In graph theory, the graph removal lemma states that when a graph contains few copies of a given subgraph, then all of the copies can be eliminated by removing a small number of edges.[1] The special case in which the subgraph is a triangle is known as the triangle removal lemma.[2]

The graph removal lemma can be used to prove Roth's theorem on 3-term arithmetic progressions,[3] and a generalization of it, the hypergraph removal lemma, can be used to prove Szemerédi's theorem.[4] It also has applications to property testing.[5]

Formulation

edit

Let   be a graph with   vertices. The graph removal lemma states that for any  , there exists a constant   such that for any  -vertex graph   with fewer than   subgraphs isomorphic to  , it is possible to eliminate all copies of   by removing at most   edges from  .[1]

An alternative way to state this is to say that for any  -vertex graph   with   subgraphs isomorphic to  , it is possible to eliminate all copies of   by removing   edges from  . Here, the   indicates the use of little o notation.

In the case when   is a triangle, resulting lemma is called triangle removal lemma.

History

edit

The original motivation for the study of triangle removal lemma was the Ruzsa–Szemerédi problem. Its initial formulation due to Imre Z. Ruzsa and Szemerédi from 1978 was slightly weaker than the triangle removal lemma used nowadays and can be roughly stated as follows: every locally linear graph on   vertices contains   edges.[6] This statement can be quickly deduced from a modern triangle removal lemma. Ruzsa and Szemerédi provided also an alternative proof of Roth's theorem on arithmetic progressions as a simple corollary.[6]

In 1986, during their work on generalizations of the Ruzsa–Szemerédi problem to arbitrary  -uniform graphs, Erdős, Frankl, and Rödl provided a statement for general graphs very close to the modern graph removal lemma: if graph   is a homomorphic image of  , then any  -free graph   on   vertices can be made  -free by removing   edges.[7]

The modern formulation of the graph removal lemma was first stated by Füredi in 1994.[8] The proof generalized earlier approaches by Ruzsa and Szemerédi and Erdős, Frankl, and Rödl, also using the Szemerédi regularity lemma.

Graph counting lemma

edit

A key component of the proof of the graph removal lemma is the graph counting lemma about counting subgraphs in systems of regular pairs. The graph counting lemma is also very useful on its own. According to Füredi, it is used "in most applications of regularity lemma".[8]

Heuristic argument

edit

Let   be a graph on   vertices, whose vertex set is   and edge set is  . Let   be sets of vertices of some graph   such that, for all  , the pair   is  -regular (in the sense of the regularity lemma). Let also   be the density between the sets   and  . Intuitively, a regular pair   with density   should behave like a random Erdős–Rényi-like graph, where every pair of vertices   is selected to be an edge independently with probability  . This suggests that the number of copies of   on vertices   such that   should be close to the expected number from the Erdős–Rényi model, where   and   are the edge set and the vertex set of  .

Precise statement

edit

The straightforward formalization of the above heuristic claim is as follows. Let   be a graph on   vertices, whose vertex set is   and whose edge set is  . Let   be arbitrary. Then there exists   such that for any   as above, satisfying   for all  , the number of graph homomorphisms from   to   such that vertex   is mapped to   is not smaller than 

Blow-up Lemma

edit

One can even find bounded-degree subgraphs of blow-ups of   in a similar setting. The following claim appears in the literature under name of the blow-up lemma and was first proven by Komlós, Sárközy, and Szemerédi.[9] The precise statement here is a slightly simplified version due to Komlós, who referred to it also as the key lemma, as it is used in numerous regularity-based proofs.[10]

Let   be an arbitrary graph and let  . Construct   by replacing each vertex   of   by an independent set   of size   and replacing every edge   of   by the complete bipartite graph on  . Let   be arbitrary reals, let   be a positive integer, and let   be a subgraph of   with   vertices and maximum degree  . Define  . Finally, let   be a graph and   be disjoint sets of vertices of   such that, whenever  , then   is a  -regular pair with density at least  . Then if   and  , then the number of injective graph homomorphisms from   to   is at least  .

In fact, one can restrict to counting only those homomorphisms such that any vertex   of   with   is mapped to a vertex in  .

Proof

edit

We will provide a proof of the counting lemma in the case when   is a triangle (triangle counting lemma). The proof of the general case, as well as the proof of the blow-up lemma, are very similar and do not require different techniques.

Take  . Let   be the set of those vertices in   which have at least   neighbors in   and at least   neighbors in  . Note that if there were more than   vertices in   with less than   neighbors in  , then these vertices together with the whole   would witness  -irregularity of the pair  . Repeating this argument for   shows that we must have  . Now take an arbitrary   and define   and   as neighbors of   in   and  , respectively. By definition,   and  , so by the regularity of   we obtain existence of at least triangles containing  . Since   was chosen arbitrarily from the set   of size at least  , we obtain a total of at least which finishes the proof as  .

Proof

edit

Proof of the triangle removal lemma

edit

To prove the triangle removal lemma, consider an  -regular partition   of the vertex set of  . This exists by the Szemerédi regularity lemma. The idea is to remove all edges between irregular pairs, low-density pairs, and small parts, and prove that if at least one triangle still remains, then many triangles remain. Specifically, remove all edges between parts   and   if

  •   is not  -regular,
  • the density   is less than  , or
  • either   or   has at most   vertices.

This procedure removes at most   edges. If there exists a triangle with vertices in   after these edges are removed, then the triangle counting lemma tells us there are at least triples in   which form a triangle. Thus, we may take 

Proof of the graph removal lemma

edit

The proof of the case of general   is analogous to the triangle case, and uses the graph counting lemma instead of the triangle counting lemma.

Induced Graph Removal Lemma

edit

A natural generalization of the graph removal lemma is to consider induced subgraphs. In property testing, it is often useful to consider how far a graph is from being induced H-free.[11] A graph   is considered to contain an induced subgraph   if there is an injective map   such that   is an edge of   if and only if   is an edge of  . Notice that non-edges are considered as well.   is induced  -free if there is no induced subgraph  . We define   as  -far from being induced  -free if we cannot add or delete   edges to make   induced  -free.

Formulation

edit

A version of the graph removal lemma for induced subgraphs was proved by Alon, Fischer, Krivelevich, and Szegedy in 2000.[12] It states that for any graph   with   vertices and  , there exists a constant   such that, if an  -vertex graph   has fewer than   induced subgraphs isomorphic to  , then it is possible to eliminate all induced copies of   by adding or removing fewer than   edges.

The problem can be reformulated as follows: Given a red-blue coloring   of the complete graph   (analogous to the graph   on the same   vertices where non-edges are blue and edges are red), and a constant  , then there exists a constant   such that for any red-blue colorings of   has fewer than   subgraphs isomorphic to  , then it is possible to eliminate all copies of   by changing the colors of fewer than   edges. Notice that our previous "cleaning" process, where we remove all edges between irregular pairs, low-density pairs, and small parts, only involves removing edges. Removing edges only corresponds to changing edge colors from red to blue. However, there are situations in the induced case where the optimal edit distance involves changing edge colors from blue to red as well. Thus, the regularity lemma is insufficient to prove the induced graph removal lemma. The proof of the induced graph removal lemma must take advantage of the strong regularity lemma.[12]

Proof

edit

Strong Regularity Lemma

edit

The strong regularity lemma[12] is a strengthened version of Szemerédi's regularity lemma. For any infinite sequence of constants  , there exists an integer   such that for any graph  , we can obtain two (equitable) partitions   and   such that the following properties are satisfied:

  •   refines  ; that is, every part of   is the union of some collection of parts in  .
  •   is  -regular and   is  -regular.
  •  
  •  

The function   is defined to be the energy function defined in Szemerédi regularity lemma. Essentially, we can find a pair of partitions   where   is extremely regular compared to  , and at the same time   are close to each other. This property is captured in the third condition.

Corollary of the Strong Regularity Lemma
edit

The following corollary of the strong regularity lemma is used in the proof of the induced graph removal lemma.[12] For any infinite sequence of constants  , there exists   such that there exists a partition   and subsets   for each   where the following properties are satisfied:

  •  
  •   is  -regular for each pair  
  •   for all but   pairs  

The main idea of the proof of this corollary is to start with two partitions   and   that satisfy the Strong Regularity Lemma where  . Then for each part  , we uniformly at random choose some part   that is a part in  . The expected number of irregular pairs   is less than 1. Thus, there exists some collection of   such that every pair is  -regular!

The important aspect of this corollary is that every pair of   is  -regular! This allows us to consider edges and non-edges when we perform our cleaning argument.

Proof Sketch of the Induced Graph Removal Lemma

edit

With these results, we are able to prove the induced graph removal lemma. Take any graph   with   vertices that has less than   copies of  . The idea is to start with a collection of vertex sets   which satisfy the conditions of the Corollary of the Strong Regularity Lemma.[12] We then can perform a "cleaning" process where we remove all edges between pairs of parts   with low density, and we can add all edges between pairs of parts   with high density. We choose the density requirements such that we added/deleted at most   edges.

If the new graph has no copies of  , then we are done. Suppose the new graph has a copy of  . Suppose the vertex   is embedded in  . Then if there is an edge connecting   in  , then   does not have low density. (Edges between   were not removed in the cleaning process.) Similarly, if there is not an edge connecting   in  , then   does not have high density. (Edges between   were not added in the cleaning process.)

Thus, by a similar counting argument to the proof of the triangle counting lemma (that is, the graph counting lemma), we can show that   has more than   copies of  .

Generalizations

edit

The graph removal lemma was later extended to directed graphs[5] and to hypergraphs.[4]

Quantitative bounds

edit

The usage of the regularity lemma in the proof of the graph removal lemma forces   to be extremely small, bounded by a tower function whose height is polynomial in  ; that is,   (here   is the tower of twos of height  ). A tower function of height   is necessary in all regularity proofs, as is implied by results of Gowers on lower bounds in the regularity lemma.[13] However, in 2011, Fox provided a new proof of the graph removal lemma which does not use the regularity lemma, improving the bound to   (here   is the number of vertices of the removed graph  ).[1] His proof, however, uses regularity-related ideas such as energy increment, but with a different notion of energy, related to entropy. This proof can be also rephrased using the Frieze-Kannan weak regularity lemma as noted by Conlon and Fox.[11] In the special case of bipartite  , it was shown that   is sufficient.

There is a large gap between the available upper and lower bounds for   in the general case. The current best result true for all graphs   is due to Alon and states that, for each nonbipartite  , there exists a constant   such that   is necessary for the graph removal lemma to hold, while for bipartite  , the optimal   has polynomial dependence on  , which matches the lower bound. The construction for the nonbipartite case is a consequence of Behrend construction of large Salem-Spencer sets. Indeed, as the triangle removal lemma implies Roth's theorem, existence of large Salem-Spencer sets may be translated to an upper bound for   in the triangle removal lemma. This method can be leveraged for arbitrary nonbipartite   to give the aforementioned bound.

Applications

edit

Additive combinatorics

edit

Graph theory

edit
  • The graph counting/removal lemma can be used to provide a quick and transparent proof of the Erdős–Stone theorem starting from Turán's theorem and to extend the result to Simonovits stability: for any graph   and any  , there exists   such that any  -free graph on   vertices with at least   edges can be transformed into a complete  -partite Turán graph   by adding or deleting at most   edges (here   is the chromatic number of  ).[15] Although both results had been proven earlier using more elementary techniques (the Erdős–Stone theorem was proved in 1966[16] by Erdős and Stone while Simonovits stability was shown in the same year by various authors[16][17][18][19]), the regularity proof provides a different viewpoint and elucidates connection with other modern proofs.
  • The graph removal lemma together with the Erdős–Stone theorem may be used to show that the number of non-isomorphic  -free graphs on   vertices is equal to

     

    where   is the Turán density of  .[7]

Property testing

edit
  • The graph removal lemma has applications for property testing, because it implies that for every graph, either the graph is near an  -free graph, or random sampling will easily find a copy of   in the graph.[5] One result is that for any fixed  , there is a constant-time algorithm that determines with high probability whether or not a given  -vertex graph   is  -far from being  -free.[20] Here,  -far from being  -free means that at least   edges must be removed to eliminate all copies of   in  .
  • The induced graph removal lemma was formulated by Alon, Fischer, Krivelevich, and Szegedy to characterize testable graph properties.[12]

See also

edit

References

edit
  1. ^ a b c Fox, Jacob (2011), "A new proof of the graph removal lemma", Annals of Mathematics, Second Series, 174 (1): 561–579, arXiv:1006.1300, doi:10.4007/annals.2011.174.1.17, MR 2811609, S2CID 8250133
  2. ^ Trevisan, Luca (May 13, 2009), "The Triangle Removal Lemma", in theory
  3. ^ a b Roth, K. F. (1953), "On certain sets of integers", Journal of the London Mathematical Society, 28 (1): 104–109, doi:10.1112/jlms/s1-28.1.104, MR 0051853
  4. ^ a b c Tao, Terence (2006), "A variant of the hypergraph removal lemma", Journal of Combinatorial Theory, Series A, 113 (7): 1257–1280, arXiv:math/0503572, doi:10.1016/j.jcta.2005.11.006, MR 2259060, S2CID 14337591
  5. ^ a b c Alon, Noga; Shapira, Asaf (2004), "Testing subgraphs in directed graphs", Journal of Computer and System Sciences, 69 (3): 353–382, doi:10.1016/j.jcss.2004.04.008, MR 2087940
  6. ^ a b Ruzsa, I. Z.; Szemerédi, E. (1978), "Triple systems with no six points carrying three triangles", Combinatorics (Proc. Fifth Hungarian Colloq., Keszthely, 1976), Vol. II, Colloq. Math. Soc. János Bolyai, vol. 18, Amsterdam and New York: North-Holland, pp. 939–945, MR 0519318
  7. ^ a b Erdős, P.; Frankl, P.; Rödl, V. (1986), "The asymptotic number of graphs not containing a fixed subgraph and a problem for hypergraphs having no exponent", Graphs and Combinatorics, 2 (2): 113–121, doi:10.1007/BF01788085, MR 0932119, S2CID 16839886
  8. ^ a b Füredi, Zoltán (1995). "Extremal Hypergraphs and Combinatorial Geometry". In Chatterji, S. D. (ed.). Proceedings of the International Congress of Mathematicians. Basel: Birkhäuser. pp. 1343–1352. doi:10.1007/978-3-0348-9078-6_129. ISBN 978-3-0348-9078-6.
  9. ^ Komlós, János; Sárközy, Gábor N.; Szemerédi, Endre (1997-03-01). "Blow-up Lemma". Combinatorica. 17 (1): 109–123. doi:10.1007/BF01196135. ISSN 1439-6912. S2CID 6720143.
  10. ^ Komlós, János (1999). "The Blow-up Lemma". Combinatorics, Probability and Computing. 8 (1–2): 161–176. doi:10.1017/S0963548398003502. ISSN 1469-2163. S2CID 6720143.
  11. ^ a b Conlon, David; Fox, Jacob (2013), "Graph removal lemmas", in Blackburn, Simon R.; Gerke, Stefanie; Wildon, Mark (eds.), Surveys in Combinatorics 2013, London Mathematical Society Lecture Note Series, vol. 409, Cambridge, UK: Cambridge University Press, pp. 1–49, arXiv:1211.3487, doi:10.1017/CBO9781139506748.002, ISBN 978-1-107-65195-1, MR 3156927, S2CID 2658118
  12. ^ a b c d e f Alon, Noga; Fischer, Eldar; Krivelevich, Michael; Szegedy, Mario (2000), "Efficient testing of large graphs", Combinatorica, 20 (4): 451–476, doi:10.1007/s004930070001, MR 1804820, S2CID 44645628
  13. ^ Gowers, W. T. (1997). "Lower bounds of tower type for Szemerédi's uniformity lemma". Geometric and Functional Analysis. 7 (2): 322–337. doi:10.1007/PL00001621. S2CID 115242956.
  14. ^ Solymosi, J. (2003), "Note on a generalization of Roth's theorem", Discrete and Computational Geometry, Algorithms and Combinatorics, 25: 825–827, doi:10.1007/978-3-642-55566-4_39, ISBN 978-3-642-62442-1, MR 2038505, S2CID 53973423
  15. ^ Alon, N. (2001-10-14). "Testing subgraphs in large graphs". Proceedings 42nd IEEE Symposium on Foundations of Computer Science. FOCS '01. USA: IEEE Computer Society. pp. 434–441. doi:10.1109/SFCS.2001.959918. ISBN 978-0-7695-1390-4. S2CID 12484006.
  16. ^ a b Erdős, P.; Simonovits, M. (1966). "A limit theorem in graph theory". Studia Sci. Math. Hung. 1: 51–57.
  17. ^ Erdős, P. (1966). "Some recent results on extremal problems in graph theory". Theory of Graphs, International Symp. Rome: 118–123.
  18. ^ Erdős, P. (1966). "On some new inequalities concerning extremal properties of graphs". Theory of Graphs, Proc. Coll. Tihany, Hungary: 77–81.
  19. ^ Erdős, P.; Katona, G. (1966). "A method for solving extremal problems in graph theory". Theory of Graphs, Proc. Coll. Tihany: 279–319.
  20. ^ Alon, Noga; Shapira, Asaf (2008), "A characterization of the (natural) graph properties testable with one-sided error", SIAM Journal on Computing, 37 (6): 1703–1727, doi:10.1137/06064888X, MR 2386211