Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated

Erdos 1997 size largest bipartite subgraphs

../

corollary_3: Shows that every connected graph contains a forest of induced stars covering all of its vertices except at most one.

lemma_1: Shows that edges captured inside bipartite blocks contribute fully while half of all remaining edges can also be retained.

proposition_5: Balances a maximum-matching block partition against one obtained from a one-factorization to prove the exact universal max-cut lower bound.

theorem_4: Bounds the largest bipartite subgraph below by half of the size plus one sixth of the order without isolated vertices, and plus a quarter of one less than the order for connected graphs.


Paul Erdős, András Gyárfás, and Yoshiharu Kohayakawa, “The size of the largest bipartite subgraphs,” Discrete Mathematics 177 (1997), 267-271.

The paper gives a constructive alternative to Edwards's original proof. It partitions a multigraph into an independent block and induced bipartite blocks arising from a maximum matching. Lemma 1 converts the total number of edges inside those blocks into a cut surplus. Proposition 5 chooses two partitions: a maximum-matching one maximizing the internal edge count, and a second, with two-vertex blocks, obtained from a one-factorization. Balancing their two lower bounds yields the exact Edwards formula.

The introduction (p. 268) records that the formula gives f(e)f(e) exactly when e=(m2)e=\binom m2, that e=19e=19 is the first case where it is not tight (f(19)=12f(19)=12 against the formula's 1111), and that the resulting question, whether the surplus of f(e)f(e) over the formula is bounded, has been answered by Alon: for even nn and e=n2/2e=n^2/2 the surplus is at least ce1/4ce^{1/4} for an absolute constant c>0c>0, while a construction of his bounds it above by Ce1/4Ce^{1/4} for every positive ee. Theorem 4 (p. 270) also gives, by the same partition method, the bound b(G)≥e/2+n/6b(G)\geq e/2+n/6 for graphs of order nn without isolated vertices and the connected-graph bound b(G)≥e/2+(n−1)/4b(G)\geq e/2+(n-1)/4, the second of which the paper credits to Edwards (Theorem 6 of his 1973 paper). The connected-graph bound rests on Lemma 2 (p. 269), which for connected GG chooses the matching partition with at most one vertex outside its blocks; the paper deduces from it Corollary 3 (p. 270), a forest of induced stars covering all vertices but at most one.

The copy read for this card is the final published five-page scan hosted on András Gyárfás's publication site at https://www.renyi.hu/~gyarfas/Cikkek/81_largebip.pdf. DOI: https://doi.org/10.1016/S0012-365X(97)00004-6. The file prints "Copyright © 1997 Elsevier Science B.V. All rights reserved" on its first page, every other right reserved.

Results.

  • Lemma 1 (p. 268): block partitions produce a cut with at least (e+s)/2(e+s)/2 edges.
  • [[extremal_graph_theory/erdos_1997_size_largest_bipartite_subgraphs/corollary_3|Corollary 3]] (p. 270), with Lemma 2 (p. 269): connected graphs have a forest of induced stars covering all vertices but at most one.
  • [[extremal_graph_theory/erdos_1997_size_largest_bipartite_subgraphs/theorem_4|Theorem 4]] (p. 270): lower bounds in terms of size and order.
  • [[extremal_graph_theory/erdos_1997_size_largest_bipartite_subgraphs/proposition_5|Proposition 5]] (p. 271): the min-max lower bound from which the paper deduces the Edwards formula.

Bears on.

  • #127: Proposition 5, with Lemma 1, gives a constructive proof of the Edwards lower bound, the baseline over which the problem measures its correction. The paper does not itself address whether the correction is unbounded; it reports Alon's answer.

No file of this source is held: no license on record permits its redistribution, and the card cites the edition it names above.