Fewest edges of a pancyclic graph: pin down h(n) between log_2 n and log_2 n + log_* n (Erdős #1016)
Statement
A graph on $n$ vertices is pancyclic if it contains a cycle on exactly $k$ vertices for every $3 \le k \le n$. Let $h(n)$ be minimal such that there is a pancyclic graph on $n$ vertices with $n + h(n)$ edges. Estimate $h(n)$. In particular, is it true that $$h(n) \ge \log_2 n + \log_* n - O(1),$$ where $\log_* n$ is the iterated logarithm (the number of times $\log_2$ must be applied to $n$ before the result drops to at most $1$)? Together with the known upper bound $h(n) \le \log_2 n + \log_* n + O(1)$, a positive answer would determine $h(n)$ up to an additive constant.
Acceptance. FULLY RESOLVES: a complete proof that $h(n) \ge \log_2 n + \log_* n - O(1)$ (machine-checkable Lean/Coq preferred, else a full written proof), which together with the known upper bound pins $h(n)$ to an additive $O(1)$; OR a proof that $h(n) \le \log_2 n + o(\log_* n) + O(1)$-type improvements of the upper bound showing the conjectured lower bound is false, i.e. constructions with fewer edges, with full proofs of pancyclicity and the counting. ADVANCES: (a) a proof that $h(n) - \log_2 n \to \infty$ (explicitly open per the background), with proof; (b) any lower bound with a strictly larger growth than the $\log_2(n-1) - 1$ bound stated in the background, or any upper-bound improvement strictly below the bound stated in the background, with proof; (c) exact values of $h(n)$ for small $n$ (with witness pancyclic graphs and a proof of minimality by exhaustive search), extending the known data/OEIS A105206, accompanied by the search code and an exhaustiveness certificate; (d) a machine-verified (Lean) proof of either published bound. Deliver the proof file (or compiling Lean sources), or the enumeration code plus the certified table of exact $h(n)$ values and witness graphs.
Background
A problem of Bondy [Bo71], transmitted and pushed by Erdős [Er71]; listed as open on erdosproblems.com/1016 (fetched 2026-07-13, status 'open', tagged 'graph theory | cycles'). Bondy claimed (without details) a proof of $$\log_2(n-1) - 1 \le h(n) \le \log_2 n + \log_* n + O(1).$$ The lower bound follows from a counting argument (a pancyclic graph needs many distinct cycle lengths, hence many edges beyond a Hamilton cycle); a written proof is provided by Griffin [Gr13], and the first published proof of the upper bound appears to be Chapter 4.5 of George, Khodkar, and Wallis [GKW16]. Erdős [Er71] believed the upper bound is the truth, but could not prove even that $h(n) - \log_2 n \to \infty$ — that weaker statement is already open and is the natural first target. The remaining gap is thus an additive $\log_* n$ term: minuscule, but structurally meaningful. Exact values of $h(n)$ for small $n$ relate to OEIS A105206 (linked as related from the problem page), and extending the exact table is a concrete computational contribution: for fixed $n$ and candidate edge count $n+h$, pancyclicity of each candidate graph is directly checkable, so $h(n)$ is decidable by isomorph-free exhaustive search. The attacker's tool: sharpened counting/entropy arguments on the set of cycle lengths for the lower bound (or explicit recursive constructions for the upper bound), plus nauty/geng-style enumeration with per-length cycle checks to extend the exact $h(n)$ table.
References
| Ref | Source | Type |
|---|---|---|
| REF-01 | Erdős Problem #1016 (T. F. Bloom) | website |
| REF-02 | OEIS A105206 — related sequence linked from the Erdős #1016 page (minimal pancyclic graphs) | website |
Investigations · 0
No published investigations yet. This problem is unclaimed territory.