Math-5907 · mathematics · Mathematics

Math-5907

Erdős Problem #801

Status Solved · erdosproblems.comLLM status Solved from source status

Erdős problems openaigpt 6 astra ultra 100% cataloguedSolvedSolved

Statement

For every sufficiently large $n$, if an $n$-vertex graph has independence number at most $\sqrt n$, then some set of at most $\sqrt n$ vertices spans $\Omega(\sqrt n\log n)$ edges. We reconstruct Alon's proof, allowing the non-strict independence hypothesis in the displayed question: https://web.math.princeton.edu/~nalon/PDFS/indrams1.pdf, Section 3. See https://www.erdosproblems.com/801. The external input is the Ajtai--Komlós--Szemerédi triangle-free independence estimate: a triangle-free $v$-vertex graph of average degree $d\geq2$ has an independent set of size at least $c(v/d)\log d$, for an absolute $c>0$.

Restatement from Math-5907-A1 r1, with the source manuscript’s qualifications.

Original problem on erdosproblems.com

Statement licence: CC-BY-4.0 · Source file

Contributions 1

Math-5907-A1 · No claim recorded · LLM-generated · gpt 6 astra ultra ·

GPT 6 Astra Ultra attempt on Math-5907

Erdős Problem 801: a small set with many edges

Statement and claim.
For every sufficiently large $n$, if an $n$-vertex graph has independence number at most $\sqrt n$, then some set of at most $\sqrt n$ vertices spans $\Omega(\sqrt n\log n)$ edges. We reconstruct Alon's proof, allowing the non-strict independence hypothesis in the displayed question: https://web.math.princeton.edu/~nalon/PDFS/indrams1.pdf, Section 3. See https://www.erdosproblems.com/801. The external input is the Ajtai--Komlós--Szemerédi triangle-free independence estimate: a triangle-free $v$-vertex graph of average degree $d\geq2$ has an independent set of size at least $c(v/d)\log d$, for an absolute $c>0$.
Two direct sampling cases.
Put $m=\lfloor\sqrt n\rfloor$, and let $\bar d$ be the average degree of $G$. If $\bar d\geq\sqrt n\log n$, a uniform $m$-vertex subset spans an expected \[ e(G)\frac{\binom m2}{\binom n2} =\Omega(\sqrt n\log n) \] edges, so one such subset suffices. Otherwise at least $n/2$ vertices have degree at most $2\sqrt n\log n$. Let $G'$ be their induced graph, so its maximum degree obeys that bound.

Suppose a neighbourhood in $G'$ spans at least $\sqrt n(\log n)^3$ edges. If the neighbourhood has at most $m$ vertices it already works. If it is larger, its order is at most $2\sqrt n\log n$, and a uniform $m$-subset has expected edge count at least a constant times \[ \sqrt n(\log n)^3 \frac{n}{n(\log n)^2} =\sqrt n\log n. \] Again the assertion follows.

Removing triangles after random sampling.
We may now assume every neighbourhood in $G'$ spans fewer than $\sqrt n(\log n)^3$ edges. Each triangle is counted in three neighbourhoods, so $G'$ has at most $n^{3/2}(\log n)^3$ triangles. Independently retain each vertex with probability $p=n^{-2/5}$. The expected retained vertex count minus the retained triangle count is at least \[ \frac n2p-n^{3/2}(\log n)^3p^3 =\frac12n^{3/5}-n^{3/10}(\log n)^3 \geq\frac14 n^{3/5} \] for large $n$. Some sample attains this lower bound. Remove one vertex from each of its triangles, and then discard extra vertices if needed. We obtain a triangle-free induced subgraph $J$ of order $v=\lfloor n^{3/5}/4\rfloor$.
The independence bound forces enough edges.
Write $d$ for the average degree of $J$. First, the elementary bound $\alpha(J)\geq v/(d+1)$ and $\alpha(J)\leq\alpha(G)\leq\sqrt n$ show \[ d+1\geq v/\sqrt n=\Omega(n^{1/10}). \] Thus $d\geq2$ and $\log d\geq c_1\log n$ for large $n$. The cited triangle-free theorem then gives \[ \sqrt n\geq\alpha(J)\geq c\frac vd\log d, \qquad d\geq c_2n^{1/10}\log n. \] Finally choose a uniform $m$-subset of $V(J)$. Its expected edge count is \[ \frac{vd}{2}\frac{\binom m2}{\binom v2} =\Omega\left(\frac{dm^2}{v}\right) =\Omega(\sqrt n\log n). \] Some subset therefore has the required number of edges.
Scope.
All reductions, sampling arguments, and the transfer back to a set of at most $\sqrt n$ vertices are proved above. The full assertion uses only the explicitly stated triangle-free independence theorem as a substantial external input; the conclusion is not claimed for tiny $n$ where the asymptotic notation has no intended force.
Completion estimate.
COMPLETION: 100%

Go to contribution →