OpenAI's released preprint 'Finite-tensor savings and exact Fourier circuits' (openai/math, dated 2026-09-25) claims a discrete Fourier transform faster than the classical n log n bound; expert acceptance of its computational model would upend a foundational algorithmic barrier, while a model-assumption flaw closes it as an error.

state: watchingheat: lowuncertainty: highconvergesscott: mediumai-for-math algorithmic-discovery openaiOpenAI

What is this?

The case concerns an OpenAI preprint ('Finite-tensor savings and exact Fourier circuits', dated 2026-09-25, HN-relayed as 'Discrete Fourier Transform faster than n log n') claiming a DFT algorithm that beats the classical O(n log n) FFT. The supplied web results cover only the classical background: Cooley–Tukey's 1965 FFT computing the DFT in O(N log N), plus known conditional speedups — the sparse-FFT literature (Gilbert/Indyk/Iwen/Schmidt) achieves sub-N-log-N scaling under sparsity assumptions, and analog/fermionic-FFT variants restructure the computation in other hardware models. None of the snippets mention the OpenAI preprint itself, so its existence, content, and authorship rest entirely on the HN relay in the case evidence, and no expert reaction is in the record. Note also that the snippets establish O(N log N) as the best-known *general-case* algorithm, not a proven lower bound — and sub-n-log-n results already exist under restricted signal models — so whether this 'upends a foundational barrier' hinges on the paper's exact computational-model assumptions, which the supplied material does not show.

Why it matters to Scott

Converges on his evidence-governance canon: a frontier-lab preprint claiming a classical algorithmic barrier is exactly the falsifiable-lab-claim shape his Evidence Class Ladder and Challenger, Never Arbiter frameworks adjudicate, and the HN headline ('faster than n log n') versus the undisclosed computational-model assumptions is a live weakest-evidence-class test whose expert verdict pays dated receipts either way — validation upgrades the AI-discovered-algorithms thesis he tracks alongside Anthropic's 3SUM/APSP and AlphaEvolve episodes, while a model-assumption flaw receipts the canon-vs-slopcannon headline-vs-fine-print gap. Medium rather than high because it touches no technology or project he actively builds, and it is a new episode in a well-populated OpenAI-preprint lineage (concluded distinct from the corpus-volume batch-release story) rather than a new argument his canon must absorb.

ip:concept.evidence-class-ladderip:framework.challenger-never-arbiterip:framework.falsifiability-spineip:concept.canon-vs-slopcannondev:concept.claim-bounded-adversarial-verificationradar:concept.algorithm-discoveryradar:concept.claim-verificationradar:openai-connes-rigidity-disproof-reviewradar:openai-millennium-maths-claimradar:agmai-openai-math-release-adviceradar:anthropic-3sum-apsp-refutationradar:alphaevolve-matrix-exponent-improvement

queries asked of Scott's wikis

AI-discovered algorithms novelty track record
frontier lab preprint credibility without peer review
expert-verdict resolution path for AI claims
FFT signal processing or algorithm dev history
falsifiable capability claims from labs
AI-for-math tooling and benchmarks

Measured heat

now 0 pts/hpeak 3 pts/hcomments 0/hpeers p16momentum: steady2 platformsage 410h

points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion

How the heat travelled

09-24 14:00	⭐ origin echo-reconstructed	Preprint titled 'Finite-tensor savings and exact Fourier circuits', relayed on HN as 'Discrete Fourier Transform faster than n log n'. OpenAI on github (echo) · attributed from hn.story.49986313	—
10-07 00:37	first on hacker news · published · +298.6h	Discrete Fourier Transform faster than n log n [pdf] E-Reverance	—
10-07 00:37	amplified on hacker news	hn.story.49986313 E-Reverance	peak 2 · 0 comments · 18% of case engagement
10-08 23:59	amplified on hacker news	hn.story.50014129 shea256	peak 2 · 0 comments · 18% of case engagement
10-09 00:43	amplified on hacker news 👑	hn.story.50014526 shea256	peak 6 · 0 comments · 55% of case engagement
10-09 10:29	amplified on hacker news	hn.story.50018530 mcmoor	peak 1 · 0 comments · 10% of case engagement
10-07 03:22	our radar first saw it · +301.4h	discovery anchor: hn.story.49986313	—

pace: p23 vs 1032 stories at the 336h mark (now 410h old) — ahead of aafp-commons-signed-agent-notebook (2.0x), behind agentgate-signed-agent-receipts (0.7x)

Evidence (5) — ⭐ canonical anchor

source	object	author	score	comments
🟧 hn	Discrete Fourier Transform faster than n log n [pdf] Retrieved article excerpt Open article · Retrieved 2026-10-07T03:33:25.479769+00:00 ## Repository navigation ## FilesExpand file tree main / # main.pdf Copy path More file actions More file actions ## Latest commit ## History [History](https://github.com/openai/math/commits/main/preprints/Finite-tensor-savings-and-exact-Fourier-circuits-September-25-2026/main.pdf) History 618 KB main / # main.pdf Copy path Top ## File metadata and controls 618 KB Download raw file Edit and raw actions LoadingViewer requires iframe.	E-Reverance	2	0
🟧 echo.github ⭐	Preprint titled 'Finite-tensor savings and exact Fourier circuits', relayed on HN as 'Discrete Fourier Transform faster than n log n'.	OpenAI	—	—
🟧 hn	Exact discrete Fourier transform below n log n: exponent saving 1e-13 to 7.3e-5	shea256	2	0
🟧 hn	Improving OpenAI's bound on the exact discrete Fourier transform below n log n	shea256	6	0
🟧 hn	Community pushed OpenAI FFT proof to over 10^-4	mcmoor	1	0

Interpretation history

2026-10-09T17:49:48Z
Community replication (mcmoor) has quantitatively advanced the claimed exponent saving to >10^-4, joining shea256's independent analysis — two researchers now publicly stress-testing the bound on HN. This is the first measurable replication progress, but it remains a single platform, zero expert review, and the computational-model assumptions that determine whether this challenges the general-case barrier are still undisclosed.
2026-10-09T13:51:10Z
evidence attached: hn.story.50018530 — Community replication effort pushing OpenAI's claimed faster-than-n-log-n FFT proof further; direct independent corroboration of the watching case.
2026-10-09T06:54:57Z
Independent researcher (shea256) now publicly analyzing and attempting to improve OpenAI's claimed bound — the claim has moved from bare preprint to live external scrutiny, though still single-source and pre-expert-verdict.
2026-10-09T04:52:45Z
evidence attached: hn.story.50014526 — Tweet discusses improving OpenAI's claimed Fourier transform bound, directly relevant to the open hypothesis.
2026-10-09T04:52:45Z
evidence attached: hn.story.50014129 — Independent GitHub implementation attempting to verify OpenAI's claimed sub-n-log-n DFT; direct corroboration attempt for the open case
2026-10-07T03:56:53Z
grounded: converges/medium — Converges on his evidence-governance canon: a frontier-lab preprint claiming a classical algorithmic barrier is exactly the falsifiable-lab-claim shape his Evid
2026-10-07T03:47:08Z
case created — A frontier lab claiming a sub-n-log-n DFT is a falsifiable challenge to a classical lower bound, a genuinely different claim from the corpus-volume story with its own expert-verdict resolution path.

Decision trace

10-10 04:50attention_routeThe editor compared this story and chose to keep watching.
10-10 04:49attention_candidatematerial_reprice
10-10 04:49repriceCommunity replication (mcmoor) has quantitatively advanced the claimed exponent saving to >10^-4, joining shea256's independent analysis — two researchers now publicly stress-testing the bound
10-10 01:26attention_routeThe editor compared this story and chose to keep watching.
10-10 00:51attention_candidateattach
10-10 00:51attachCommunity replication effort pushing OpenAI's claimed faster-than-n-log-n FFT proof further; direct independent corroboration of the watching case.
10-10 00:48propose_attachCommunity replication effort pushing OpenAI's claimed faster-than-n-log-n FFT proof further; direct independent corroboration of the watching case.
10-09 17:58attention_routeThe editor compared this story and chose to keep watching.
10-09 17:54attention_candidatematerial_reprice
10-09 17:54repriceIndependent researcher (shea256) now publicly analyzing and attempting to improve OpenAI's claimed bound — the claim has moved from bare preprint to live external scrutiny, though still single-so
10-09 15:58attention_routeThe editor compared this story and chose to keep watching.
10-09 15:52attention_candidateattach
10-09 15:52attachTweet discusses improving OpenAI's claimed Fourier transform bound, directly relevant to the open hypothesis.
10-09 15:52attachIndependent GitHub implementation attempting to verify OpenAI's claimed sub-n-log-n DFT; direct corroboration attempt for the open case
10-09 15:52propose_attachTweet discusses improving OpenAI's claimed Fourier transform bound, directly relevant to the open hypothesis.
10-09 15:50propose_attachIndependent GitHub implementation attempting to verify OpenAI's claimed sub-n-log-n DFT; direct corroboration attempt for the open case
10-08 00:59attention_routeThe editor compared this story and chose to keep watching.
10-07 14:56groundConverges on his evidence-governance canon: a frontier-lab preprint claiming a classical algorithmic barrier is exactly the falsifiable-lab-claim shape his Evidence Class Ladder and Challenger, Never
10-07 14:47createA frontier lab claiming a sub-n-log-n DFT is a falsifiable challenge to a classical lower bound, a genuinely different claim from the corpus-volume story with its own expert-verdict resolution path.