cd /news/artificial-intelligence/a-sticky-kakeya-set-over-f3-that-bea… · home › topics › artificial-intelligence › article
[ARTICLE · art-173123] src=dev.to ↗ pub= topic=artificial-intelligence verified=true sentiment=· neutral

A sticky Kakeya set over F3 that beats the classical construction, and where the question stands

MV2 of Munim, Inc. reported finding an explicit 30-point sticky Kakeya set in F_3^4, one point smaller than the classical 31-point construction, which lifts to sticky sets of size 31·2^(n-4) − 1 for all n ≥ 4 and shows the classical set is minimal only up to n = 3. The team also showed that no sticky Kakeya tree over F_3 beats the classical growth rate below height 5, with the minimum slice product π_k equal to 8^k for k = 1 through 4 (π_4 = 4096), and introduced a single-subtree 'cyclic triangle' formulation that reduces the open growth-rate question to a conjecture that κ_h = 2^(h+1). The results, produced with AI agents and not yet checked by a human mathematician, are posted with code and pre-registrations on GitHub.

key takeaway

An explicit 30-point sticky Kakeya set in F_3^4 beats the classical 31-point construction, and exhaustive computation shows the minimum slice product π_k equals 8^k for k ≤ 4, so any tree beating the classical growth rate must have height at least 5.

by read3 min views1 publishedNov 20, 2026

By MV2 of Munim, Inc. (mv2@munimtech.com). The work described here was carried out with AI agents working in MV2's name. No human mathematician has checked these results yet; corrections and pointers are very welcome.

A Kakeya set in F_3^n contains a full line in every direction. Over the three-element field, the Kakeya sets that are easiest to build are sticky: the line in direction (d, 1) starts at a point P(d), and coordinate j of P(d) depends only on the first j coordinates of d. The classical sticky construction, P(d) = (d_1^2, ..., d_n^2), has 2^(n+1) - 1 points, which gives a growth rate of 2^n.

We asked two questions: is the classical sticky set the smallest sticky Kakeya set, and does the best growth rate for sticky sets stay at 2?

The classical set is not the smallest. We found an explicit 30-point sticky Kakeya set in F_3^4, where the classical set has 31 points. A colour-lifting step then gives sticky sets of size 31 * 2^(n-4) - 1 for every n >= 4, so the classical set is minimal only up to n = 3. Write-up: A sticky Kakeya set over F3 that beats the classical construction.

The growth rate is still open, and we pinned down where to look. The rate question reduces to "trees": a sticky map of height k has three slices S_t = {P(d) + t d}, and the growth base is below 2 exactly when some tree has |S_0| |S_1| |S_2| < 8^k. Exhaustive computations show that the minimum of this product, pi_k, is exactly 8^k for k = 1, 2, 3, 4 (pi_4 = 4096). So any tree that beats the classical construction has height at least 5. Write-up: No sticky Kakeya tree over F3 beats the classical one below height 5.

A cleaner target (newest, note a195). A height-(k+1) tree normally depends on three different subtrees and on unions of slices taken from different subtrees. We showed that one family, the cyclic triangles, needs only one subtree A: its three slices all have size kappa(A) = min |A_i| + |A_{i+1} ∪ (A_{i+2} + w)|. So one tree of height h with kappa(A) < 2^(h+1) would show that sticky Kakeya sets over F_3 grow with base below 2. The extremal height-4 tree we found earlier is a cyclic triangle. We conjecture that kappa_h = 2^(h+1) for every h; this is now proved for h <= 3 and open from h = 4.

The best lower bound we have for sticky sets is about 1.905^n. Every Kakeya set in F_3^n, sticky or not, has at least 1.82^n points (via the Katz-Tao sum-difference inequality). We also showed that bounds of the "relaxation" kind, which replace unions of slices by maxima, can never prove that the classical rate 2 is optimal. Only exact structure, like the computations above, can settle it.

Code, notes, pre-registrations and raw outputs are in munimtechnologies/kakeya-small-fields (code Apache-2.0, text CC BY 4.0). Two pre-registered computations are queued: a proved lower bound for pi_5, and a search for a single tree that breaks the kappa conjecture (sticky_kappa.py, note a195).

If you know of earlier work on sticky Kakeya sets over finite fields, or see a mistake, please write to mv2@munimtech.com. MV2, Munim, Inc., mv2@munimtech.com. Made with AI agents working in MV2's name.

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @mv2 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/a-sticky-kakeya-set-…] indexed:0 read:3min 2026-11-20 · —