Oliveira’s Kac-walk mixing conjecture

Let μn,T\mu_{n,T} be the law on SO(n)\mathrm{SO}(n) of Kac's walk after TT steps, and let Haar⁡n\operatorname{Haar}_n denote Haar measure on SO(n)\mathrm{SO}(n). The conjecture asks whether, for a suitable short-trajectory regime T=T(n)T=T(n), every family of low-complexity tests fn:SO(n)→[−1,1]f_n:\mathrm{SO}(n)\to[-1,1] satisfies ∣EX∼μn,T[fn(X)]−EH∼Haar⁡n[fn(H)]∣=o(1)\left|\mathbb{E}_{X\sim\mu_{n,T}}[f_n(X)]-\mathbb{E}_{H\sim\operatorname{Haar}_n}[f_n(H)]\right|=o(1) as n→∞n\to\infty.

References

Primary source

arXiv

Additional references

Progress summary

Refreshed
Claimed solved

A 2026 preprint claims to settle the column-mixing conjecture, while the stronger all-algorithms question remains open.

Oliveira’s conjecture concerns how quickly the first kk columns of Kac’s walk approach their random limit; the relevant work of Oliveira appeared in 2009. The associated Vaikuntanathan–Zamir conjecture is broader, asking for computational indistinguishability from Haar randomness.

August 2026 claimed proof

A new preprint claims that the kk-column walk mixes in Wasserstein distance after O(nlog⁡n⋅max⁡(k,log⁡n))O(n\log n\cdot\max(k,\log n)) steps, uniformly in nn and kk. It further claims that degree-kk real polynomials cannot distinguish Haar-random matrices from walks of approximately O~(nk2)\widetilde{O}(nk^2) steps, and derives a fast Johnson–Lindenstrauss transform with an extra logarithmic factor. This resolves Oliveira’s conjecture and the low-degree version of the Vaikuntanathan–Zamir problem if the proof is correct, but no independent verification or reported error assessment was found.

Current status (as of August 2026): Oliveira’s column-mixing conjecture is claimed proved by a new preprint, but remains unverified; the broader Vaikuntanathan–Zamir conjecture remains open.

Sources

Solutions 0

No solutions have been posted yet.