Skip to main content

The Complexity of Perfect Packings in Dense Graphs

  • Conference paper
  • First Online:
Theory and Applications of Models of Computation (TAMC 2017)

Part of the book series: Lecture Notes in Computer Science ((LNTCS,volume 10185))

  • 800 Accesses

Abstract

Let H be a graph of order h and let G be a graph of order n such that \(h\mid n\). A perfect H-packing in G is a collection of vertex disjoint copies of H in G that covers all vertices of G. Hell and Kirkpatrick showed that the decision problem whether a graph G has a perfect H-packing is NP-complete if and only if H has a component which contains at least 3 vertices.

We consider the decision problem of containment of a perfect H-packing in graphs G under the additional minimum degree condition. Our main result shows that given any \(\gamma >0\) and any n-vertex graph G with minimum degree at least \((1-1/\chi _{cr}(H)+\gamma )n\), the problem of determining whether G has a perfect H-packing can be solved in polynomial time, where \(\chi _{cr}(H)\) is the critical chromatic number of H. This answers a question of Yuster negatively. Moreover, a hardness result of Kühn and Osthus shows that our main result is essentially best possible and closes a long-standing hardness gap for all complete multi-partite graphs H whose second smallest color class has size at least 2.

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

Access this chapter

Chapter
USD 29.95
Price excludes VAT (USA)
  • Available as PDF
  • Read on any device
  • Instant download
  • Own it forever
eBook
USD 39.99
Price excludes VAT (USA)
  • Available as EPUB and PDF
  • Read on any device
  • Instant download
  • Own it forever
Softcover Book
USD 54.99
Price excludes VAT (USA)
  • Compact, lightweight edition
  • Dispatched in 3 to 5 business days
  • Free shipping worldwide - see info

Tax calculation will be finalised at checkout

Purchases are for personal use only

Institutional subscriptions

Notes

  1. 1.

    More precisely, here one should take the complete \(\chi (H)\)-partite graphs with \(\sigma (H)n/|H|-1\) vertices in one color class, and other color classes of sizes as equal as possible.

References

  1. Alon, N., Fischer, E.: Refining the graph density condition for the existence of almost \(K\)-factors. ARS Combin. 52, 296–308 (1999)

    MathSciNet  MATH  Google Scholar 

  2. Alon, N., Yuster, R.: \(H\)-factors in dense graphs. J. Combin. Theory Ser. B 66(2), 269–282 (1996)

    Article  MathSciNet  MATH  Google Scholar 

  3. Cooley, O., Kühn, D., Osthus, D.: Perfect packings with complete graphs minus an edge. Eur. J. Combin. 28(8), 2143–2155 (2007)

    Article  MathSciNet  MATH  Google Scholar 

  4. Edmonds, J.: Paths, trees, and flowers. Canad. J. Math. 17, 449–467 (1965)

    Article  MathSciNet  MATH  Google Scholar 

  5. Erdős, P.: On extremal problems of graphs and generalized graphs. Isr. J. Math. 2(3), 183–190 (1964)

    Article  MathSciNet  MATH  Google Scholar 

  6. Han, J.: Decision problem for perfect matchings in dense uniform hypergraphs. Trans. Amer. Math. Soc. (accepted)

    Google Scholar 

  7. Hell, P., Kirkpatrick, D.G.: On the complexity of general graph factor problems. SIAM J. Comput. 12(3), 601–609 (1983)

    Article  MathSciNet  MATH  Google Scholar 

  8. Hurkens, C.A.J., Schrijver, A.: On the size of systems of sets every \(t\) of which have an SDR, with an application to the worst-case ratio of heuristics for packing problems. SIAM J. Discrete Math. 2(1), 68–72 (1989)

    Article  MathSciNet  MATH  Google Scholar 

  9. Kann, V.: Maximum bounded \(H\)-matching is MAX SNP-complete. Inform. Process. Lett. 49(6), 309–318 (1994)

    Article  MathSciNet  MATH  Google Scholar 

  10. Karpiński, M., Ruciński, A., Szymańska, E.: Computational complexity of the perfect matching problem in hypergraphs with subcritical density. Internat. J. Found. Comput. Sci. 21(6), 905–924 (2010)

    Article  MathSciNet  MATH  Google Scholar 

  11. Kawarabayashi, K.: \(K^-_4\)-factor in a graph. J. Graph Theor. 39(2), 111–128 (2002)

    Article  MathSciNet  MATH  Google Scholar 

  12. Keevash, P., Knox, F., Mycroft, R.: Polynomial-time perfect matchings in dense hypergraphs. Adv. Math. 269, 265–334 (2015)

    Article  MathSciNet  MATH  Google Scholar 

  13. Keevash, P., Mycroft, R.: A geometric theory for hypergraph matching. Mem. Am. Math. Soc. 233(1098) (2014). Monograph

    Google Scholar 

  14. Komlós, J.: Tiling Turán theorems. Combinatorica 20(2), 203–218 (2000)

    Article  MathSciNet  MATH  Google Scholar 

  15. Komlós, J., Sárközy, G., Szemerédi, E.: Proof of the Alon-Yuster conjecture. Discrete Math. 235(1–3), 255–269 (2001). Combinatorics (prague, 1998)

    Article  MathSciNet  MATH  Google Scholar 

  16. Kühn, D., Osthus, D.: Critical chromatic number and the complexity of perfect packings in graphs. In: Proceedings of the Seventeenth Annual ACM-SIAM Symposium on Discrete Algorithms, pp. 851–859 (2006)

    Google Scholar 

  17. Kühn, D., Osthus, D.: The minimum degree threshold for perfect graph packings. Combinatorica 29(1), 65–107 (2009)

    Article  MathSciNet  Google Scholar 

  18. Le Gall, F.: Powers of tensors and fast matrix multiplication. In: Proceedings of the 39th International Symposium on Symbolic and Algebraic Computation, ISSAC 2014, pp. 296–303. ACM, New York (2014)

    Google Scholar 

  19. Lo, A., Markström, K.: F-factors in hypergraphs via absorption. Graphs Comb. 31(3), 679–712 (2015)

    Article  MathSciNet  MATH  Google Scholar 

  20. Rödl, V., Ruciński, A., Szemerédi, E.: A Dirac-type theorem for 3-uniform hypergraphs. Comb. Probab. Comput. 15(1–2), 229–251 (2006)

    Article  MathSciNet  MATH  Google Scholar 

  21. Shokoufandeh, A., Zhao, Y.: Proof of a tiling conjecture of Komlós. Random Struct. Algor. 23(2), 180–205 (2003)

    Article  MATH  Google Scholar 

  22. Treglown, A.: Personal communication

    Google Scholar 

  23. Yuster, R.: Combinatorial and computational aspects of graph packing and graph decomposition. Comput. Sci. Rev. 1(1), 12–26 (2007)

    Article  MATH  Google Scholar 

Download references

Acknowledgments

The author’s research is supported by FAPESP Proc. (2013/03447-6, 2014/18641-5, 2015/07869-8). The author would like to thank Andrew Treglown for helpful discussions and thank Andrew Treglown and two anonymous reviewers for valuable comments.

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Jie Han .

Editor information

Editors and Affiliations

Appendices

A Proof of Proposition 1

We need the following simple counting result, which, for example, follows from the result of Erdős [5] on supersaturation.

Proposition 2

Given \(\gamma '> 0\), \(\ell _1, \dots , \ell _ k\in \mathbb {N}\), there exists \(\mu >0\) such that the following holds for sufficiently large n. Let T be an n-vertex graph with a vertex partition \(V_1 \cup \dots \cup V_d\). Suppose \(i_1, \dots , i_k\in [d]\) and T contains at least \(\gamma ' {n}^{k}\) copies of \(K_k\) with vertex set \(\{ v_1, \dots , v_k \}\) such that \(v_1\in V_{i_1}\), \(\dots , v_k\in V_{i_k}\). Then T contains at least \(\mu {n}^{\ell _1+\cdots + \ell _k}\) copies of \(K_{\ell _1, \dots , \ell _ k}\) whose jth part is contained in \(V_{i_j}\) for all \(j\in [k]\).

Proof

(Proof of Proposition 1 ). Write \(L=L_{\mathcal {P}}^{\mu }(G)\) and \(Q=Q(\mathcal {P}, L)\). It suffices to show that for any element \(\mathbf {v}\in L_{\max }^{d}\), there exists \(\mathbf {v}'=(v_1',\dots , v_d')\) such that \(-(m-1)\le v_i'\le m-1\) for all \(i\in [d]\) and \(\mathbf {v}- \mathbf {v}'\in L\) – since the number of such \(\mathbf {v}'\) is at most \((2m-1)^d\). Recall that since H is unbalanced, there exists a k-coloring with color class sizes \(a_1\le \cdots \le a_k\) and \(a_1< a_k\). Let \(a=a_{k} - a_{1}<m\).

Define a graph P on the vertex set [d] such that \((i, j)\in P\) if and only if \(e(G[V_i, V_j])\ge \gamma n^2\). We claim that if i and j are connected in P, then \(a(\mathbf {u}_i-\mathbf {u}_j)\in L\). Indeed, first assume that \((i,j)\in P\). For each edge uv in \((V_i, V_j)\), by

$$ \delta (G)\ge (1-1/\chi _{cr}(H)+\gamma ) n\ge \left( 1- \frac{m-1}{(k-1)m} +\gamma \right) n, $$

it is easy to see that uv is contained in at least \(\frac{1}{m^{k-2}} \left( {\begin{array}{c}n\\ k-2\end{array}}\right) \) copies of \(K_k\). So there are at least \(\gamma n^2 \cdot \frac{1}{m^{k-2}} \left( {\begin{array}{c}n\\ k-2\end{array}}\right) /\left( {\begin{array}{c}k\\ 2\end{array}}\right) \) copies of \(K_k\) in G intersecting both \(V_i\) and \(V_j\). By averaging, there exists a k-array \((i_1, \dots , i_k)\), \(i_j\in [d]\) where \(i_1=i\) and \(i_k=j\) such that G contains at least

$$ \frac{1}{d^{k-2}}\gamma n^2 \cdot \frac{1}{m^{k-2}} \left( {\begin{array}{c}n\\ k-2\end{array}}\right) /\left( {\begin{array}{c}k\\ 2\end{array}}\right) \ge \frac{\gamma }{m^{k-2}d^{k-2}k!}n^{k} $$

copies of \(K_k\) with vertex set \(\{ v_1, \dots , v_k \}\) such that \(v_1\in V_{i_1}\), \(\dots , v_k\in V_{i_k}\). By applying Proposition 2 with \(\ell _i=a_i\), \(i\in [k]\), we get that there are at least \(\mu n^{m}\) copies of \(K_{a_1, \dots , a_ k}\) in G whose jth part is contained in \(V_{i_j}\) for all \(j\in [k]\). We apply Proposition 2 again, this time with \(\ell _i=a_i\), for all \(2\le i\le k-1\) and \(\ell _{1}=a_{k}\), \(\ell _{k}=a_{1}\) and thus conclude that there are at least \(\mu n^{m}\) copies of \(K_{a_k, a_2,\dots , a_{k-1}, a_1}\) (with \(a_1\) and \(a_k\) exchanged) in G whose jth part is contained in \(V_{i_j}\) for all \(j\in [k]\). Taking subtraction of index vectors of these two types of copies gives that \(a(\mathbf {u}_i-\mathbf {u}_j)\in L\). Furthermore, note that if i and j are connected by a path in P, we can apply the argument above to every edge in the path and conclude that \(a(\mathbf {u}_i-\mathbf {u}_j)\in L\), so the claim is proved.

Now we separate two cases.

Case 1: \(k\ge 3\) . In this case, we first show that P is connected. Indeed, we prove that for any bipartition \(A\cup B\) of [d], there exists \(i\in A\) and \(j\in B\) such that \((i,j)\in P\). Let \(V_A=\bigcup _{i\in A} V_i\) and \(V_B=\bigcup _{j\in B}V_j\). Without loss of generality, assume that \(|V_A|\le n/2\). Since \(\delta (G)\ge \frac{1+(k-2)m}{(k-1)m} n \ge (1/2 + 1/(2m))n\), the number of edges in G that are incident to \(V_A\) is at least

$$ |V_A| \cdot \left( \frac{1}{2} + \frac{1}{2m} \right) n - \left( {\begin{array}{c}|V_A|\\ 2\end{array}}\right) \ge \left( {\begin{array}{c}|V_A|\\ 2\end{array}}\right) + \frac{n}{4m}|V_A| \ge \left( {\begin{array}{c}|V_A|\\ 2\end{array}}\right) + \gamma n^2 |A| |B|, $$

where the last inequality follows from \(|A| |B|\le d^2/4\) and \(|V_i|\ge n/m\), \(i\in [d]\). By averaging, there exists \(i\in A\) and \(j\in B\) such that \(e(G[V_i, V_j])\ge \gamma n^2\) and thus \((i,j)\in P\).

Now let \(\mathbf {v}=(v_1,\dots , v_d)\in L_{\max }^{d}\). We fix an arbitrary m-vector \(\mathbf {w}\in L\) and let \(\mathbf {v}_1 = \mathbf {v}- ( |\mathbf {v}|/m ) \mathbf {w}\) and thus \(|\mathbf {v}_1| =0\). Since P is connected, the claim above implies that for any \(i,j\in [d]\), \(a(\mathbf {u}_i-\mathbf {u}_j)\in L\). Thus, we obtain the desired vector \(\mathbf {v}'\) by making the difference of any two digits at most a. Since \(|\mathbf {v}'|= |\mathbf {v}_1| =0\), we know that \(-(m-1)\le v_i'\le m-1\) for all \(i\in [d]\) and we are done.

Case 2: \(k = 2 \) . In this case we cannot guarantee that P is connected (we may even have some isolated vertices). First let i be an isolated vertex in P. By the definition of P, we know that \(e(G[V_i, V\setminus V_i]) \le (d-1)\gamma n^2\). Since \(\delta (G)\ge n/m\),

$$ e(G[V_i]) \ge \frac{1}{2} (|V_i| n/m - (d-1)\gamma n^2) \ge \frac{1}{4m} |V_i|^2. $$

Applying Proposition 2 on \(V_i\) shows that there are at least \(\mu n^m\) copies of \(K_{a_1, a_2}\) in \(V_i\), i.e., \(m\mathbf {u}_i\in L\). Second, if \((i,j)\in P\), then applying Proposition 2 on \([V_i, V_j]\) gives that \(a_1\mathbf {u}_i + a_2\mathbf {u}_j\in L\). So in both cases, for any component C in P, there exists an m-vector \(\mathbf {w}\in L\) such that \(\mathbf {w}|_{[d]\setminus C}=\mathbf 0 \).

Now let \(\mathbf {v}=(v_1,\dots , v_d)\in L_{\max }^{d}\). Consider the connected components \(C_1\), \(C_2, \dots , C_q\) of P, for some \(1\le q\le d\). By the conclusion in the last paragraph, there exists \(\mathbf {v}_1\in \mathbb {Z}^d\) such that \(\mathbf {v}- \mathbf {v}_1\in L\) and for each component \(C_i\), \(0\le |\mathbf {v}_1|_{C_i}|\le m-1\) (for each component C, we take out a multiple of the vector \(\mathbf {w}\) given by the last paragraph). Next, within each nontrivial component \(C_i\), we can use the claim to ‘balance’ the digits, as in Case 1. At last we obtain a vector \(\mathbf {v}'\) with the desired property.

B Proof of Lemma 1

In this subsection we prove Lemma 1. We will build a partition \(\mathcal {P}=\{V_1,\dots , V_d\}\) of V(G) for some \(d\le m\) such that every \(V_i\) is \((H,\beta , t)\)-closed for some \(\beta >0\) and integer \(t\ge 1\) in polynomial time. For any \(v\in V(G)\), let \(\tilde{N}_{H,\beta , i}(v)\) be the set of vertices in V(G) that are \((H,\beta , i)\)-reachable to v.

We need the following lemma [19, Lemma 4.2], which was originally stated for k-uniform hypergraphs. We remark that the current form can be easily derived by defining a k-uniform hypergraph \(G'\) where each k-set forms a hyperedge if and only if it spans a \(K_k\) in G. For any vertex \(u\in V(G)\), let W(u) be the collection of \((k-1)\)-cliques \(S\subseteq N(u)\).

Lemma 3

[19]. Given H and \(\gamma '>0\), there exists \(\alpha >0\) such that the following holds for sufficiently large n. For any n-vertex graph G, two vertices \(x, y \in V(G)\) are \((H,\alpha ,1)\)-reachable if the number of \((k-1)\)-sets \(S\in W(x)\cap W(y)\) with \(|N(S)|\ge \gamma ' n\) is at least \(\gamma ' \left( {\begin{array}{c}n\\ k-1\end{array}}\right) \), where \(N(S)=\bigcap _{v\in S}N(v)\).

Proposition 3

Suppose \(0< 1/n \ll \alpha \ll \gamma \ll 1/m\) and let G be an n-vertex graph with \(\delta (G)\ge (1-1/\chi _{cr}(H)+\gamma ) n\). Then for any \(v\in V(G)\), \(|\tilde{N}_{H,\alpha , 1}(v)| \ge (1/m+\gamma )n\).

Proof

First note that for each \((k-1)\)-clique S, we have \(|N(S)|\ge (1/m + k\gamma )n\). Then by Lemma 3, for any distinct \(u, v\in V(G)\), \(u\in \tilde{N}_{H,\alpha , 1}(v)\) if \(|W(u)\cap W(v)|\ge \gamma ^2 \left( {\begin{array}{c}n\\ k-1\end{array}}\right) \). By double counting, we have

$$ \sum _{S\in W(v)} |N(S)| < |\tilde{N}_{H,\alpha , 1}(v)|\cdot |W(v)|+n\cdot \gamma ^2 \left( {\begin{array}{c}n\\ k-1\end{array}}\right) . $$

Note that any S in the above inequality is a \((k-1)\)-clique, thus \(|N(S)| \ge (1/m + k\gamma )n\). On the other hand, it is easy to see that \(|W(v)|\ge \frac{1}{m^{k-1}}\left( {\begin{array}{c}n\\ k-1\end{array}}\right) \). By \(\gamma \ll 1/m\), we have

$$ |\tilde{N}_{H,\alpha , 1}(v)|> (1/m + k\gamma )n - \frac{\gamma ^2 n^k}{|W(v)|}\ge (1/m + \gamma )n. $$

   \(\square \)

The following lemma will be used to find the partition \(\mathcal {P}\) in Lemma 1. Its proof is almost identical to the one of [6, Lemma 3.8] and thus we omit it.

Lemma 4

Given \(0<\alpha \ll \{1/c, \delta '\}\), there exists constant \(\beta >0\) such that the following holds for all sufficiently large n. Assume an n-vertex graph T satisfies that \(|\tilde{N}_{H,\alpha , 1}(v)| \ge \delta ' n\) for any \(v\in V(T)\) and every set of \(c+1\) vertices in V(T) contains two vertices that are \((H,\alpha , 1)\)-reachable. Then we can find a partition \(\mathcal {P}\) of V(T) into \(V_1,\dots , V_d\) with \(d\le \min \{c, 1/\delta ' \}\) such that for any \(i\in [d]\), \(|V_i|\ge (\delta ' - \alpha ) n\) and \(V_i\) is \((H,\beta , 2^{c-1})\)-closed in T, in time \(O(n^{2^{c-1} m+1})\).

Now we are ready to prove Lemma 1.

Proof

(Proof of Lemma 1 ). Fix \(\gamma >0\). Without loss of generality, we may assume that \(\gamma \ll 1/m\). We apply Lemma 4 with \(c=m^{k-1}\), \(\delta '=1/m+\gamma \) and \(\alpha \ll \gamma \) and get \(\beta >0\). Suppose

$$ 1/n_0\ll \beta \ll \alpha \ll \gamma \ll \delta '. $$

Let G be a graph on \(n\ge n_0\) vertices satisfying \(\delta (G)\ge (1-1/\chi _{cr}(H)+\gamma ) n\). By Proposition 3, for any \(v\in V(G)\), \(|\tilde{N}_{H,\alpha , 1}(v)| \ge \delta ' n\). By the degree condition and Lemma 3, for distinct \(u, v\in V(G)\), u and v are \((H,\alpha , 1)\)-reachable if \(|W(u)\cap W(v)|\ge \gamma ^2 \left( {\begin{array}{c}n\\ k-1\end{array}}\right) \). So any set of \(c+1\) vertices in V(G) contains two vertices that are \((H,\alpha , 1)\)-reachable because \(|W(u)|\ge \frac{1}{c}\left( {\begin{array}{c}n-1\\ k-1\end{array}}\right) \) for any \(u\in V(G)\), and \((c+1)/c-1\ge \left( {\begin{array}{c}c+1\\ 2\end{array}}\right) \gamma ^2\). So we can apply Lemma 4 on G and get a partition \(\mathcal {P}=\{V_1, \dots , V_d\}\) of V(G) in time \(O(n^{2^{c-1}m+1})\). Note that \(|V_i|\ge (\delta '-\alpha )n \ge n/m\) for all \(i\in [d]\). Also \(d\le 1/\delta '\le m\) and each \(V_i\) is \((H,\beta , 2^{c-1})\)-closed.

Rights and permissions

Reprints and permissions

Copyright information

© 2017 Springer International Publishing AG

About this paper

Cite this paper

Han, J. (2017). The Complexity of Perfect Packings in Dense Graphs. In: Gopal, T., Jäger , G., Steila, S. (eds) Theory and Applications of Models of Computation. TAMC 2017. Lecture Notes in Computer Science(), vol 10185. Springer, Cham. https://doi.org/10.1007/978-3-319-55911-7_21

Download citation

  • DOI: https://doi.org/10.1007/978-3-319-55911-7_21

  • Published:

  • Publisher Name: Springer, Cham

  • Print ISBN: 978-3-319-55910-0

  • Online ISBN: 978-3-319-55911-7

  • eBook Packages: Computer ScienceComputer Science (R0)

Publish with us

Policies and ethics