The biclique partition number bp(G)\mathrm{bp}(G) is the minimum number of complete bipartite subgraphs (bicliques) whose edge sets partition the edge set of GG. For split graphs — graphs whose vertices partition into a clique KK and a stable set SS — Lyu and Hicks proved mc(Gc)−2≤bp(G)≤mc(Gc)−1\mathrm{mc}(G^c)-2\le\mathrm{bp}(G)\le\mathrm{mc}(G^c)-1, where mc(Gc)\mathrm{mc}(G^c) is the number of maximal cliques of the complement, and posed as an open problem whether bp(G)=mc(Gc)−1\mathrm{bp}(G)=\mathrm{mc}(G^c)-1 always holds for co-chordal or split graphs, which would extend the Graham–Pollak theorem. Babu and Jacob first announced an affirmative answer for all split graphs, then withdrew it, and in April 2026 posted a counterexample: a 14-vertex balanced split graph with bp(G)=6=mc(Gc)−2\mathrm{bp}(G)=6=\mathrm{mc}(G^c)-2, together with an infinite family. Their note settles unbalanced split graphs completely (bp(G)=ω(G)−1\mathrm{bp}(G)=\omega(G)-1) and proves bp(G)≤ω(G)\mathrm{bp}(G)\le\omega(G) for balanced split graphs, but leaves the balanced case open. The problem: for a balanced split graph — a split graph admitting a split partition (K,S)(K,S) into a clique KK and a stable set SS with ∣K∣=ω(G)|K|=\omega(G) and ∣S∣=α(G)|S|=\alpha(G), equivalently ω(G)+α(G)=∣V(G)∣\omega(G)+\alpha(G)=|V(G)| — decide whether bp(G)=ω(G)−1\mathrm{bp}(G)=\omega(G)-1 or bp(G)=ω(G)\mathrm{bp}(G)=\omega(G), and characterize structurally which bipartite adjacency between KK and SS forces each outcome. The problem is open.

Background

A biclique is a complete bipartite subgraph, and the biclique partition number bp(G)\mathrm{bp}(G) of a graph GG is the minimum number of bicliques whose edge sets partition the edge set of GG. The celebrated Graham–Pollak theorem (1972) states that the edges of the complete graph KnK_n cannot be partitioned into fewer than n−1n-1 complete bipartite graphs, i.e. bp(Kn)=n−1\mathrm{bp}(K_n)=n-1. A clique is a set of pairwise adjacent vertices (its maximum size is the clique number ω(G)\omega(G)) and a stable set is a set of pairwise non-adjacent vertices (its maximum size is the independence number α(G)\alpha(G)). A split graph is a graph whose vertex set partitions into a clique KK and a stable set SS; split graphs are exactly the graphs that are simultaneously chordal (no induced cycle longer than a triangle) and co-chordal (complement of a chordal graph). A maximal clique is a clique not contained in a larger one; write mc(Gc)\mathrm{mc}(G^c) for the number of maximal cliques of the complement GcG^c. For co-chordal graphs Lyu and Hicks proved bp(G)<mc(Gc)\mathrm{bp}(G)<\mathrm{mc}(G^c), and for split graphs the two-sided bound mc(Gc)−2≤bp(G)≤mc(Gc)−1\mathrm{mc}(G^c)-2\le\mathrm{bp}(G)\le\mathrm{mc}(G^c)-1, with exact solvability when GcG^c is chordal and clique-vertex irreducible (a restriction on how the maximal cliques of GcG^c overlap); they posed as an open problem whether the equality bp(G)=mc(Gc)−1\mathrm{bp}(G)=\mathrm{mc}(G^c)-1 holds for every co-chordal graph or split graph. This equality would be a genuine extension of Graham–Pollak, since for G=KnG=K_n the complement is edgeless, mc(Gc)=n\mathrm{mc}(G^c)=n, and the equality reads bp(Kn)=n−1\mathrm{bp}(K_n)=n-1. Babu and Jacob first claimed the affirmative answer bp(G)=mc(Gc)−1\mathrm{bp}(G)=\mathrm{mc}(G^c)-1 for all split graphs (arXiv:2507.08114, July 2025), but withdrew the preprint in March 2026 with the comment that a key lemma was 'not correct in its entirety'. In April 2026 they posted a counterexample note (arXiv:2604.05491): a 14-vertex balanced split graph with bp(G)=6=ω(G)−1=mc(Gc)−2\mathrm{bp}(G)=6=\omega(G)-1=\mathrm{mc}(G^c)-2, plus an infinite family of such counterexamples. Here a balanced split graph is a split graph that admits a split partition (K,S)(K,S) with ∣K∣=ω(G)|K|=\omega(G) and ∣S∣=α(G)|S|=\alpha(G), equivalently ω(G)+α(G)=∣V(G)∣\omega(G)+\alpha(G)=|V(G)|; by the Hammer–Simeone trichotomy, every split partition of a split graph is of exactly one of three types — balanced (∣K∣=ω(G)|K|=\omega(G), ∣S∣=α(G)|S|=\alpha(G)), SS-max (∣K∣=ω(G)−1|K|=\omega(G)-1, ∣S∣=α(G)|S|=\alpha(G)), or KK-max (∣K∣=ω(G)|K|=\omega(G), ∣S∣=α(G)−1|S|=\alpha(G)-1) — so the split partition with ∣K∣=ω(G)|K|=\omega(G) and ∣S∣=α(G)|S|=\alpha(G), the balanced split partition, is unique, and in it no vertex of SS is adjacent to all of KK and every vertex of KK has a neighbor in SS. The Babu–Jacob counterexample and its infinite family are the symmetric case ω(G)=α(G)\omega(G)=\alpha(G); the balanced class is strictly larger and also contains graphs with ω(G)≠α(G)\omega(G)\neq\alpha(G). For any balanced split graph one checks that the maximal cliques of GcG^c are exactly SS together with {v}∪(S∖NG(v))\{v\}\cup(S\setminus N_G(v)) for each v∈Kv\in K, where NG(v)N_G(v) denotes the set of neighbors of vv in GG, so mc(Gc)=ω(G)+1\mathrm{mc}(G^c)=\omega(G)+1; hence the Lyu–Hicks bounds and the note's Lemma 7 (bp(G)≤ω(G)\mathrm{bp}(G)\le\omega(G) for balanced split graphs) pin the biclique partition number of a balanced split graph to the two values ω(G)−1\omega(G)-1 and ω(G)\omega(G). Both values occur: the note's infinite family attains ω(G)−1\omega(G)-1, while the 4-vertex path, viewed as a balanced split graph whose clique side is its middle edge and whose stable side is its two endpoints, has bp=2=ω(G)\mathrm{bp}=2=\omega(G). The note settles the unbalanced case completely (bp(G)=ω(G)−1=mc(Gc)−1\mathrm{bp}(G)=\omega(G)-1=\mathrm{mc}(G^c)-1, its Lemma 2) and proves structural restrictions on minimum biclique partitions of balanced split graphs (its Lemmas 3–6, among them: no biclique of a minimum partition is a star centered in SS, and no biclique has one side lying entirely inside SS), but does not determine which of the two values a given balanced split graph attains.

Problem Statement

Let GG be a balanced split graph, i.e. a split graph admitting a split partition (K,S)(K,S) into a clique KK and a stable set SS with ∣K∣=ω(G)|K|=\omega(G) and ∣S∣=α(G)|S|=\alpha(G), equivalently ω(G)+α(G)=∣V(G)∣\omega(G)+\alpha(G)=|V(G)|; this balanced split partition is unique, no vertex of SS is adjacent to all of KK, and every vertex of KK has a neighbor in SS. Published results pin its biclique partition number to ω(G)−1≤bp(G)≤ω(G)\omega(G)-1\le\mathrm{bp}(G)\le\omega(G), equivalently bp(G)∈{mc(Gc)−2,mc(Gc)−1}\mathrm{bp}(G)\in\{\mathrm{mc}(G^c)-2,\mathrm{mc}(G^c)-1\}. Determine which value it attains, by carrying out both of the following: (i) give a criterion that decides, for an arbitrary balanced split graph GG, whether bp(G)=ω(G)−1\mathrm{bp}(G)=\omega(G)-1 or bp(G)=ω(G)\mathrm{bp}(G)=\omega(G) — equivalently, whether the Lyu–Hicks equality bp(G)=mc(Gc)−1\mathrm{bp}(G)=\mathrm{mc}(G^c)-1 holds for GG; and (ii) characterize structurally the balanced split graphs that attain each of the two values — since the balanced split partition is unique, this structure is captured by the bipartite adjacency between the clique side KK and the stable side SS.

References

  1. Anand Babu and Ashwin Jacob, 'A counterexample to the conjecture on Biclique Partition number of Split Graphs and related problems', arXiv:2604.05491, 2026, DOI 10.48550/arXiv.2604.05491, https://arxiv.org/abs/2604.05491
  2. Bochuan Lyu and Illya V. Hicks, 'Finding biclique partitions of co-chordal graphs', Discrete Applied Mathematics 337 (2023) 278–287, DOI 10.1016/j.dam.2023.05.001, https://doi.org/10.1016/j.dam.2023.05.001 (preprint arXiv:2203.02837, https://arxiv.org/abs/2203.02837)
  3. Anand Babu and Ashwin Jacob, 'Exact Biclique Partition number of Split Graphs', arXiv:2507.08114, 2025 (v2 withdrawn March 2026), DOI 10.48550/arXiv.2507.08114, https://arxiv.org/abs/2507.08114
  4. R. L. Graham and H. O. Pollak, 'On embedding graphs in squashed cubes', Lecture Notes in Mathematics (Graph Theory and Applications), 1972, pp. 99–110, DOI 10.1007/bfb0067362, https://doi.org/10.1007/bfb0067362
  5. Peter L. Hammer and Bruno Simeone, 'The splittance of a graph', Combinatorica 1 (1981) 275–284, DOI 10.1007/bf02579333, https://doi.org/10.1007/bf02579333
  6. Zhen-Peng Xu, Jie Wang, Qi Ye, Gereon Koßmann, René Schwonnek and Andreas Winter, 'Simultaneous variances of Pauli strings, weighted independence numbers, and a new kind of perfection of graphs', arXiv:2511.13531, 2025, DOI 10.48550/arXiv.2511.13531, https://arxiv.org/abs/2511.13531
  7. Bochuan Lyu and Illya V. Hicks, 'Maximal Clique and Edge-Ranking Bounds of Biclique Cover Number', arXiv:2302.12775, 2023, DOI 10.48550/arXiv.2302.12775, https://arxiv.org/abs/2302.12775