the missing coordinates. By the proof of (i) w.r.t. K, this implies that for each pair of adjacent edges f_1 \neq f_2 \in E(K_n-v-w_1) we have \chi^{\{f_1\}}-\chi^{\{f_2\}} \in U, and for each vertex w \in V(K_n) \setminus \{v,w_1\} we have \chi^{\{\{v,w\}\}}+\chi^{\{f'\}} \in U for some f' \in E(K_n-v-w_1). Again by the proof of (i), the span of these vectors, denoted by U'_1 \subseteq U, has dimension \vert E(K_n-w_1) \vert-1=\vert E(K_n)\vert-n. We remark that U'_1 \subseteq \{x \in \mathbb{R}^{E(K_n)} \mid x_{f'}=0 \text{ for all } f' \in \delta(\{w_1\})\}.
For the second type of vectors, let f'=\{w_1,w\} \in \delta(\{w_1\})\setminus\{f,h\} for some w \in V(K_n) \setminus \{v,w_1,w_2\}. Let u_1,\ldots,u_{n-4} be the vertices V(K_n) \setminus \{v,w_1,w_2,w\} and let H_1^{(w)} \subseteq K_n be the 2-edge-connected subgraph with vertex set V(H_1^{(w)})=V(K_n)\setminus\{w_1,u_{n-4}\} and edge set
Consider the graph H_2^{(w)} with vertices V(H_2^{(w)})=V(H_1^{(w)}) \cup \{w_1\} and edges of E(H_2^{(w)})=E(H_1^{(w)})\setminus \{\{w_2,w\}\}\cup\{h,f'\}. It is immediate that \chi^{H_1^{(w)}},\chi^{H_2^{(w)}} \in \mathcal{F}. This means {\chi^{H_2^{(w)}}-\chi^{H_1^{(w)}}} {=\chi^{\{f'\}}+\chi^{\{h\}}-\chi^{\{\{w_2,w\}\}} \in U} for all f'=\{w_1,w\}\in \delta(\{w_1\})\setminus \{f,h\}. Hence, S'_2 \coloneqq \{\chi^{\{f'\}}+\chi^{\{h\}}-\chi^{\{\{w_2,w\}\}} \mid f'=\{w_1,w\} \in \delta(\{w_1\}) \setminus \{f,h\}\} \subseteq U. Projecting the vectors in S'_2 onto the coordinates \delta(\{w_1\})\setminus \{f,h\} yields the standard unit vectors of \mathbb{R}^{\delta(\{w_1\})\setminus \{f,h\}}. Hence, S'_2 spans an (n-3)-dimensional linear subspace U'_2. Clearly, the only vector \chi \in U'_2 with \chi_{\bar{f}}=0 for all \bar{f} \in \delta(\{w_1\}) is the all-zeros vector \textbf{0}, which means U'_1 \cap U'_2=\{\textbf{0}\}. This implies \dim(U) \ge \dim(U'_1)+\dim(U'_2) = \vert E(K_n) \vert -3.
For the last two missing vectors, let u_1,\ldots,u_{n-3} denote the vertices V(K_n)\setminus\{v,w_1,w_2\}. Let H_1' \subseteq K_n be the 2-edge-connected subgraph with vertices V(H_1')=V(K_n) \setminus \{w_1,w_2\} and edges E(H_1')=\delta(\{v\})\setminus\{f,\{v,w_2\}\}\cup\{\{u_1,u_2\},\ldots, \{u_{n-4},u_{n-3}\}\}. Let the subgraph H_2' contain all vertices V(K_n), all edges of H_1', and additionally the edges f, h, and \{v,w_2\}. H_2' is 2-edge-connected, too, and we have \chi^{H_1'},\chi^{H_2'} \in \mathcal{F}. Hence, \chi' \coloneqq \chi^{H_2'}-\chi^{H_1'}=\chi^{\{f\}}+\chi^{\{h\}}+\chi^{\{\{v,w_2\}\}} \in U. Since U'_1,U'_2 \subseteq \{x \in \mathbb{R}^{E(G)} \mid x_f=0\}, but \chi'_f=1, the dimension of U is at least \vert E(K_n) \vert -2.
For the last vector, let H_3' \subseteq K_n be the 2-edge-connected subgraph with vertex set V(H_3')=V(K_n) and edge set
Furthermore, let H_4' be the subgraph, with vertex set V(H_4')=V(K_n)\setminus\{w_1,w_2\}, and edge set
It is easy to check that \chi^{H_3'},\chi^{H_4'} \in \mathcal{F}. Hence, \chi^{H_3'}-\chi^{H_4'}=\chi^{\{h\}}+\chi^{\{\{w_1,u_1\}\}}+\chi^{\{\{w_1,u_2\}\}}+\chi^{\{v,w_2\}}-\chi^{\{\{u_1,u_2\}\}} \in U. All of the previously considered vectors are contained in the linear subspace \{x \in \mathbb{R}^{E(G)} \mid x_h-x(\delta(\{w_1\}) \setminus \{h\})=0\}, which is not true for \chi^{H_3'}-\chi^{H_4'}. Hence, we have \dim(U) \ge \vert E(K_n)\vert -1, which finishes the proof.
What is a complete facet-description of \mathrm{TECSP}(G) for any 2-edge-connected graph G?
What are other classes of supporting hyperplanes for the \mathrm{TECSP}(G), and when are they facet-defining?
For which graphs G is \mathrm{TECSP}(G) completely described by the so far studied inequalities?
Let G be a graph, let E_3 \subseteq E(G) be the edges not contained in a cut of size at most 3 and let E' \subseteq E(G) be a maximal set of edges not containing bridges or 2-cuts. Then \mathrm{ESP}(G) is given by
x_e=0 for each bridge e \in E(G),
x_e-x_f=0 for each minimal cut \{e,f\} of size two,
0 \le x_e \le 1 for each e \in E(G) \setminus E_3 that is not a bridge,
x(F)-x(C\setminus F) \le \vert F \vert -1 for each minimal cut C \subseteq E' without chord, and each F \subseteq C, \vert F \vert odd.
Moreover, the system above is nonredundant.
Let \mathcal{M}'=(E,\mathcal{C}) be a binary matroid without coloops and without coparallel elements. If e \in E does not belong to a cotriangle, i.e., a cocircuit with three elements, then the inequalities
define facets for \mathrm{Cyc}(\mathcal{M}').
If the binary matroid \mathcal{M} has no F_7^\ast minor, and C=\{e_1,e_2,\ldots,e_k\}, k \ge 3, is a cocircuit without chord, then the inequality
defines a facet of \mathrm{Cyc}(\mathcal{M}).
If the binary matroid \mathcal{M} has no F_7^\ast minor, and C=\{e_1,e_2,\ldots,e_k\}, k \ge 3, is a cocircuit, then the inquality
defines a facet if and only if \pi(C) has no chord in \overline{\mathcal{M}}.
The text on this page is from arXiv:2410.18564v1 by Justus Bruckamp, Markus Chimani and Martina Juhnke, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.
The record of each item is in analysis/objects/2410/2410.18564.jsonl.