Flexibility of measurable and topological nilfactors in dynamical systems
Seljon Akhmedli · 2026
Read Flexibility of measurable and topological nilfactors in dynamical systems by Seljon Akhmedli (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-20855.txt.
Introduction
Sebastiano Monti and four co-authors test how well AI models plan far ahead. Their 2026 paper presents SokoBench, a benchmark for long-horizon planning and reasoning. The name nods to Sokoban, the classic box-pushing puzzle. Large language models have been tested hard on complex reasoning. Their long-horizon planning, though, has not been studied as closely. The team set out to assess it in a systematic way.
The benchmark targets state-of-the-art large reasoning models. Long-horizon tasks need many steps that stay consistent to the end. One early error can ruin the whole plan. SokoBench measures where models break down as plans lengthen. That gives developers a clear view of a weak spot. Better planning tests are a step toward agents that can carry out long jobs alone.
Catalog record
| CREATOR | Seljon Akhmedli |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-20855
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Flexibility of measurable and topological nilfactors in dynamical systems Thanks: The author was partially supported by NSF grant DMS-2136217 Seljon Akhmedli Abstract . We construct examples of minimal and uniquely ergodic systems realizing all possible behaviors in the interplay of measurable and topological nilfactors. To build such examples, we adapt an idea that stems from Furstenberg’s construction of a minimal but not uniquely ergodic system on 𝕋 2 \mathbb{T}^{2} . 1. Introduction Nilsystems and nilfactors of systems arise in convergence and recurrence problems in ergodic theory and topological dynamics. Prime examples of these can be found in [ 8 ] and [ 12 ] . Numerous measure-theoretic and topological properties, such as transitivity, minimality, ergodicity, and unique ergodicity, are equivalent in nilsystems, combining results of [ 1 , 20 , 21 , 23 ] . These equivalences are key in many recent developments in dynamics. To describe the behaviors that may arise, we assume throughout that our dynamical systems are endowed with both measurable and topological structure. Every topological factor naturally has the structure of a measurable factor, but the converse does not hold in general. The measurable k -step nilfactor of a system has been shown in works of Host and Kra in [ 12 ] to control the dynamical behavior of certain multiple ergodic averages. The topological analog of these factors was first introduced by Host, Kra and Maass in [ 14 ] (see Section 2 for definitions). A more recent interplay of the measurable and topological nilfactors has appeared in [ 17 ] where Kra, Moreira, Richter and Robertson, solved a conjecture of Erdős on sumsets. There, it was shown how these nilfactors control certain infinite patterns in sets with positive density. The well-known Jewett-Krieger theorem says that every ergodic system has a strictly ergodic (minimal and uniquely ergodic) topological model [ 15 , 18 ] . By a topological model, or just model, we mean a topological dynamical system equipped with an invariant probability measure that is measurably isomorphic to the original system. In [ 11 ] , it is shown that every ergodic system has a strictly ergodic model where the measurable and topological k -step nilfactors, Z_{k} and X_{k} , are isomorphic. An older result of Lehrer [ 19 ] obtaining topologically mixing models of ergodic systems is most relevant for this work. An important consequence of his theorem is the existence of systems where Z_{1} and X_{1} are not isomorphic; it then follows Z_{i} and X_{i} are not isomorphic as well for all i>1 . Concrete examples of such systems can be found in, for instance, [ 3 , 9 , 19 , 22 ] . We can illustrate these phenomena in the diagram below X Z_{k} X_{k} Z_{k-1} X_{k-1} Z_{1} X_{1} where Z_{i}\dasharrow X_{i} indicates either a measurable isomorphism for all i or a proper factor map for each i . We construct examples realizing all possible behaviors in between these two phenomena. Understanding when the measurable and topological nilfactors are different constitutes partly to the goal of this paper. We note in a weak mixing system all nilfactors are trivial (see, for example, Proposition 20 of Chapter 8 in [ 13 ] ). We also note that nilsystems are classic examples of systems with Z_{k} and X_{k} isomorphic. Theorem 1.1 . For all 0\leq j\leq\ell\leq k,\;k\geq 1 , there exists a strictly ergodic system which admits the following: 1. Z_{k} is a nontrivial extension of Z_{k-1} 2. Z_{i} is topologically isomorphic to X_{i} for all 0\leq i\leq j 3. Z_{i} is not measurably isomorphic to X_{i} for all j+1\leq i\leq k 4. X_{i} is topologically isomorphic to X_{i+1} for all i\geq\ell . Furthermore, for every minimal, ergodic, and non-weak mixing (X,\mu,T) there exist j,\ell , and k , such that X satisfies properties (1)-(4). To prove Theorem 1.1 , we draw inspiration from an example of a minimal but not uniquely ergodic system constructed by Furstenberg in [ 5 ] . There, it is shown that for \alpha\notin\mathbb{Q} , the system T(x,y)=(x+\alpha,y+f(x)),\;\;T\colon\mathbb{T}^{2}\rightarrow\mathbb{T}^{2} is minimal and not uniquely ergodic if and only if (1.1) nf\equiv F(x+\alpha)-F(x)\mod 1 has a measurable solution for some n\in\mathbb{Z}\setminus\{0\} but no continuous solution unless n=0 . In [ 5 ] , an f which has a measurable solution for n=1 but no continuous solution except when n=0 is constructed. We refer to f as a measurable but not continuous coboundary over the system (\mathbb{T},m_{\mathbb{T}},T_{\alpha}) . In Section 4 , we use these coboundaries to construct systems where the measurable and topological k -step nilfactors are not measurably isomorphic after some point. In Proposition 4.9 , we exhibit another flexibility of the nilfactors where for each k\in\mathbb{N} , we build a strictly ergodic system which has Z_{k} as a nontrivial extension of Z_{k-1} yet has trivial topological nilfactors. Such examples are known to exist by Lehrer [ 19 ] , however, we provide a construction that is more explicit using a certain minimal Cantor system built by Durand, Frank, and Maass, in [ 3 ] along with techniques we develop in this article from Proposition 4.4 . 2. Preliminaries 2.1. Definitions and notations Throughout \mathbb{N} denotes the set \{1,2,3,\dots\} . By a dynamical system we mean any triple of the form (X,\mu,T) where X is a compact metric space endowed with the Borel \sigma -algebra and T is a homeomorphism of X preserving a Borel probability measure \mu . Namely, for any Borel set B\subseteq X we have \mu(T^{-1}(B))=\mu(B). Given two compact metric spaces X , Y , and a measurable map f\colon X\rightarrow Y , the push forward of \mu under f is the invariant measure on Y defined by (f_{*}\mu)(B)=\mu(f^{-1}(B)) for all B\subseteq Y . Since we have assumed throughout that the setting of our dynamical systems has an underlying topological structure, this allows us to discuss both measurable and topological factors. For a system (X,\mu,T) , we say (Y,\nu,S) is a measurable factor of (X,\mu,T) if there exists a measurable map \pi\colon X\rightarrow Y , the measurable factor map , such that (S\circ\pi)(x)=(\pi\circ T)(x) for \mu -almost every x\in X and \pi_{*}\mu=\nu. Furthermore, if \pi is a bijection with \pi^{-1} measurable, then it is a measurable isomorphism . We say (Y,\nu,S) is a topological factor of (X,\mu,T) if there exists a continuous and surjective map \rho\colon X\rightarrow Y , the topological factor map , such that (S\circ\rho)(x)=(\rho\circ T)(x) for all x\in X and \rho_{*}\mu=\nu. If \rho\colon X\rightarrow Y is a homeomorphism, then it is a topological isomorphism . We use the notation \overset{m}{\cong} and \overset{t}{\cong} to denote measurable and topological isomorphisms, respectively. Given a k -step nilpotent Lie group G and \Gamma\leq G a discrete cocompact subgroup, X=G/\Gamma is a k -step nilmanifold . Now fix g\in G . Then T\colon X\rightarrow X given by x\mapsto g\cdot x is a nilrotation. The Haar measure on X is the unique measure invariant under all nilrotations and we denote it by m_{X} . A system (X,m_{X},T) where X is a k -step nilmanifold and T is a nilrotation is a k -step nilsystem . A special class of nilsystems we consider are affine nilsystems . Let A be a k\times k unipotent integer matrix, b\in\mathbb{T}^{k} , and