Abstract
The weighted \({\mathcal {T}}\)-free 2-matching problem is the following problem: given an undirected graph G, a weight function on its edge set, and a set \({\mathcal {T}}\) of triangles in G, find a maximum weight 2-matching containing no triangle in \({\mathcal {T}}\). When \({\mathcal {T}}\) is the set of all triangles in G, this problem is known as the weighted triangle-free 2-matching problem, which is a long-standing open problem. A main contribution of this paper is to give the first polynomial-time algorithm for the weighted \({\mathcal {T}}\)-free 2-matching problem under the assumption that \({\mathcal {T}}\) is a set of edge-disjoint triangles. In our algorithm, a key ingredient is to give an extended formulation representing the solution set, that is, we introduce new variables and represent the convex hull of the feasible solutions as a projection of another polytope in a higher dimensional space. Although our extended formulation has exponentially many inequalities, we show that the separation problem can be solved in polynomial time, which leads to a polynomial-time algorithm for the weighted \({\mathcal {T}}\)-free 2-matching problem.
Similar content being viewed by others
Notes
Although such an edge set is often called a simple 2-matching in the literature, we call it a 2-matching to simplify the description.
References
Babenko, Maxim A.: Improved algorithms for even factors and square-free simple \(b\)-matchings. Algorithmica 64(3), 362–383 (2012). https://doi.org/10.1007/s00453-012-9642-6
Bérczi, Kristóf.: The triangle-free \(2\)-matching polytope of subcubic graphs. Technical Report TR-2012-2, Egerváry Research Group (2012)
Bérczi, Kristóf, Yusuke, K.: An algorithm for \((n-3)\)-connectivity augmentation problem: jump system approach. J. Comb. Theory Ser. B 102(3), 565–587 (2012). https://doi.org/10.1016/j.jctb.2011.08.007
Bérczi, K., Végh, L.A.: Restricted \(b\)-matchings in degree-bounded graphs. In: Integer Programming and Combinatorial Optimization, pp. 43–56. Springer, Berlin (2010). https://doi.org/10.1007/978-3-642-13036-6_4
Conforti, M., Cornuéjols, G., Zambelli, Giacomo: Extended formulations in combinatorial optimization. 4OR 8(1), 1–48 (2010). https://doi.org/10.1007/s10288-010-0122-z
Cornuéjols, G., Pulleyblank, W.: A matching problem with side conditions. Discrete Math. 29(2), 135–159 (1980). https://doi.org/10.1016/0012-365x(80)90002-3
Cornuejols, G., Pulleyblank, W.: Perfect triangle-free 2-matchings. In: Mathematical Programming Studies, pp. 1–7. (1980). https://doi.org/10.1007/bfb0120901
Cunningham, William H.: Matching, matroids, and extensions. Math. Program. 91(3), 515–542 (2002). https://doi.org/10.1007/s101070100256
Edmonds, Jack: Maximum matching and a polyhedron with \(0, 1\)-vertices. J. Res. Natl Bur. Stand. B 69, 125–130 (1965)
Frank, András: Restricted \(t\)-matchings in bipartite graphs. Discrete Appl. Math. 131(2), 337–346 (2003). https://doi.org/10.1016/s0166-218x(02)00461-4
Grötschel, M., Lovász, L., Schrijver, A.: Geometric Algorithms and Combinatorial Optimization, volume 2 of Algorithms and Combinatorics. Springer, New York (1988)
Hartvigsen, D.: Extensions of Matching Theory. PhD thesis, Carnegie Mellon University (1984)
Hartvigsen, D.: The square-free 2-factor problem in bipartite graphs. In: Integer Programming and Combinatorial Optimization, pp. 234–241. Springer, Berlin, Heidelberg (1999). https://doi.org/10.1007/3-540-48777-8_18
Hartvigsen, David: Finding maximum square-free 2-matchings in bipartite graphs. J. Comb. Theory Ser. B 96(5), 693–705 (2006). https://doi.org/10.1016/j.jctb.2006.01.004
Hartvigsen, David, Li, Y.: Polyhedron of triangle-free simple 2-matchings in subcubic graphs. Math. Program. 138(1–2), 43–82 (2012). https://doi.org/10.1007/s10107-012-0516-0
Iwata, S., Kobayashi, Y.: A weighted linear matroid parity algorithm. In Proceedings of the 49th Annual ACM SIGACT Symposium on Theory of Computing - STOC 2017, pp. 264–276. ACM Press (2017). https://doi.org/10.1145/3055399.3055436
Kaibel, V.: Extended formulations in combinatorial optimization. Technical report, arXiv:1104.1023, (2011)
Király, Z.: \({C}_4\)–free 2-factors in bipartite graphs. Technical Report TR-2012-2, Egerváry Research Group (1999)
Kobayashi, Y.: A simple algorithm for finding a maximum triangle-free \(2\)-matching in subcubic graphs. Discrete Optim. 7(4), 197–202 (2010). https://doi.org/10.1016/j.disopt.2010.04.001
Kobayashi, Y.: Triangle-free 2-matchings and M-concave functions on jump systems. Discrete Appl. Math. 175, 35–42 (2014). https://doi.org/10.1016/j.dam.2014.05.016
Kobayashi, Y.: Weighted triangle-free 2-matching problem with edge-disjoint forbidden triangles. In: Integer Programming and Combinatorial Optimization, pp. 280–293. Springer, New York (2020). https://doi.org/10.1007/978-3-030-45771-6_22
Kobayashi, Y., Szabó, J., Takazawa, K.: A proof of Cunningham’s conjecture on restricted subgraphs and jump systems. J. Comb. Theory Ser B 102(4), 948–966 (2012). https://doi.org/10.1016/j.jctb.2012.03.003
Letchford, A.N., Reinelt, G., Theis, D.O.: Odd minimum cut sets and \(b\)-matchings revisited. SIAM J. Discrete Math. 22(4), 1480–1487 (2008). https://doi.org/10.1137/060664793
Makai, M.: On maximum cost \({K}_{t, t}\)-free \(t\)-matchings of bipartite graphs. SIAM J. Discrete Math. 21(2), 349–360 (2007). https://doi.org/10.1137/060652282
Nam, Y.: Matching Theory: Subgraphs With Degree Constraints And Other Properties. PhD thesis, University of British Columbia (1994)
Padberg, M.W., Rao, M.R.: Odd minimum cut-sets and \(b\)-matchings. Math. Oper. Res. 7(1), 67–80 (1982)
Pap, G.: Combinatorial algorithms for matchings, even factors and square-free 2-factors. Math. Program. 110(1), 57–69 (2007). https://doi.org/10.1007/s10107-006-0053-9
Takazawa, K.: A weighted \({K}_{t, t}\)-free \(t\)-factor algorithm for bipartite graphs. Math. Oper. Res. 34(2), 351–362 (2009). https://doi.org/10.1287/moor.1080.0365
Takazawa, K.: Decomposition theorems for square-free 2-matchings in bipartite graphs. Discrete Appl. Math. 233, 215–223 (2017). https://doi.org/10.1016/j.dam.2017.07.035
Takazawa, K.: Excluded \(t\)-factors in bipartite graphs: a unified framework for nonbipartite matchings and restricted 2-matchings. In: Integer Programming and Combinatorial Optimization, pp. 430–441. Springer, New York (2017). https://doi.org/10.1007/978-3-319-59250-3_35
Takazawa, K.: Finding a maximum 2-matching excluding prescribed cycles in bipartite graphs. Discrete Optim. 26, 26–40 (2017). https://doi.org/10.1016/j.disopt.2017.05.003
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
A preliminary version of this paper appears in [21]. Supported by JSPS KAKENHI Grant Numbers JP16K16010, JP16H03118, JP18H05291, and JP20K11692, Japan.
Appendices
A Proof of Lemma 5
By symmetry, it suffices to consider \((G_1, b_1, {\mathcal {T}}_1)\). Since the tightness of (10) for \((S^*, F^*_0, F^*_1)\) implies that \(x_1(\delta _{G_1}(r)) = 1\), we can easily see that \((x_1, y_1)\) satisfies (1), (2), (4)–(7). In what follows, we consider (10) for \((x_1, y_1)\) in \((G_1, b_1, {\mathcal {T}}_1)\). For edge sets \(F'_0, F'_1 \subseteq E_1\), we denote \(g(F'_0, F'_1) = \sum _{e \in F'_0} x_1(e) + \sum _{e \in F'_1} (1-x_1(e))\) to simplify the notation. For \((S', F'_0, F'_1) \in {\mathcal {F}}_1\), let \(h(S', F'_0, F'_1)\) denote the left-hand side of (10). To derive a contradiction, let \((S', F'_0, F'_1) \in {\mathcal {F}}_1\) be a minimizer of \(h(S', F'_0, F'_1)\) and assume that \(h(S', F'_0, F'_1) < 1\). By changing the roles of \(S'\) and \(V' {\setminus } S'\) if necessary, we may assume that \(r \not \in S'\).
For \(T \in {\mathcal {T}}^+_{S^*}\), let \(v_1, v_2, v_3, \alpha , \beta \), and \(\gamma \) be as in Figs. 7–10. Let \(G'_T = (V'_T, E'_T)\) be the subgraph of \(G_1\) corresponding to T, that is, the subgraph induced by \(\{r, p_1, p_2, v_2, v_3\}\) (Fig. 7), \(\{r, p_3, v_1 \}\) (Fig. 8), \(\{r, p_1, p_2, p_3, v_2, v_3\}\) (Fig. 9), or \(\{r, p_4, v_1 \}\) (Fig. 10). Let \({\hat{S}} = S' \cap (V'_T {\setminus } \{v_1, v_2, v_3\})\), \({\hat{F}}_0 = F'_0 \cap E'_T\), and \({\hat{F}}_1 = F'_1 \cap E'_T\).
We show the following properties (P1)–(P9) in Sect. 1, and show that \((x_1, y_1)\) satisfies (10) by using these properties in Sect. 2.
-
(P1)
If T is of type (A) or (B) and \(v_2, v_3 \not \in S'\), then \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even.
-
(P2)
If T is of type (A), \(v_2, v_3 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even, then \(g({\hat{F}}_0, {\hat{F}}_1) \ge \min \{x(\alpha )+x(\gamma ), 2-x(\alpha )-x(\gamma ) - 2 y_\beta \}\).
-
(P3)
If T is of type (B), \(v_2, v_3 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even, then \(g({\hat{F}}_0, {\hat{F}}_1) \ge y_{\alpha } + y_{\gamma } + y_{\alpha \beta } + y_{\beta \gamma }\).
-
(P4)
If T is of type (A) or (B), \(v_2, v_3 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is odd, then \(g({\hat{F}}_0, {\hat{F}}_1) \ge y_\emptyset + y_\beta + y_{\alpha \gamma }\).
-
(P5)
If T is of type (A) or (B), \(v_2 \in S'\), \(v_3 \not \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even, then \(g({\hat{F}}_0, {\hat{F}}_1) \ge \min \{ x(\alpha )+x(\beta ), 2 - x(\alpha ) - x(\beta ) - 2 y_{\gamma }\}\).
-
(P6)
If T is of type (A) or (B), \(v_2 \in S'\), \(v_3 \not \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is odd, then \(g({\hat{F}}_0, {\hat{F}}_1) \ge y_\emptyset + y_\gamma + y_{\alpha \beta }\).
-
(P7)
If T is of type (A\('\)) or type (B\('\)) and \(v_1 \not \in S'\), then \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even.
-
(P8)
If T is of type (A\('\)) or type (B\('\)), \(v_1 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even, then \(g({\hat{F}}_0, {\hat{F}}_1) = \min \{x(\alpha )+x(\gamma ), 2-x(\alpha )-x(\gamma ) - 2 y_\beta \}\).
-
(P9)
If T is of type (A\('\)) or type (B\('\)), \(v_1 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is odd, then \(g({\hat{F}}_0, {\hat{F}}_1) = y_\emptyset + y_\beta + y_{\alpha \gamma }\).
Note that each \(T \in {\mathcal {T}}^+_{S^*}\) satisfies exactly one of (P1)–(P9) by changing the labels of \(v_2\) and \(v_3\) if necessary.
1.1 A.1 Proofs of (P1)–(P9)
1.1.1 A.1.1 When T is of type (A)
We first consider the case when T is of type (A).
Proof of (P1). Suppose that T is of type (A) and \(v_2, v_3 \not \in S'\). If \(b_1({{\hat{S}}}) + |{{\hat{F}}}_1|\) is odd, then either \(p_1 \in {{\hat{S}}}\) and \(|{{\hat{F}}}_1 \cap \delta _{G_1}(p_1)|\) is even or \(p_2 \in {{\hat{S}}}\) and \(|{{\hat{F}}}_1 \cap \delta _{G_1}(p_2)|\) is even. In the former case, \(h(S', F'_0, F'_1) \ge \min \{x_1(e_1) + x_1(e_5), 2 - x_1 (e_1) - x_1(e_5) \} = 1\), which is a contradiction. The same argument can be applied to the latter case. Therefore, \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even.
Proof of (P2). Suppose that T is of type (A), \(v_2, v_3 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even. If \(p_1 \not \in S'\), then we define \((S'', F''_0, F''_1) \in {\mathcal {F}}_1\) as \((S'', F''_0, F''_1) = (S' \cup \{p_1\}, F'_0 {\setminus } \{e_5\}, F'_1\cup \{e_1\})\) if \(e_5 \in F'_0\) and \((S'', F''_0, F''_1) = (S' \cup \{p_1\}, F'_0 \cup \{e_1\}, F'_1 {\setminus } \{e_5\})\) if \(e_5 \in F'_1\). Since \(h(S'', F''_0, F''_1) = h(S', F'_0, F'_1)\) holds, by replacing \((S', F'_0, F'_1)\) with \((S'', F''_0, F''_1)\), we may assume that \(p_1 \in S'\). Similarly, we may assume that \(p_2 \in S'\), which implies that \({{\hat{S}}} = \{p_1, p_2\}\), \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_1, e_2, e_3, e_4\}\), and \(|{{\hat{F}}}_1|\) is even. Then, \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge \min \{ x(\alpha ) + x(\gamma ), 2-x(\alpha )-x(\gamma ) - 2 y_\beta \}\) by the following case analysis.
-
If \({{\hat{F}}}_1 = \emptyset \), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) = x_1(e_1) + x_1(e_2) + x_1(e_3) + x_1(e_4) = 2-x(\alpha )-x(\gamma ) - 2 y_\beta \).
-
If \(|{{\hat{F}}}_1| \ge 2\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge 2 - (x_1(e_1) + x_1(e_2) + x_1(e_3) + x_1(e_4)) = x(\alpha ) + x(\gamma ) + 2 y_\beta \ge x(\alpha ) + x(\gamma )\).
Proof of (P4). Suppose that T is of type (A), \(v_2, v_3 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is odd. In the same way as (P2), we may assume that \({{\hat{S}}} = \{p_1, p_2\}\), \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_1, e_2, e_3, e_4\}\), and \(|{{\hat{F}}}_1|\) is odd. Then, \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge y_\emptyset + y_\beta + y_{\alpha \gamma }\) by the following case analysis and by the symmetry of \(v_2\) and \(v_3\).
-
If \(|{{\hat{F}}}_1| = 3\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge 3 - (x_1(e_1) + x_1(e_2) + x_1(e_3) + x_1(e_4)) \ge 1 \ge y_\emptyset + y_\beta + y_{\alpha \gamma }\).
-
If \({{\hat{F}}}_1 = \{e_1\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge (1 - x_1(e_1)) + x_1(e_2) \ge y_\emptyset + y_\beta + y_{\alpha \gamma }\).
-
If \({{\hat{F}}}_1 = \{e_3\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge 1-x_1(e_3) \ge y_\emptyset + y_\beta + y_{\alpha \gamma }\).
Proof of (P5). Suppose that T is of type (A), \(v_2 \in S'\), \(v_3 \not \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even. In the same way as (P2), we may assume that \(p_1 \in S'\). If \(p_2 \in S'\), then \(b_1(p_2) + |F'_1 \cap \delta _{G_1}(p_2)|\) is even by the same calculation as (P1). Therefore, we may assume that \(p_2 \not \in S'\), since otherwise we can replace \((S', F'_0, F'_1)\) with \((S' {\setminus } \{p_2\}, F'_0 {\setminus } \delta _{G_1}(p_2), F'_1{\setminus } \delta _{G_1}(p_2))\) without increasing the value of \(h(S', F'_0, F'_1)\). That is, we may assume that \({{\hat{S}}} = \{p_1\}\), \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_1, e_3\}\), and \(|{{\hat{F}}}_1|\) is odd. Then, \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge \min \{(1 - x_1(e_1)) + x_1(e_3), x_1(e_1) + (1 - x_1(e_3)) \} = \min \{ x(\alpha )+x(\beta ), 2 - x(\alpha ) - x(\beta ) \} \ge \min \{ x(\alpha )+x(\beta ), 2 - x(\alpha ) - x(\beta ) - 2 y_{\gamma }\}\).
Proof of (P6). Suppose that T is of type (A), \(v_2 \in S'\), \(v_3 \not \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is odd. In the same way as (P5), we may assume that \({{\hat{S}}} = \{p_1\}\), \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_1, e_3\}\), and \(|{{\hat{F}}}_1|\) is even. Then, \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge \min \{ x_1(e_1) + x_1(e_3), 2 - x_1(e_1) - x_1(e_3) \} = \min \{ y_\emptyset + y_\gamma + y_{\alpha \beta }, 2 - (y_\emptyset + y_\gamma + y_{\alpha \beta })\} = y_\emptyset + y_\gamma + y_{\alpha \beta }\).
1.1.2 A.1.2 When T is of type (A\('\))
Second, we consider the case when T is of type (A\('\)).
Proof of (P7). Suppose that T is of type (A\('\)) and \(v_1 \not \in S'\). If \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is odd, then \({{\hat{S}}} =\{p_3\}\) and \(|{{\hat{F}}}_1|\) is odd. This shows that \(h(S', F'_0, F'_1) \ge g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge 1\) by the following case analysis, which is a contradiction.
-
If \({{\hat{F}}}_1 = \{e_1\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge (1-x_1(e_1)) + x_1(e_2) + x_1(e_9) \ge 1\). The same argument can be applied to the case of \({{\hat{F}}}_1 = \{e_2\}\) by the symmetry of \(\alpha \) and \(\gamma \).
-
If \({{\hat{F}}}_1 = \{e_i\}\) for some \(i \in \{3, 4, 8\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge (1 - x_1(e_i)) + x_1(e_9) \ge 1\).
-
If \({{\hat{F}}}_1 = \{e_9\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) = 1 + 2 y_\emptyset \ge 1\).
-
If \(|{{\hat{F}}}_1| \ge 3\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge 3 - (x_1(e_1) + x_1(e_2) + x_1(e_3) + x_1(e_4) + x_1(e_8) + x_1(e_9)) \ge 1\).
Therefore, \(b_1({{\hat{S}}}) + |{{\hat{F}}}_1|\) is even.
Proof of (P8). Suppose that T is of type (A\('\)), \(v_1 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even. Then, \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge \min \{ x(\alpha ) + x(\gamma ), 2-x(\alpha )-x(\gamma ) - 2 y_\beta \}\) by the following case analysis.
-
If \({{\hat{F}}}_0 = \{e_8, e_9\}\) and \({{\hat{F}}}_1 = \emptyset \), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) = x_1(e_8) + x_1(e_9) = x(\alpha ) + x(\gamma )\).
-
If \({{\hat{F}}}_0 = \emptyset \) and \({{\hat{F}}}_1 = \{e_8, e_9\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) = (1 - x_1(e_8)) + (1-x_1(e_9)) = 2-x(\alpha )-x(\gamma ) \ge 2-x(\alpha )-x(\gamma ) - 2 y_\beta \).
-
If \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_1, e_2, e_3, e_4\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge \min \{ x(\alpha ) + x(\gamma ), 2-x(\alpha )-x(\gamma ) - 2 y_\beta \}\) by the same calculation as (P2) in Sect. 1.
Proof of (P9). Suppose that T is of type (A\('\)), \(v_1 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is odd. Then, \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge y_\emptyset + y_\beta + y_{\alpha \gamma }\) by the following case analysis.
-
If \({{\hat{F}}}_0 = \{e_8 \}\) and \({{\hat{F}}}_1 = \{e_9 \}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) = x_1(e_8) + (1-x_1(e_9)) = y_\emptyset + y_\beta + y_{\alpha \gamma }\).
-
If \({{\hat{F}}}_0 = \{e_9 \}\) and \({{\hat{F}}}_1 = \{e_8 \}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) = (1 - x_1(e_8)) + x_1(e_9) \ge 1 \ge y_\emptyset + y_\beta + y_{\alpha \gamma }\).
-
If \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_1, e_2, e_3, e_4\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge y_\emptyset + y_\beta + y_{\alpha \gamma }\) by the same calculation as (P4) in Sect. 1.
1.1.3 A.1.3 When T is of type (B)
Third, we consider the case when T is of type (B). Let \(G^+=(V^+, E^+)\) be the graph obtained from \(G'_T = (V'_T, E'_T)\) in Fig. 9 by adding a new vertex \(r^*\), edges \(e_{11}= r r^*\), \(e_{12}= v_2 r^*\), \(e_{13}= v_3 r^*\), and self-loops \(e_{14}, e_{15}, e_{16}\) that are incident to \(v_2\), \(v_3\), and \(r^*\), respectively (Fig. 12). We define \(b_T: V^+ \rightarrow {\mathbf {Z}}_{\ge 0}\) as \(b_T(v) = 1\) for \(v \in \{r, p_1, p_2, p_3\}\) and \(b_T(v) = 2\) for \(v \in \{r^*, v_2, v_3\}\). We also define \(x_T: E^+ \rightarrow {\mathbf {Z}}_{\ge 0}\) as \(x_T(e) = x_1(e)\) for \(e \in E'_T\) and \(x_T(e_{11}) = y_{\alpha } + y_{\gamma } + y_{\alpha \beta } + y_{\beta \gamma }\), \(x_T(e_{12}) = y_{\alpha } + y_{\beta } + y_{\alpha \gamma } + y_{\beta \gamma }\), \(x_T(e_{13}) = y_{\beta } + y_{\gamma } + y_{\alpha \beta } + y_{\alpha \gamma }\), \(x_T(e_{14}) = y_{\emptyset } + y_{\gamma }\), \(x_T(e_{15}) = y_{\emptyset } + y_{\alpha }\), and \(x_T(e_{16}) = y_{\emptyset }\). For \(J \in {\mathcal {E}}_T\), define \(b_T\)-factors \(M_J\) in \(G^+\) as follows:
Then, we obtain \(\sum _{J \in {\mathcal {E}}_T} y_1(J) =1\) and \(\sum _{J \in {\mathcal {E}}_T} y_1(J) x_{M_J} = x_T\), where \(x_{M_J} \in {\mathbf {R}}^{E^+}\) is the characteristic vector of \(M_J\). This shows that \(x_T\) is in the \(b_T\)-factor polytope in \(G^+\). Therefore, \(x_T\) satisfies (3) with respect to \(G^+\) and \(b_T\). We now show (P1), (P3), (P4), (P5), and (P6).
Proof of (P1). Suppose that T is of type (B) and \(v_2, v_3 \not \in S'\). If \(b_1({{\hat{S}}}) + |{{\hat{F}}}_1|\) is odd, then \(b_T({{\hat{S}}}) + |{{\hat{F}}}_1|\) is also odd. Since \(x_T\) satisfies (3) with respect to \(G^+\) and \(b_T\), we obtain \(g({\hat{F}}_0, {\hat{F}}_1) \ge 1\). This shows that \(h(S', F'_0, F'_1) \ge 1\), which is a contradiction. Therefore, \(b_1({{\hat{S}}}) + |{{\hat{F}}}_1|\) is even.
Proof of (P3). Suppose that T is of type (B), \(v_2, v_3 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even. Since \(b_T({{\hat{S}}} \cup \{r^*, v_2, v_3\}) + |{{\hat{F}}}_1 \cup \{e_{11}\}|\) is odd and \(x_T\) satisfies (3), we obtain \(g({{\hat{F}}}_0, {{\hat{F}}}_1) + (1-x_T(e_{11})) \ge 1\). Therefore, \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge x_T(e_{11}) = y_\alpha + y_\gamma + y_{\alpha \beta } + y_{\beta \gamma }\).
Proof of (P4). Suppose that T is of type (B), \(v_2, v_3 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is odd. Since \(b_T({{\hat{S}}} \cup \{r^*, v_2, v_3\}) + |{{\hat{F}}}_1|\) is odd and \(x_T\) satisfies (3), we obtain \(g({{\hat{F}}}_0, {{\hat{F}}}_1) + x_T(e_{11}) \ge 1\). Therefore, \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge 1 - x_T(e_{11}) = y_\emptyset + y_\beta + y_{\alpha \gamma }\).
Proof of (P5). Suppose that T is of type (B), \(v_2 \in S'\), \(v_3 \not \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even. If \({{\hat{S}}} \cap \{p_1, p_3\} \not = \emptyset \) and \(p_2 \not \in {{\hat{S}}}\), then we can add \(p_2\) to \(S'\) without decreasing the value of \(h(S', F'_0, F'_1)\). Therefore, we can show \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge \{x(\alpha ) + x(\beta ), 2 - x(\alpha ) - x(\beta ) - 2 y_\gamma \}\) by the following case analysis.
-
Suppose that \({{\hat{S}}} = \{p_1, p_2, p_3\}\), which implies that \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_1, e_2, e_6, e_9\}\) and \(|{{\hat{F}}}_1|\) is odd.
-
If \({{\hat{F}}}_1 = \{e_i\}\) for \(i \in \{1, 2, 6\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge (1- x_1(e_i)) + x_1(e_9) \ge x(\alpha ) + x(\beta )\).
-
If \({{\hat{F}}}_1 = \{e_9\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) = y_\alpha + y_\beta + y_{\alpha \gamma } + y_{\beta \gamma } + 2 y_\emptyset = 2 - x(\alpha ) - x(\beta ) - 2 y_\gamma \).
-
If \(|{{\hat{F}}}_1| = 3\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge 3 - (x_1(e_1) + x_1(e_2) + x_1(e_6) + x_1(e_9)) \ge x(\alpha ) + x(\beta )\).
-
-
Suppose that \({{\hat{S}}} = \{p_1, p_2\}\), which implies that \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_1, e_2, e_4, e_6, e_7\}\) and \(|{{\hat{F}}}_1|\) is even.
-
If \({{\hat{F}}}_1 = \emptyset \), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) = x_1(e_1) + x_1(e_2) + x_1(e_4) + x_1(e_6) + x_1(e_7) = 2 - x(\alpha ) - x(\beta ) - 2 y_\gamma \).
-
If \(|{{\hat{F}}}_1| \ge 2\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge 2 - (x_1(e_1) + x_1(e_2) + x_1(e_4) + x_1(e_6) + x_1(e_7)) \ge x(\alpha ) + x(\beta )\).
-
-
Suppose that \({{\hat{S}}} = \{p_2\}\), which implies that \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_3, e_5, e_7\}\) and \(|{{\hat{F}}}_1|\) is odd.
-
If \({{\hat{F}}}_1 = \{e_i\}\) for \(i \in \{3, 7\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge (1- x_1(e_i)) + x_1(e_5) \ge x(\alpha ) + x(\beta )\).
-
If \({{\hat{F}}}_1 = \{e_5\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge (1-x_1(e_5)) + x_1(e_7) \ge 2 - x(\alpha ) - x(\beta ) - 2 y_\gamma \).
-
If \({{\hat{F}}}_1 = \{e_3, e_5, e_7\}\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) = 3 - (x_1(e_3) + x_1(e_5) + x_1(e_7)) \ge x(\alpha ) + x(\beta )\).
-
-
Suppose that \({{\hat{S}}} = \{p_2, p_3\}\), which implies that \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_3, e_4, e_5, e_9\}\) and \(|{{\hat{F}}}_1|\) is even.
-
If \({{\hat{F}}}_1 = \emptyset \), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) = x_1(e_3) + x_1(e_4) + x_1(e_5) + x_1(e_9) = x(\alpha ) + x(\beta ) + 2 y_{\gamma } \ge x(\alpha ) + x(\beta )\).
-
If \(|{{\hat{F}}}_1| \ge 2\), then \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge 2 - (x_1(e_3) + x_1(e_4) + x_1(e_5) + x_1(e_9)) = 2- x(\alpha ) - x(\beta ) - 2 y_{\gamma }\).
-
-
If \({{\hat{S}}} = \emptyset \), then \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_5, e_8\}\) and \(|{{\hat{F}}}_1|\) is even. Therefore, \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge \min \{x_1(e_5) + x_1(e_8), 2-x_1(e_5)-x_1(e_8) \} \ge \min \{x(\alpha ) + x(\beta ), 2 - x(\alpha ) - x(\beta ) - 2 y_\gamma \}\).
Proof of (P6). Suppose that T is of type (B), \(v_2 \in S'\), \(v_3 \not \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is odd. Since \(b_T({{\hat{S}}} \cup \{v_2 \}) + |{{\hat{F}}}_1|\) is odd and \(x_T\) satisfies (3), we obtain \(g({{\hat{F}}}_0, {{\hat{F}}}_1) + x_T(e_{12}) \ge 1\). Therefore, \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge 1 - x_T(e_{12}) = y_\emptyset + y_\gamma + y_{\alpha \beta }\).
1.1.4 A.1.4 When T is of type (B\('\))
Finally, we consider the case when T is of type (B\('\)).
Proof of (P7). Suppose that T is of type (B\('\)) and \(v_1 \not \in S'\). If \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is odd, then \({{\hat{S}}} =\{p_4\}\) and \(h(S', F'_0, F'_1) \ge \min \{x_1(e_1) + x_1(e_{10}), 2 - x_1(e_1) - x_1(e_{10})\} = 1\), which is a contradiction. Therefore, \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even.
Proof of (P8). Suppose that T is of type (B\('\)), \(v_1 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even. If \(p_4 \not \in S'\), then we define \((S'', F''_0, F''_1) \in {\mathcal {F}}_1\) as \((S'', F''_0, F''_1) = (S' \cup \{p_4\}, F'_0 {\setminus } \{e_{10}\}, F'_1\cup \{e_1\})\) if \(e_{10} \in F'_0\) and \((S'', F''_0, F''_1) = (S' \cup \{p_4\}, F'_0 \cup \{e_1\}, F'_1 {\setminus } \{e_{10}\})\) if \(e_{10} \in F'_1\). Since \(h(S'', F''_0, F''_1) = h(S', F'_0, F'_1)\), by replacing \((S', F'_0, F'_1)\) with \((S'', F''_0, F''_1)\), we may assume that \(p_4 \in S'\). Then, since \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_1, e_2\}\) and \(|{{\hat{F}}}_1|\) is odd, we obtain \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge \min \{ (1-x_1(e_1)) + x_1(e_2), x_1(e_1) + (1-x_1(e_2)) \} \ge \min \{ x(\alpha ) + x(\gamma ), 2 - x(\alpha ) - x(\gamma ) - 2y_\beta \}\).
Proof of (P9). Suppose that T is of type (B\('\)), \(v_1 \in S'\), and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is odd. In the same way as (P8), we may assume that \({{\hat{S}}} = \{p_4\}\), \({{\hat{F}}}_0 \cup {{\hat{F}}}_1 = \{e_1, e_2\}\), and \(|{{\hat{F}}}_1|\) is even. Then, \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge \min \{ x_1(e_1) + x_1(e_2), (1-x_1(e_1)) + (1-x_1(e_2)) \} = \min \{ y_\emptyset + y_\beta + y_{\alpha \gamma }, 2 - (y_\emptyset + y_\beta + y_{\alpha \gamma })\} = y_\emptyset + y_\beta + y_{\alpha \gamma }\).
1.2 A.2 Condition (10)
Recall that \(r \not \in S'\) is assumed and note that \(x_1(\delta _{G_1}(r)) = 1\). Let \({\mathcal {T}}_{(P3)} \subseteq {\mathcal {T}}^+_{S^*}\) be the set of triangles satisfying the conditions in (P3), i.e., the set of triangles of type (B) such that \(v_2, v_3 \in S'\) and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is even. Since \(y_{\alpha } + y_{\gamma } + y_{\alpha \beta } + y_{\beta \gamma } = 1 - x_1(e^{T}_1) - x_1(e^{T}_2)\) holds for each triangle \(T \in {\mathcal {T}}^+_{S^*}\) of type (B), if there exist two triangles \(T, T' \in {\mathcal {T}}_{(P3)}\), then \(h(S', F'_0, F'_1) \ge (1- x_1(e^{T}_1) - x_1(e^{T}_2)) + (1- x_1(e^{T'}_1) - x_1(e^{T'}_2)) \ge 2 - x_1(\delta _{G_1}(r)) = 1\), which is a contradiction. Similarly, if there exists a triangle \(T \in {\mathcal {T}}_{(P3)}\) and an edge \(e \in (\delta _{G_1}(r) {\setminus } E'_{T}) \cap F'_1\), then \(h(S', F'_0, F'_1) \ge (1- x_1(e^{T}_1) - x_1(e^{T}_2)) + (1- x_1(e)) \ge 2 - x_1(\delta _{G_1}(r)) = 1\), which is a contradiction. Therefore, either \({\mathcal {T}}_{(P3)} = \emptyset \) holds or \({\mathcal {T}}_{(P3)}\) consists of exactly one triangle, say T, and \((\delta _{G_1}(r) {\setminus } E'_{T}) \cap F'_1 = \emptyset \).
Assume that \({\mathcal {T}}_{(P3)} = \{T\}\) and \((\delta _{G_1}(r) {\setminus } E'_{T}) \cap F'_1 = \emptyset \). Define \((S'', F''_0, F''_1) \in {\mathcal {F}}_1\) as \(S'' = S' \cup V'_{T}\), \(F''_0 = (F'_0 \triangle \delta _{G_1}(r)) {\setminus } E'_{T}\), and \(F''_1 = F'_1 {\setminus } E'_{T}\), where \(\triangle \) denotes the symmetric difference. Note that \((F''_0, F''_1)\) is a partition of \(\delta _{G_1} (S'')\), \(b_1(S'') + |F''_1| = (b_1 (S') + b_1({\hat{S}}) ) + (|F'_1| - |{{\hat{F}}}_1|) \equiv 1 \pmod {2}\), and \(h(S', F'_0, F'_1) - h(S'', F''_0, F''_1) \ge (1 - x_1(e^{T}_1) - x_1(e^{T}_2)) - x_1( \delta _{G_1}(r) {\setminus } \{ x_1(e^{T}_1), x_1(e^{T}_2)\}) = 0\). By these observations, \((S'', F''_0, F''_1) \in {\mathcal {F}}_1\) is also a minimizer of h. This shows that \((V'' {\setminus } S'', F''_0, F''_1) \in {\mathcal {F}}_1\) is a minimizer of h such that \(r \in V'' {\setminus } S''\). Furthermore, if a triangle \(T' \in {\mathcal {T}}^+_{S^*}\) satisfies the conditions in (P3) with respect to \((V'' {\setminus } S'', F''_0, F''_1)\), then \(T'\) is a triangle of type (B) such that \(v_2, v_3 \not \in S'\) and \(b_1({\hat{S}}) + |{{\hat{F}}}_1|\) is odd with respect to \((S', F'_0, F'_1)\), which contradicts (P1). Therefore, by replacing \((S', F'_0, F'_1)\) with \((V'' {\setminus } S'', F''_0, F''_1)\), we may assume that \({\mathcal {T}}_{(P3)} = \emptyset \).
In what follows, we construct \((S, F_0, F_1) \in {\mathcal {F}}\) for which (x, y) violates (10) to derive a contradiction. We initialize \((S, F_0, F_1)\) as \(S = S' \cap V\), \(F_0 = F'_0 \cap E\), and \(F_1 = F'_1 \cap E\), and apply the following procedures for each triangle \(T \in {\mathcal {T}}^+_{S^*}\).
-
Suppose that T satisfies the condition in (P1) or (P7). In this case, we do nothing.
-
Suppose that T satisfies the condition in (P2) or (P8). If \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge x(\alpha ) + x(\gamma )\), then add \(\alpha \) and \(\gamma \) to \(F_0\). Otherwise, since \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge 2-x(\alpha )-x(\gamma ) - 2 y_\beta \), add \(\alpha \) and \(\gamma \) to \(F_1\).
-
Suppose that T satisfies the condition in (P4) or (P9). In this case, add \(\alpha \) to \(F_0\) and add \(\gamma \) to \(F_1\).
-
Suppose that T satisfies the condition in (P5). If \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge x(\alpha ) + x(\beta )\), then add \(\alpha \) and \(\beta \) to \(F_0\). Otherwise, since \(g({{\hat{F}}}_0, {{\hat{F}}}_1) \ge 2-x(\alpha )-x(\beta ) - 2 y_\gamma \), add \(\alpha \) and \(\beta \) to \(F_1\).
-
Suppose that T satisfies the condition in (P6). In this case, add \(\alpha \) to \(F_0\) and add \(\beta \) to \(F_1\).
Note that exactly one of the above procedures is applied for each \(T \in {\mathcal {T}}^+_{S^*}\), because \({\mathcal {T}}_{(P3)} = \emptyset \).
Then, we see that \((S, F_0, F_1) \in {\mathcal {F}}\) holds and the left-hand side of (10) with respect to \((S, F_0, F_1)\) is at most \(h(S', F'_0, F'_1)\) by (P1)–(P9). Since \(h(S', F'_0, F'_1) < 1\) is assumed, (x, y) violates (10) for \((S, F_0, F_1) \in {\mathcal {F}}\), which is a contradiction. \(\square \)
B Proof of Lemma 6
We first show that \(M_1 \oplus M_2\) forms a \({\mathcal {T}}\)-free b-factor. We can easily see that replacing \((M_1 \cup M_2) \cap \{e^f \mid f \in {\tilde{F}}^*_0\}\) with \(\{ f \in {\tilde{F}}^*_0 \mid {e}^f \in M_1 \cap M_2 \}\) does not affect the degrees of vertices in V. Since \(M_1 \cup M_2\) contains exactly one of \(\{e^f_u, e^f_v\}\) or \({e}^f_r (= {e}^f_{r'})\) for \(f = uv \in {\tilde{F}}^*_1\), replacing \((M_1 \cup M_2) \cap \{e^f_u, e^f_r, e^f_v \mid f = uv \in {\tilde{F}}^*_1\}\) with \(\{ f \in {\tilde{F}}^*_1 \mid e^f_r \not \in M_1 \cap M_2 \}\) does not affect the degrees of vertices in V.
For every \(T \in {\mathcal {T}}^+_{S^*}\) of type (A) or (A\('\)), since \(|\varphi (M_1, M_2, T) \cap \{\alpha , \gamma \}| = |M_T \cap \{e_8, e_9\}|\), \(|\varphi (M_1, M_2, T) \cap \{\alpha , \beta \}| = |M_T \cap \{e_3, e_5\}|\), and \(|\varphi (M_1, M_2, T) \cap \{\beta , \gamma \}| = |M_T \cap \{e_4, e_6\}|\) hold by the definition of \(\varphi (M_1, M_2, T)\), replacing \(M_T\) with \(\varphi (M_1, M_2, T)\) does not affect the degrees of vertices in V.
Furthermore, for every \(T \in {\mathcal {T}}^+_{S^*}\) of type (B) or (B\('\)), since \(|\varphi (M_1, M_2, T) \cap \{\alpha , \gamma \}| = |M_T \cap \{e_2, e_{10}\}|\), \(|\varphi (M_1, M_2, T) \cap \{\alpha , \beta \}| = |M_T \cap \{e_5, e_8\}|\), and \(|\varphi (M_1, M_2, T) \cap \{\beta , \gamma \}| = |M_T \cap \{e_6, e_9\}|\) hold by the definition of \(\varphi (M_1, M_2, T)\), replacing \(M_T\) with \(\varphi (M_1, M_2, T)\) does not affect the degrees of vertices in V.
Since \(b(v) = b_1(v)\) for \(v \in S^*\) and \(b(v) = b_2(v)\) for \(v \in V^* {\setminus } S^*\), this shows that \(M_1 \oplus M_2\) forms a b-factor. Since \(M_j\) is \({\mathcal {T}}_j\)-free for \(j \in \{1, 2\}\), \(M_1 \oplus M_2\) is a \({\mathcal {T}}\)-free b-factor.
We next show that \(x = \sum _{(M_1, M_2) \in {\mathcal {M}}} \lambda _{(M_1, M_2)} x_{M_1 \oplus M_2}\). By the definitions of \(x_1, x_2, M_1 \oplus M_2\), and \(\lambda _{(M_1, M_2)}\), it holds that
for \(e \in E {\setminus } \bigcup _{T \in {\mathcal {T}}^+_{S^*}} E(T)\).
Let \(T \in {\mathcal {T}}^+_{S^*}\) be a triangle of type (A) for \((G_1, b_1, {\mathcal {T}}_1)\) and let \(\alpha , \beta \), and \(\gamma \) be as in Figs. 7 and 8. By the definition of \(\varphi (M_1, M_2, T)\), we obtain
We also obtain
Since a similar equality holds for \(\gamma \) by symmetry, (11) holds for \(e \in \{\alpha , \beta , \gamma \}\). Since T is a triangle of type (A\('\)’) for \((G_1, b_1, {\mathcal {T}}_1)\) if and only if it is of type (A) for \((G_2, b_2, {\mathcal {T}}_2)\), the same argument can be applied when T is a triangle of type (A\('\)) for \((G_1, b_1, {\mathcal {T}}_1)\).
Let \(T \in {\mathcal {T}}^+_{S^*}\) be a triangle of type (B) for \((G_1, b_1, {\mathcal {T}}_1)\) and let \(\alpha , \beta \), and \(\gamma \) be as in Figs. 9 and 10. By the definition of \(\varphi (M_1, M_2, T)\), we obtain
We also obtain
Since a similar equality holds for \(\gamma \) by symmetry, (11) holds for \(e \in \{\alpha , \beta , \gamma \}\). The same argument can be applied when T is a triangle of type (B\('\)) for \((G_1, b_1, {\mathcal {T}}_1)\).
Therefore, (11) holds for every \(e \in E\), which complete the proof. \(\square \)
Rights and permissions
About this article
Cite this article
Kobayashi, Y. Weighted Triangle-free 2-matching Problem with Edge-disjoint Forbidden Triangles. Math. Program. 192, 675–702 (2022). https://doi.org/10.1007/s10107-021-01661-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10107-021-01661-y