Skip to main content
Log in

Separation routine and extended formulations for the stable set problem in claw-free graphs

  • Full Length Paper
  • Series A
  • Published:
Mathematical Programming Submit manuscript

Abstract

The maximum weighted stable set problem in claw-free graphs is a well-known generalization of the maximum weighted matching problem, and a classical problem in combinatorial optimization. In spite of the recent development of fast(er) combinatorial algorithms and some progresses in the characterization of the corresponding stable set polytope, the problem of “providing a decent linear description” for this polytope (Grötschel et al. in Geometric algorithms and combinatorial optimization, Springer, New York, 1988) is still open. The main contribution of this paper is to propose an algorithmic answer to that question by providing a polynomial-time and computationally attractive separation routine for the stable set polytope of claw-free graphs, that only requires a combinatorial decomposition algorithm, the solution of (moderate sized) compact linear programs, and Padberg and Rao’s algorithm for separating over the matching polytope. In particular, it is a generalization of the latter and avoids the heavy computational burden of resorting to the ellipsoid method, on which the only poly-time separation routine known so far relied. Besides, our separation routine comes with a ‘small’ (but not polynomial) extended linear programming formulation and a procedure to derive a linear description of the stable set polytope of claw-free graphs in the original space.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5

Similar content being viewed by others

Notes

  1. In this paper, by fast, we mean computationally fast, which might entail polynomiality—as for Padberg and Rao’s separation routine—but not necessarily: for instance, we consider the simplex method a fast algorithm for linear programming.

  2. See [17] for a definition.

  3. While the facets of the stable set polytope of claw-free graphs with stability number at least four and no 1-join are reasonnably nice and could possibily allow for an elegant min–max characterization, the facets for claw-free graphs with stability number three can exhibit arbitrary many (complicated) coefficients [31] and this leaves very little hope for a neat min–max characterization through linear programming duality in the original space.

  4. More precisely, we mean here that there is no need to call the ellipsoid method within the separation routine (as would a separation routine building upon the polytime equivalence between optimization and separation). Now depending on the application, one might still want to call the separation routine within the ellipsoid method to guarantee polytime solvability of the corresponding cutting plane approach, or one could use the routine within another form of cutting plane approach (or within a branch-and-cut framework) that might not guarantee polynomiality but would be efficient computationally.

  5. Another simpler extended formulation we developed appeared in [10, 12], see also [22]; however it could not be used for the separation in the original space.

  6. A sketch of the separation routine presented in this paper and the corresponding extended formulation was published in SODA 2012 [12].

  7. In fact, it outputs a most violated inequality, i.e., one maximizing the difference between the left-hand side and the right-hand side.

  8. Different from clique inequalities and non-negativity constraints.

References

  1. Balas, E.: Disjunctive programming: properties of the convex hull of feasible points. SIAM J. Algebr. Discrete Methods 6, 466–486 (1985)

    Article  Google Scholar 

  2. Bonomo, F., Oriolo, G., Snels, C.: Minimum weighted clique cover on strip-composed perfect graphs. In: WG 2012, pp. 22–33 (2012)

  3. Chudnovsky, M., Seymour, P.: The structure of claw-free graphs. Surv. Comb. 327, 153–171 (2005)

    MathSciNet  MATH  Google Scholar 

  4. Chudnovsky, M., Seymour, P.: Claw free graphs IV. Global structure. J. Comb. Theory Ser. B 98, 1373–1410 (2008)

    Article  Google Scholar 

  5. Chvátal, V.: Edmonds polytopes and a hierarchy of combinatorial problems. Discrete Math. 4, 305–337 (1973)

    Article  MathSciNet  Google Scholar 

  6. Cook, W.J., Cunningham, W.H., Pulleyblank, W.R., Schrijver, A.: Combinatorial Optimization. Wiley, New York (1998)

    MATH  Google Scholar 

  7. Dubois, A., Rouire, M.: Algorithmes de décomposition dans les graphes sans griffe et couplage: Travail d’Étude et de Recherche. Master MIMSE, Université de Bordeaux (2011)

  8. Edmonds, J.: Maximum matching and a polyhedron with 0,1-vertices. J. Res. Natl. Bur. Stand. 69, 125–130 (1965)

    Article  MathSciNet  Google Scholar 

  9. Eisenbrand, F., Oriolo, G., Stauffer, G., Ventura, P.: Circular one matrices and the stable set polytope of quasi-line graphs. Combinatorica 28(1), 45–67 (2008)

    Article  MathSciNet  Google Scholar 

  10. Faenza, Y.: On the interplay between extended formulations and algorithms in some combinatorial problems. Ph.D. thesis, Sapienza university of Rome (2010)

  11. Faenza, Y., Oriolo, G., Stauffer, G.: An algorithmic decomposition of claw-free graphs leading to an \(O(n^3)\)-algorithm for the weighted stable set problem. In: Proceedings of SODA 2011, pp. 630–646 (2011)

  12. Faenza, Y., Oriolo, G., Stauffer, G.: Separating stable sets in claw-free graphs via Padberg–Rao and compact linear programs. In: Proceedings of SODA 2012, pp. 1298–1308 (2012)

  13. Faenza, Y., Oriolo, G., Stauffer, G.: Solving the maximum weighted stable set problem in claw-free graphs via decomposition. J. ACM 61(4), 20–41 (2014)

    Article  Google Scholar 

  14. Fiorini, S., Massar, S., Pokutta, S., Tiwary, H.R., De Wolf, R.: Exponential lower bounds for polytopes in combinatorial optimization. J. ACM 62(2), 17:1–17:23 (2015)

    Article  MathSciNet  Google Scholar 

  15. Galluccio, A., Gentile, C., Ventura, P.: Gear composition and the stable set polytope. Oper. Res. Lett. 36(4), 419–423 (2008)

    Article  MathSciNet  Google Scholar 

  16. Galluccio, A., Gentile, C., Ventura, P.: The stable set polytope of claw-free graphs with stability number at least four. II. Striped graphs are-perfect. J. Comb. Theory Ser. B 108, 1–28 (2014)

    Article  MathSciNet  Google Scholar 

  17. Galluccio, A., Gentile, C., Ventura, P.: The stable set polytope of claw-free graphs with stability number at least four. I. Fuzzy antihat graphs are-perfect. J. Comb. Theory Ser. B 107, 92–122 (2014)

    Article  MathSciNet  Google Scholar 

  18. Grötschel, M., Lovasz, L., Schrijver, A.: Geometric Algorithms and Combinatorial Optimization. Springer, New York (1988)

    Book  Google Scholar 

  19. Grötschel, M., Holland, O.: A cutting-plane algorithm for minimum perfect 2-matching. Computing 39, 327–344 (1987)

    Article  MathSciNet  Google Scholar 

  20. Hopf, M., Thielen, C., Wendt, O.: Competitive algorithms for multistage online scheduling. Eur. J. Oper. Res. 260(2), 468–481 (2017)

    Article  MathSciNet  Google Scholar 

  21. Kaibel, V.: Extended formulations in combinatorial optimization. Optima 85, 2–7 (2011)

    Google Scholar 

  22. Kaibel, V., Loos, A.: Branched polyhedral systems. In: Eisenbrand, F., Shepherd, B. (eds.) Proceedings of IPCO XIV, pp. 177–190 (2010)

  23. Kose, A., Medard, M.: Scheduling wireless ad hoc networks in polynomial time using claw-free conflict graphs. (2017). arXiv:1711.01620

  24. Martin, R.K.: Using separation algorithms to generate mixed integer model reformulations. Oper. Res. Lett. 10(3), 119–128 (1991)

    Article  MathSciNet  Google Scholar 

  25. Minty, G.J.: On maximal independent sets of vertices in claw-free graphs. J. Comb. Theory 28, 284–304 (1980)

    Article  MathSciNet  Google Scholar 

  26. Nakamura, D., Tamura, A.: A revision of Minty’s algorithm for finding a maximum weighted stable set of a claw-free graph. J. Oper. Res. Soc. Jpn. 44(2), 194–2004 (2001)

    MATH  Google Scholar 

  27. Nobili, P., Sassano A.: An \({\cal{O}}(n^2 \log (n))\) algorithm for the weighted stable set problem in claw-free graphs. (2015). arXiv:1501.05775

  28. Oriolo, G., Pietropaoli, U., Stauffer, G.: A new algorithm for the maximum weighted stable set problem in claw-free graphs. In: Lodi, A., Panconesi, A., Rinaldi, G. (eds.) Proceedings Thirteenth IPCO Conference, pp. 77–96 (2008)

  29. Padberg, M.W.: Node packings in graphs and claw free graphs. Optima 86, 8–10 (2011)

    Google Scholar 

  30. Padberg, M.W., Rao, M.R.: Odd minimum cut-sets and b-matchings. Math. Oper. Res. 7(1), 67–80 (1982)

    Article  MathSciNet  Google Scholar 

  31. Pecher, A., Wagler, A.: On facets of stable set polytopes of claw-free graphs with stability number three. Discrete Math. 310, 493–498 (2010)

    Article  MathSciNet  Google Scholar 

  32. Pulleyblank W.R., Shepherd, F.B.: Formulations for the stable set polytope of a claw-free graph. In: Rinaldi, G., Wolsey, L. (eds.) Proceedings Third IPCO Conference, pp. 267–279 (1993)

  33. Rothvoss, T.: The matching polytope has exponential extension complexity. J. ACM 64(6), 41:1–41:19 (2017)

    Article  MathSciNet  Google Scholar 

  34. Schrijver, A.: Combinatorial Optimization. Polyhedra and Efficiency (3 volumes). Algorithms and Combinatorics, vol. 24. Springer, Berlin (2003)

    MATH  Google Scholar 

  35. Stauffer, G.: On the stable set polytope of claw-free graphs. PhD thesis, Ecole Polytechnique Fédérale de Lausanne (2005)

  36. Weller, A., Jebara, T.: Revisiting the limits of MAP inference by MWSS on perfect graphs. In: Artificial Intelligence and Statistics, pp. 1061–1069 (2015)

Download references

Acknowledgements

We acknowledge support from the Ambizione Grant PZ00P2_154779 Tight formulations of 0-1 problems funded by the Swiss National Science Foundation and from a start-up Grant from the Columbia IEOR Department and SEAS.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Gautier Stauffer.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

An extended abstract containing a few results from the present paper appeared in [12].

A Proof of Lemma 6

A Proof of Lemma 6

Proof

Claim Let \(x^* \in \hbox {STAB}(G)\) with \(x^*_i=x^*[G_i]\in \hbox {STAB}(G_i)\). Consider two vectors \(\underline{s},\overline{s}\in [0,1]^k\) such that \(\underline{t}\le \underline{s}\le \overline{s}\le \overline{t}\) and we define \(y=f(x^*,\underline{s},\overline{s}) \in \mathbb {R}^{V(H(G))}\). We denote by \(y_i{:}{=}y[gg_i]\). Let \((G_i,\mathcal{A}_i)\) be a strip of G with \(\mathcal{A}_i=\{A_i,B_i\}\). Then, for all \(i\in [k]\), \(\underline{s}_i=\min \{t_i: y_i\in \hbox {STAB}(gg_i,t_i,\{u_i\},\{v_i\})\) and \(\overline{s}_i=\max \{t_i: y_i \in \hbox {STAB}(gg_i,t_i,\{u_i\},\{v_i\})\).

Proof

Fix \(i\in [k]\) and \(t_i \in [0,1]\). We will show that \(y_i \in \hbox {STAB}(gg_i,t_i,\{u_i\},\{v_i\})\) if and only if \(\underline{s_i}\le t_i \le \overline{s_i}\). In order to prove the latter fact, we take a small detour. We define the graph \(h_i\) obtained from \(gg_i\) by adding the edge \(\{u_i,v_i\}\). We will show that:

  1. (i)

    \(y_i\) is in \(\hbox {STAB}(gg_i,t_i,\{u_i\},\{v_i\})\) if and only if \(y_i' = y_i - t_i \cdot \chi ^{\{u_i,v_i\}}\) is in \((1-t_i) \cdot \hbox {STAB}(h_i)\) (a point y is in \(\rho \cdot Q\) with \(Q=\{x:Ax\le b\}\) if \(A y \le \rho \cdot b\));

  2. (ii)

    \(y_i'\in (1-t_i) \cdot \hbox {STAB}(h_i)\) if and only if \(\underline{s_i}\le t_i \le \overline{s_i}\).

Clearly, (i) and (ii) are enough to prove that \(y_i\) is in \(\hbox {STAB}(gg_i,t_i,\{u_i\},\{v_i\})\) if and only if \(\underline{s_i}\le t_i \le \overline{s_i}\), and therefore the claim.

  1. (i)

    If \(y_i \in \hbox {STAB}(gg_i,t_i,\{u_i\},\{v_i\})\), then there exist \(\lambda \in \mathbb {R}^{\mathcal{S}(gg_i)}_+ : y_i=\sum _{S \in \mathcal{S}(gg_i)} \lambda _S \chi ^S\), \(\sum _{S \in \mathcal{S}(gg_i)} \lambda _S =1\), with \(\lambda _{\{u_i,v_i\}} =t_i\), since a stable set of \(gg_i\) containing \(u_i\) and \(v_i\) does not contain any other node. If we remove the stable set \(\{u_i,v_i\}\) from the combination we get \(y_i'\), which is then expressed as a non-negative combination of stable sets of \(h_i\) with coefficients summing to \(1-t_i\). Hence \( y_i' \in (1-t_i) \cdot \hbox {STAB}(h_i)\); by reversing the operations above we deduce that the converse also holds and the claim follows.

  2. (ii)

    Since \(h_i\) is a perfect graph, \(\hbox {STAB}(h_i)\) is described by clique inequalities and non-negativity constraints. The clique inequalities in \((1-t_i) \cdot \hbox {STAB}(h_i)\) read: \(y'(u_i)+ y'({w^1_i}) + y'(w_i) \le 1-t_i, y'(w_i)+y'({w^1_i})+y'({w^2_i}) \le 1-t_i, y'({w_i})+y'({w^2_i})+y'({v_i}) \le 1-t_i\) and \(y'({u_i}) + y'({v_i}) + y'(w_i) \le 1-t_i\). The non-negativity constraints read: \(y'({u_i}) , y'({v_i}) , y'(w_i),y'({w^1_i}),y'({w^2_i}) \ge 0\). Substituting \(y_i'\) by its expression in term of \(x^*_i\) and \(t_i\), performing simple calculations and using the bounds from Lemma 5, it follows that \(y_i'\in (1-t_i) \cdot \hbox {STAB}(h_i)\) if and only if \(\underline{s_i}\le t_i \le \overline{s_i}\), as required.

\(\square \)

We need one more technical claim, where we assume that G is the gluing of a strip \((G_i, \{A_i, B_i\})\) with a strip \((G',\{A',B'\})\) and show how to reduce the membership problem for \(\hbox {STAB}(G)\) to the membership problems for two suitable sets that are derived from \(\hbox {STAB}(G_i)\) and \(\hbox {STAB}(G')\). (Note that the second strip is denoted differently since, in general, it will not be a strip from our original strip decomposition.) In the following, we let \(\overline{G'}\) be the graph obtained from \(G'\) by adding two non-adjacent vertices \(z^1, z^2\) that are complete respectively to \(A'\) and \(B'\), and anti-complete to the other vertices of \(G'\) (i.e., \(G'\) is the gluing of \((G',\{A',B'\})\) with \((L,\{\{z^1\}, \{z^2\}\})\), where L is the graph with vertices \(z^1\), \(z^2\) and no edges). Note that \((\overline{G'}, \{\{z^1\},\{z^2\}\})\) is a 2-strip.

Claim 1

Let G be the gluing of \((G_i,\{A_i,B_i\})\) and \((G',\{A',B'\})\), and let \( x^*\in \mathbb {R}^{V(G)}_+\). We extend \(x^*[G']\) to vertices of \(\overline{G'}\) by setting \(x^*(z^1)=x^*(A_i)\) and \(x^*(z^2)=x^*(B_i)\), and denote by \(x^*[\overline{G'}]\) the resulting vector. Let \(t_i\) such that \(\underline{t_i}\le t_i \le \overline{t_i}\). Then \(x^* \in \hbox {STAB}(G,t_i,A_i,B_i)\) if and only if \(x^*[\overline{G'}] \in \hbox {STAB}(\overline{G'}, t_i, \{z^1\},\{z^2\})\).

Proof

Necessity. Let \(\lambda \in \mathbb {R}^{\mathcal{S}(G)}_+\) such that \(e\sum _{S \in \mathcal{S}(G)} \lambda _S=1\), \(x^*=\sum _{S \in \mathcal{S}(G)} \lambda _S \chi ^S\) with \(t_i=\sum _{S \in \mathcal{S}_{A_i,B_i}(G)} \lambda _S\) and \(\underline{t_i}\le t_i \le \overline{t_i}\). We associate to each \(S \in \mathcal{S}(G)\) a stable set \(S[\overline{G'}]\in \mathcal{S}(\overline{G'})\) by first taking the restriction of S to \(V(G')\), and then adding \(z^1\) and/or \(z^2\) whenever respectively \(S\cap A_i\ne \emptyset \) and/or \(S\cap B_i\ne \emptyset \). Then \(x^*[\overline{G'}]=\sum _{S \in \mathcal{S}(G)} \lambda _S \chi ^{S[\overline{G'}]}\), and therefore \(x^*[\overline{G'}]\in \hbox {STAB}(\overline{G'}, t, \{z^1\},\{z^2\})\).

Sufficiency. \(t_i\in [\underline{t_i},\overline{t_i}]\) implies that \(x^*[G_i] \in \hbox {STAB}(G_i,t_i,A_i,B_i)\). In particular there exist multipliers \(\{\lambda _S\}_{S \in \mathcal{S}(G_i)} \ge 0\) such that \(\sum _{S \in \mathcal{S}(G_i)} \lambda _S=1\), \(x^*[G_i]=\sum _{S \in \mathcal{S}(G_i)} \lambda _S \chi ^S\), and \(\sum _{\mathcal{S}_{A_i,B_i}(G)} \lambda _S=t_i\). Now \(x^*[\overline{G'}] \in \hbox {STAB}(\overline{G'}, t_i, \{z^1\},\{z^2\})\) implies that there exists multipliers \(\{\mu _W\}_{W \in \mathcal{S}(\overline{G'})} \ge 0\) such that \(\sum _{W \in \mathcal{S}(\overline{G'})} \mu _W=1\), \(x^*[\overline{G'}]=\sum _{W \in \mathcal{S}(\overline{G'})} \mu _W \chi ^W\), and \(\sum _{W \in \mathcal{S}(\overline{G'}) : z^1, z^2 \in W} \mu _W=t_i\). We want to show that \(x^* \in \hbox {STAB}(G,t_i,A_i,B_i)\). It is easy to associate stable sets from \(\mathcal{S}(G_i)\) with stable sets from \(\mathcal{S}(\overline{G'})\) into stable sets of G as \(x^*(z^1)=x^*(A_i)\) and \(x^*(z^2)=x^*(B_i)\). Indeed, consider the union of the four complete bipartite networks with vertex set \(\mathcal{S}_{a,b}(G_i) \cup \mathcal{S}_{c,d}(\bar{G}')\) for \(a \in \{A_i,\lnot A_i\},b \in \{B_i,\lnot B_i\}\), \(c=\{z^1\}\) if \(a=A_i\) and \(\lnot \{z^1\}\) otherwise, and \(d=\{z^2\}\) if \(b=B_i\) and \(\lnot \{z^2\}\) otherwise. The offer in vertex \(S \in \mathcal{S}(G_i)\) is \(\lambda _S\) and the demand in vertex \(W \in \mathcal{S}(\bar{G}')\) is \(\mu _W\). Arcs have no capacities. It is easy to see that any feasible flow (i.e., a flow meeting offers and demands), and there is a least one, in this network will correspond to a certificate of feasibility for x, since we can interprete f(SW) as the multiplier for stable set \(S \cup W{\setminus } \{z^1,z^2\}\). The sum over all multipliers f(SW) intersecting both \(A_i\) and \(B_i\) is by construction \(t_i\). \(\square \)

Now express G as the gluing of \((G_1,A_1,B_1)\) and the appropriate strip \((G',A',B')\). Let \(x^*\in \mathbb {R}^{V(G)}\) such that \(x^*_i=x^*[G_i]\in STAB(G_i)\) for \(i \in [k]\), and let \( \underline{s}_1, \overline{s}_1\) such that \(\underline{t}_1 \le \underline{s}_1 \le \overline{s}_1 \le \overline{t}_1\).

We next show that \( x^* \in \hbox {STAB}(G, t^*,A_1,B_1)\) for some \(t^* \in [\underline{s}_1,\overline{s}_1]\) if and only if \(\widetilde{x}=\left( \begin{array}{c}x^*[G']\\ y_1\\ \end{array}\right) \in \hbox {STAB}(G'+gg_1)\), where \(y_1\) is the restriction of \(f_1(x^*_1,\underline{s}_1,\overline{s}_1)\) to \(\mathbb {R}^{V(gg_i)}\). Then we can iterate the argument replacing at each step one strip \((G_i,\mathcal{A}_i)\) with the corresponding strip \((gg_i,\mathcal{A}^{gg}_i)\): we will finally obtain the graph H(G) and conclude that \(\exists t\) such that \(\underline{s}\le t \le \overline{s}\) and \(x^*\in \hbox {STAB}(G, \{ t_i,A_i,B_i\})\) if and only if \(y=f(x^*,\underline{s},\overline{s}) \in \hbox {STAB}(H(G))\). This proves the equivalence of (i) and (ii).

We observe first that we have (*) \(\underline{s}_1=\min \{t^*: y_1\in \hbox {STAB}(gg_1,t^*,\{u_1\},\{v_1\})\) and \(\overline{s}_1=\max \{t^*: y_1 \in \hbox {STAB}(gg_1,t^*,\{u_1\},\{v_1\})\) by Claim A. Then:

$$\begin{aligned} \begin{array}{cl} \exists t^* \in [\underline{s}_1, \overline{s}_1] : x^* \in \hbox {STAB}(G, t^*,A_1,B_1)\\ \Leftrightarrow &{} \hbox {(by Claim A)}\\ \exists t^* \in [\underline{s}_1, \overline{s}_1]: x^*[\overline{G'}]\in \hbox {STAB}(\overline{G'}, t^*, \{z^1\},\{z^2\}) \\ \Leftrightarrow &{} \hbox {(by Claim A, with } gg1\hbox { playing the role of } G_1\\ &{} \hbox { and because of (*))}\\ \exists t^* \in [\underline{s}_1, \overline{s}_1]: \widetilde{x} \in \hbox {STAB}(G'+gg_1,t^*,\{u_1\},\{v_1\}) \\ \Leftrightarrow &{} (\text{ because } \text{ of } \text{(*) }) \\ \widetilde{x} \in \hbox {STAB}(G'+gg_1) \end{array} \end{aligned}$$

Now we need to prove equivalence of (iii) with (i) and (ii). It’s enough to observe that we can apply equivalence between (i) and (ii), by choosing \(\underline{s}=\overline{s}=t\), to derive equivalence between (i) and (iii).

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Faenza, Y., Oriolo, G. & Stauffer, G. Separation routine and extended formulations for the stable set problem in claw-free graphs. Math. Program. 188, 53–84 (2021). https://doi.org/10.1007/s10107-020-01502-4

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s10107-020-01502-4

Keywords

Mathematics Subject Classification

Navigation