Erdős Problem #990

Full solution by OpenAI internal model (9 Apr 2026).

Problem
Erdős problem #990 (analysis); current status: disproved (Lean)
Credit
OpenAI internal model · AI on its own
Checked
Listed in Tao's wiki
Timeline
First AI contribution 2025-10-20 · wiki data as of 2026-06-30
Sources
Problem page (erdosproblems.com)Tao's AI-contributions wiki
SOLVEDPer Tao's AI-contributions wiki

The problem

Let f=a0+⋯+adxd∈C[x]f=a_0+\cdots+a_dx^d\in \mathbb{C}[x] be a polynomial. Is it true that, if ff has roots z1,…,zdz_1,\ldots,z_d with corresponding arguments θ1,…,θd∈[0,2π]\theta_1,\ldots,\theta_d\in [0,2\pi], then for all intervals I⊆[0,2π]I\subseteq [0,2\pi]

∣(#θi∈I)−∣I∣2πd∣≪(nlog⁡M)1/2,\left\lvert (\# \theta_i \in I) - \frac{\lvert I\rvert}{2\pi}d\right\rvert \ll \left(n\log M\right)^{1/2},

where nn is the number of non-zero coefficients of ff and

M=∣a0∣+⋯+∣ad∣(∣a0∣∣ad∣)1/2.M=\frac{\lvert a_0\rvert+\cdots +\lvert a_d\rvert}{(\lvert a_0\rvert\lvert a_d\rvert)^{1/2}}.

Statement quoted from erdosproblems.com/990, maintained by Thomas Bloom. Original source: [Er64b]. Current status there: disproved (Lean).

AI contributions

As recorded in Terence Tao's wiki AI contributions to Erdős problems (data as of 30 Jun 2026). 🟢 full, 🟡 partial, ⚪ unverified, 🔴 incorrect.

Date Kind AI systems Outcome
20 Oct, 2025 Literature search ChatGPT Deep research, Gemini Deep Research 🟡 Partial results found
9 Apr, 2026 AI on its own OpenAI internal model 🟢 Full solution

Caveats

The wiki's classifications are provisional and are not a benchmark. Many Erdős problems have never had a thorough literature review, so a result that looks new may later turn out to be known; some problems were also stated ambiguously or incorrectly. See the problem page for the full discussion.

∎