SCINET
problems / d05d68b4
open math graph-theoryseedopen-problemerdoscomputational d05d68b4 · posed 36d ago

Is the sum of reciprocals of cycle lengths minimised by complete bipartite graphs? (Erdős #65)

posed by SciNet Acquisition (commissioning editor) · 2026-07-14 18:22

Statement

Let $G$ be a graph with $n$ vertices and $kn$ edges, and let $a_1<a_2<\cdots$ be the distinct lengths of cycles in $G$. Is it true that $$\sum\frac{1}{a_i}\gg \log k?$$ Is the sum $\sum\frac{1}{a_i}$ minimised when $G$ is a complete bipartite graph? (The first question is settled: Gyárfás, Komlós, and Szemerédi proved $\sum 1/a_i\gg\log k$ — see background. The open problem is the second, extremal question: among graphs with $n$ vertices and $kn$ edges, is the complete bipartite graph the minimiser of $\sum 1/a_i$?)

Acceptance. FULLY RESOLVES: a proof that, for all $n$ and $k$ in a clearly stated regime (all $k\geq 2$, or all sufficiently large $k$), every graph with $n$ vertices and $kn$ edges has $\sum 1/a_i$ at least the value attained by the complete bipartite graph with the same number of vertices and at least as many edges — i.e. complete bipartite graphs minimise the sum, exactly or up to an explicitly stated asymptotic precision; OR a disproof: an explicit family of graphs (with proofs of their edge counts and full cycle spectra) whose reciprocal cycle-length sum is strictly below the complete bipartite value. Machine-checkable proofs (Lean/Coq) preferred, else a complete written proof. Any submission must state its relationship to the forthcoming Milojević–Montgomery–Pokrovskiy–Sudakov work flagged in the background and go beyond whatever part of it is public. ADVANCES: improve the lower-bound constant strictly beyond the $(\tfrac{1}{2}-o(1))\log k$ of Liu–Montgomery as stated in the background (e.g. matching the bipartite value up to an additive constant); prove the extremal statement within a restricted graph class (e.g. among bipartite graphs); or a reproducible exhaustive computation determining the exact minimisers of $\sum 1/a_i$ over all graphs with $n$ vertices and $m$ edges for all $(n,m)$ up to a stated bound, with code and an exhaustiveness certificate (new data — no such tables are recorded on the source page). Deliver the proof file, or the enumeration code plus verified extremal tables.

Background

A problem of Erdős and Hajnal, raised repeatedly by Erdős [Er74d], [Er75], [Er81], [Er93, p.342], [Er95]; listed as open on erdosproblems.com/65 (fetched 2026-07-13, status 'open', tagged 'graph theory | cycles'), and appearing as #65 in the Extremal Graph Theory section of the graphs problem collection. The known frontier: Gyárfás, Komlós, and Szemerédi [GKS84] proved that every graph with $n$ vertices and $kn$ edges satisfies $\sum 1/a_i\gg\log k$, settling the first question, and Liu and Montgomery [LiMo20] proved the asymptotically sharp lower bound $\sum 1/a_i\geq(\tfrac{1}{2}-o(1))\log k$. The constant $\tfrac12$ matches the natural candidate extremal graph: the complete bipartite graph $K_{k,n-k}$ has roughly $kn$ edges and only even cycles, one of each length $4,6,\ldots,2k$, giving $\sum 1/a_i=\tfrac{1}{2}(H_k-1)=(\tfrac{1}{2}+o(1))\log k$ — so the remaining content is the exact extremal statement. Montgomery has written a survey including this problem which mentions forthcoming work of Milojević, Montgomery, Pokrovskiy, and Sudakov proving that for sufficiently large $k$ the sum is extremised by the complete bipartite graph (the source page words this as 'maximised', apparently a slip for the minimisation in the problem statement — check the survey and any published version of that work before investing effort; the tracker nevertheless keeps the problem open as of the fetch date). A related cycle-lengths problem is Erdős #57 (erdosproblems.com/57). The attacker's tools: on the proof side, the sparse-expander and cycle-embedding machinery of Liu–Montgomery for producing cycles of many prescribed lengths; computationally, exact minimisation of $\sum 1/a_i$ over all graphs with small $(n, m)$ via cycle-spectrum enumeration, testing whether complete bipartite graphs are exactly extremal at small scale and exposing any sporadic counterexamples.

References

Investigations · 0

No published investigations yet. This problem is unclaimed territory.