## Abstract

We show that 3-graphs on *n* vertices whose minimum codegree is at least \((2/3 + o(1))n\) can be decomposed into tight cycles and admit Euler tours, subject to the trivial necessary divisibility conditions. We also provide a construction showing that our bounds are best possible up to a *o*(1) term. All together, our results answer negatively some recent questions of Glock, Joos, Kühn, and Osthus.

This is a preview of subscription content,

to check access.## Notes

These conjectures were modified in [3] after the original preprint version of this paper appeared.

## References

Keevash, P.: The existence of designs. Preprint (2014). arXiv: 1401.3665 [math.CO] (cit. on p. 1)

Glock, S., Kühn, D., Lo, A., Osthus, D.: The existence of designs via iterative absorption: hypergraph F-designs for arbitrary F. Mem. Am. Math. Soc.

**16**, 1–2 (2020)Glock, S., Kühn, D., Osthus, D.: “Extremal aspects of graph and hypergraph decomposition problems”. In: Surveys in combinatorics 2021. Vol. 470. London Mathematical Society. Lecture Note Ser. Cambridge University Press, Cambridge, (2021), pp. 235–265. https://doi.org/10.1017/9781009036214.007(cit. on pp. 1, 3, 24)

Barber, B., Kühn, D., Lo, A., Osthus, D.: Edge-decompositions of graphs with high minimum degree. Adv. Math.

**288**, 337–385 (2016). https://doi.org/10.1016/j.aim.2015.09.032Nash-Williams, C.: An unsolved problem concerning decomposition of graphs into triangles. In: Combinatorial Theory and its Applications, Vol. III, Proceedings of the Colloquium on Combinatorial Theory and its Applications held at Balatonfüred, August 24–29, 1969, North-Holland (1970), pp. 1179–1183 (cit. on p. 2)

Delcourt, M., Postle, L.: Progress towards Nash–Williams’ conjecture on triangle decompositions. J. Combin. Theory Ser. B

**146**, 286–416 (2021). https://doi.org/10.1016/j.jctb.2020.09.008Joos, F., Köhn, M.: Fractional cycle decompositions in hypergraphs’. Random Struct. Algorithms (2021), in press. https://doi.org/10.1002/rsa.21070(cit. on pp. 2, 14, 15)

Barber, B., Glock, S., Kühn, D., Lo, A., Montgomery, R., Osthus, D.: Minimalist designs. Random Struct. Algorithms

**57**(1), 47–63 (2020)Glock, S., Kühn, D., Lo, A., Montgomery, R., Osthus, D.: On the decomposition threshold of a given graph. J. Combin. Theory Ser. B

**139**, 47–127 (2019). https://doi.org/10.1016/j.jctb.2019.02.010Taylor, A.: On the exact decomposition threshold for even cycles. J. Graph Theory

**90**(3), 231–266 (2019). https://doi.org/10.1002/jgt.22399Euler, L.: Solutio problematis ad geometriam situs pertinentis. Comment. Academiae Sci. I. Petropolitanae

**8**, 128–140 (1741)Hierholzer, C., Wiener, C.: Ueber die Möglichkeit, einen Linienzug ohne Wiederholung und ohne Unterbrechung zu umfahren. Math. Ann.

**6**(1), 30–32 (1873). https://doi.org/10.1007/BF01442866Chung, F., Diaconis, P., Graham, R.: Universal cycles for combinatorial structures. Discrete Math.

**110**(1), 43–59 (1992). https://doi.org/10.1016/0012-365X(92)90699-GGlock, S., Joos, F., Kühn, D., Osthus, D.: Euler tours in hypergraphs. Combinatorica

**40**(5), 679–690 (2020). https://doi.org/10.1007/s00493-020-4046-8Glock, S., Kühn, D., Osthus, D.: Extremal aspects of graph and hypergraph decomposition problems. Preprint (2020). arXiv: 2008.00926v1 [math.CO] (cit. on p. 3)

Janson, S., Łuczak, T., Rucinski, A.: Random Graphs. Discrete Mathematics and Optimization, pp. 12–333. Wiley-Interscience, New York (2000). https://doi.org/10.1002/9781118032718

Raman, R.: The power of collision: randomized parallel algorithms for chaining and integer sorting. In: Foundations of software technology and theoretical computer science (Bangalore, 1990). Vol. 472. Lecture Notes in Computer Science Springer, Berlin, 1990, pp. 161–175. https://doi.org/10.1007/3-540-53487-3_42(cit. on p. 11)

Erdös, P.: On extremal problems of graphs and generalized graphs. Isr. J. Math.

**2**, 183–190 (1964). https://doi.org/10.1007/BF02759942Alon, N., Yuster, R.: On a hypergraph matching problem. Graphs Combin.

**21**(4), 377–384 (2005). https://doi.org/10.1007/s00373-005-0628-xKahn, J.: A linear programming perspective on the Frankl-Rödl-Pippenger theorem. Random Struct. Algorithms

**8**, 149–157 (1996)Ehard, S., Glock, S., Joos, F.: Pseudorandom hypergraph matchings. Combin. Probab. Comput.

**29**(6), 868–885 (2020). https://doi.org/10.1017/s0963548320000280Thomassen, C.: Decompositions of highly connected graphs into paths of length 3. J. Graph Theory

**58**(4), 286–292 (2008). https://doi.org/10.1002/jgt.20311

## Acknowledgements

We thank Felix Joos and Allan Lo for helpful discussions and suggestions, and the second author thanks Vincent Pfenninger for useful conversations.

## 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.

The research leading to these results was supported by the Czech Science Foundation, Grant Number GA19-08740 S with institutional support RVO: 67985807 (N. Sanhueza-Matamala) while the second author was affiliated with the Institute of Computer Sciences of the Czech Academy of Sciences; and by ANID/CONICYT Acuerdo Bilateral DAAD/62170017 through a Ph.D. Scholarship (S. Piga).

## Appendix A: Proof of Lemma 7.3

### Appendix A: Proof of Lemma 7.3

**Proof **The proof proceeds in three steps. First, we find \(H_p\subseteq H\) by including each edge with probability *p*, and in the remainder \(H_0 = H \setminus H_p\) we find an almost perfect \(C_{\ell }\)-packing \({\mathcal {C}}_0\), let \(L_0 = H_0 \setminus E({\mathcal {C}}_0)\) be the leftover edges. Secondly, we correct the leftover \(L_0\) in the vertices incident with \(\Omega (n^2)\) many edges of \(L_0\) by constructing cycles with the help of the edges in \(H_p\). This provides us with a new cycle packing \({\mathcal {C}}_1\subseteq L_0 \cup H_p\) whose new leftover \(L_1 = H_0 \setminus E({\mathcal {C}}_0 \cup {\mathcal {C}}_1)\) satisfies \(\Delta _1(L_1) = o(n^2)\). Finally, we correct the new leftover \(L_1\) in a similar way, fixing the pairs incident to \(\Omega (n)\) edges in \(L_1\). We get a cycle packing \({\mathcal {C}}_2\subseteq L_1 \cup H_p\), and \({\mathcal {C}}_0 \cup {\mathcal {C}}_1 \cup {\mathcal {C}}_2\) will be the desired cycle packing.

*Step 1: Random slice and aproximate decomposition.* Note that \(\delta ^{(3)}_2(H) \ge 3 \varepsilon n\). Now let \(p = \gamma / 4\), and let \(H_p\subseteq H\) be obtained from *H* by including each edge independently with probability *p*. Using concentration inequalities (e.g. Theorem 5.4) we see that with non-zero probability

hold simultaneously for \(H_p\). From now on we suppose \(H_p\) is fixed and satisfies (A.1).

Let \(H_0 = H \setminus H_p\). In \(H_0\), construct a \(C_{\ell }\)-packing by removing edge-disjoint cycles, one by one, until no longer possible. We get a \(C_\ell \)-packing \({\mathcal {C}}_0\) in \(H_0\), let \(F_0 = E({\mathcal {C}}_0)\). By Erdős’ Theorem [18, Theorem 1] there exists \(c > 0\) such that \(L_0 = H_0 \setminus F_0\) has at most \(n^{3 - 3c}\) edges.

*Step 2: Eliminating bad vertices.* Let \(B_0 = \{ v \in V: \deg _{L_0}(v) \ge n^{2 - 2c} \}\). Since \(|L_0| \le n^{3 - 3c}\), by double-counting we have \(|B_0| \le 3 n^{1 - c}\).

For each \(b \in B_0\), let \(G_b\) be the subgraph of \(L_0(b)\) obtained after removing the vertices of \(B_0\). Note that \(L_0(b) - G_0\) has at most \(|B_0|n \le 3n^{2-c}\) edges. Now, let \({\mathcal {P}}_b\) be a maximal edge-disjoint collection of paths of length 3 in \(G_b\). Since every graph on *n* vertices with at least \(n+1\) edges contains a path of length 3, then \(G_b - E({\mathcal {P}}_b)\) has at most *n* edges. All together, we deduce that the number of edges in \(L_0(b) - E({\mathcal {P}}_b)\) satisfies

Since \(G_b\) contains at most \(n^2\) edges, we certainly have \(|{\mathcal {P}}_b| \le n^2\). Let \({\mathcal {P}}_b\) be a collection of tight paths on five vertices obtained by replacing each \(v_0 v_1 v_2 v_3\) in \({\mathcal {P}}_b\) with the tight path \(v_0 v_1 b v_2 v_3\) in \(L_0\). Note that any two distinct \(P_1, P_2 \in {\mathcal {P}}_b\) are edge-disjoint, and for two distinct \(b, b' \in B_0\), and \(P \in {\mathcal {P}}_b\), \(P' \in {\mathcal {P}}_{b'}\), since \(b' \notin V(G_b)\) we have \(P, P'\) are edge-disjoint. Thus the union \({\mathcal {P}}= \bigcup _{b \in B_0} {\mathcal {P}}_b\) is an edge-disjoint collection of tight paths on 5 vertices.

Select \(\gamma ', \mu ', \varepsilon '\) such that \(1/n \ll \gamma ' \ll \mu ' \ll \varepsilon ' \ll \gamma , \varepsilon , 1/\ell \). We wish to apply Lemma 7.1 to extend \({\mathcal {P}}\) into cycles. We claim \({\mathcal {P}}\) is \(\gamma '\)-sparse. Let \(S \in \left( {\begin{array}{c}V(H)\\ 2\end{array}}\right) \). Since \(|{\mathcal {P}}| \le |B_0| n^2 \le 3 n^{3 - c} \le \gamma ' n^{3}\), certainly \({\mathcal {P}}\) contains at most \(|{\mathcal {P}}| \le \gamma ' n^{3}\) paths of type 0 for *S*. Now, note that for each \(b \in B_0\), \(P \in {\mathcal {P}}_b\) can have at most 2*n* paths of type 1 for *S*, thus \({\mathcal {P}}\) has at most \(|B_0|2n \le 6n^{2-c} \le \gamma ' n^{2}\) paths of type 1 for *S*. Analogously, for each \(b \in B_0\), \(P \in {\mathcal {P}}_b\) can have at most 1 path of type 2 for *S*, thus \({\mathcal {P}}\) has at most \(|B_0| \le 3n^{1-c} \le \gamma ' n\) paths of type 2 for *S*. Thus \({\mathcal {P}}\) is \(\gamma '\)-sparse.

Recall that \(L_0\) is edge-disjoint with \(H_p\). Inequalities (A.1) together with \(p = \gamma /4\) and \(\varepsilon ' \ll \gamma , 1/\ell \), show that we can use Corollary 5.3 (with \(U = V(H)\)) and deduce that for each \(P \in {\mathcal {P}}\), there exists at least \(\varepsilon ' n^{\ell - 5}\) copies of \(C_{\ell }\) in \(L_0 \cup H_p\) that extend \(P_i\) using extra edges of \(H_p\) only.

We apply Lemma 7.1 with \(\varepsilon ', \mu ', \gamma ', \ell , 5, L_0, H_p, {\mathcal {P}}\) playing the rôle of \(\varepsilon , \mu , \gamma , \ell , \ell ', H_1, H_2, {\mathcal {P}}\) respectively, to obtain a \(C_{\ell }\)-decomposable graph \(F_1\subseteq L_0 \cup H_p\) such that \(E({\mathcal {P}})\subseteq F_1\) and

Since \(F_0\), \(F_1\) are edge-disjoint, \(F_0 \cup F_1\) is \(C_{\ell }\)-decomposable. Let \(L_1 = H_0 \setminus ( F_0 \cup F_1 )\). Observe that, if \(v \notin B_0\), then \(\deg _{L_1}(v) \le \deg _{L_0}(v) <n^{2-2c}\) by definition. Moreover, if \(v \in B_0\), then each edge in \(E({\mathcal {P}}_v)\) is in \(F_1\), and hence (A.2) implies \(\deg _{L_1}(v) \le |L_0(v)| - |E({\mathcal {P}}_v)| \le 4n^{2-c}\). Therefore,

*Step 3: Eliminating bad pairs.* Let \(f = c/2\) and \(B_1 = \{ xy \in \left( {\begin{array}{c}V\\ 2\end{array}}\right) : \deg _{L_1}(xy) \ge n^{1 - f} \}\). From \(|L_1| \le |L_0| \le n^{3 - 3c} \le n^{3 - 6f}\) we deduce \(|B_1| \le n^{2 - 4f}\). Now consider \(B_1\) as the set of edges of a 2-graph in *V*. Each edge of \(B_1\) incident to a vertex *x* implies that *x* belongs to at least \(n^{1 - f}\) edges in \(L_1\), and each of those edges participates in at most two of the edges in \(B_1\) incident to *x*. So we have \(\deg _{L_1}(x) \ge \frac{1}{2} n^{1 - f} \deg _{B_1}(x)\). Together with inequality (A.4) we deduce \(\Delta (B_1) \le 8 n^{1 - f}\).

A path *P* on \(L_1\) is \(B_1\)-*based* if \(P =zxyw\) and \(xy \in B_1\). Let \({\mathcal {P}}_2\) be a maximal packing of \(B_1\)-based paths. For all \(xy \in B_1\), it holds that \(\deg _{L_1}(xy) - \deg _{E({\mathcal {P}}_2)}(xy) \le 1\). Otherwise it would exist distinct \(z, w \in N_{L_1 \setminus E({\mathcal {P}}_2)}(xy)\), and then *zxyw* would a \(B_1\)-based path not in \({\mathcal {P}}_2\), which contradicts its maximality.

We claim \({\mathcal {P}}_2\) is \(\gamma '\)-sparse. For each \(xy \in B_1\), let \({\mathcal {P}}_{xy}\subseteq {\mathcal {P}}_2\) be the paths whose two interior vertices are precisely *xy*. Clearly \(|{\mathcal {P}}_{xy}| \le n\) and \({\mathcal {P}}_2 = \bigcup _{xy \in B_1} {\mathcal {P}}_{xy}\). Let \(e \in \left( {\begin{array}{c}V\\ 2\end{array}}\right) \). Since \(|{\mathcal {P}}_2| \le \sum _{xy \in B_1} |{\mathcal {P}}_{xy}| \le n |B_1| \le n^{3 - 4f} \le \gamma ' n^3\), there are at most \(\gamma ' n^3\) paths of type 0 for *e* in \({\mathcal {P}}_2\). Recall that if \(P = zxyw\) is a path of type 1 for *e*, then we have \(|e \cap \{z, x, y, w\}| = 1\). If \(xy \in B_1\) satisfies \(e \cap \{x, y\} = \emptyset \), then at most two paths in \({\mathcal {P}}_{xy}\) can be of type 1 for *e* and therefore there are at most \(2|B_1| \le 2n^{2 - 4f}\) paths of type 1 for *e* in \({\mathcal {P}}_2\). We estimate the contribution of the pairs \(xy \in B_1\) such that \(|e \cap \{x,y\}| = 1\). Each such *xy* contributes at most *n* paths of type 1 for *e* in \({\mathcal {P}}_{xy}\). By (A.4), the number of such *xy* is at most \(2 \Delta (B_1) \le 16 n^{1 - f}\), thus the total contribution of those pairs is at most \(16 n^{2-f}\). All together, the total number of paths of type 1 for *e* in \({\mathcal {P}}_2\) is at most \(2n^{2 - 4f}+16 n^{2-f} \le \gamma ' n^2\). If \(e = \{a,b\}\) then \({\mathcal {P}}_{a,b}\) does not contain any path of type 2 for *e*, by definition of the path types. Thus the only possible contributions come from the pairs in \({\mathcal {P}}_{a,x}\) and \({\mathcal {P}}_{b,y}\) for some \(x, y \in V(H)\); and each one of those sets contains at most 1 path of type 2 for *e*. Thus the total number of pairs of type 2 for *e* in \({\mathcal {P}}_2\) is at most \(2 \Delta (B_1) \le 16 n^{1 - f} \le \gamma ' n\). Thus \({\mathcal {P}}_2\) is \(\gamma '\)-sparse.

Let \(H'_p = H_p \setminus ( F_0 \cup F_1 )\). (A.1) and (A.3), together with \(\mu ' \ll \varepsilon ' \ll \gamma , 1/\ell \), allow us to use Corollary 5.3 with \(U = V(H)\), thus for each \(P \in {\mathcal {P}}_2\), there exist at least \(\varepsilon ' n^{\ell - 4}\) copies of \(C_{\ell }\) in \(L_1 \cup H'_p\) that extend *P* using extra edges of \(H'_p\) only. Apply Lemma 7.1 with the parameters \(\varepsilon ', \mu ', \gamma ', \ell , 4, L_1, H'_p, {\mathcal {P}}_2\) playing the roles of \(\varepsilon , \mu , \gamma , \ell , \ell ', H_1, H_2, {\mathcal {P}}\) respectively, to obtain a \(C_{\ell }\)-decomposable \(F_2\subseteq L_1 \cup H'_p\) such that \(E({\mathcal {P}}_2)\subseteq F_2\) and \(\Delta _2(F_2 \setminus E({\mathcal {P}}_2)) \le \mu 'n\).

We claim that \(\Delta _2(L_1 \setminus F_2) \le n^{1 - f}\). Indeed, if \(xy \in B_1\), \(\deg _{L_1 \setminus F_2}(xy) \le \deg _{L_1}(xy) \le n^{1-f}\) follows by definition, otherwise, \(E({\mathcal {P}}_2) \subseteq F_2\) implies \(\deg _{L_1 \setminus F_2}(xy) \le \deg _{L_1}(xy) - \deg _{F_2}(xy) \le 1\). Since \(F_2\) and \(F_0 \cup F_1\) are edge-disjoint, \(F = F_0 \cup F_1 \cup F_2\) is a \(C_{\ell }\)-decomposable subgraph of *H*. We claim \(L = H \setminus F\) satisfies \(\Delta _2(L) \le \gamma n\). Indeed, an edge not covered by *F* is either in \(H_p\) or in \(L_1 \setminus F_2\). Thus we have \(\Delta _2(L) \le \Delta _2(H_p) + \Delta _2(L_1 \setminus F_2) \le 2pn + n^{1-f} \le \gamma n\), as required. \(\square \)

## Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

## About this article

### Cite this article

Piga, S., Sanhueza-Matamala, N. Cycle Decompositions in 3-Uniform Hypergraphs.
*Combinatorica* **43**, 1–36 (2023). https://doi.org/10.1007/s00493-023-00001-2

Received:

Revised:

Accepted:

Published:

Issue Date:

DOI: https://doi.org/10.1007/s00493-023-00001-2