Sublinear growth of 1 1 -cocycles and uniform convexity
Andreas Thom · 2026
Read Sublinear growth of 1 1 -cocycles and uniform convexity by Andreas Thom (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-2603-22049.txt.
Introduction
Thom proves a growth bound for 1-cocycles tied to group representations on Banach spaces. The group G is finitely generated, and its representation is uniformly bounded and of c0 type. Thom shows any 1-cocycle for that action grows slower than the word length of group elements. In other words, its growth is sublinear. As a corollary, he obtains the same statement for Hilbert spaces. That version covers strongly mixing unitary representations.
Catalog record
| CREATOR | Andreas Thom |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 17187 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.22049
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Sublinear growth of 1 1 -cocycles and uniform convexity Andreas Thom Abstract . Let G be a finitely generated group, let π : G → GL ( ℰ ) \pi\colon G\to{\rm GL}(\mathcal{E}) be a uniformly bounded c 0 c_{0} -representation on a superreflexive Banach space ℰ \mathcal{E} , and let b : G → ℰ b\colon G\to\mathcal{E} be a 1 1 -cocycle for π \pi . Then b b has sublinear growth with respect to the word length. As a corollary we obtain the corresponding Hilbert space statement for strongly mixing unitary representations. 2010 Mathematics Subject Classification 20F65, 22D12, 46B20 1. Introduction Let G be a finitely generated group with a fixed finite symmetric generating set, and let |g| denote the associated word length. Let \mathcal{E} be a Banach space. Let \pi\colon G\to{\rm GL}(\mathcal{E}) be a representation by linear isomorphisms. The representation \pi is called uniformly bounded if there exists C>0 such that \left\lVert\pi(g)\right\rVert\leq C for every g\in G . We say that \pi is a c_{0} -representation if for every \xi\in\mathcal{E} and every \varphi\in\mathcal{E}^{\ast} one has \varphi(\pi(g)\xi)\to 0\qquad\text{as }g\to\infty. Equivalently, every orbit vector tends weakly to 0 at infinity; this terminology is standard in [ 1 , 7 ] . An affine isometric action with linear part \pi is encoded by a 1 -cocycle b(gh)=b(g)+\pi(g)b(h)\qquad(g,h\in G). The modulus of convexity is defined for a Banach space \mathcal{E} by \delta_{\mathcal{E}}(\varepsilon):=\inf\Bigl\{1-\Bigl\lVert\frac{x+y}{2}\Bigr\rVert:\left\lVert x\right\rVert=\left\lVert y\right\rVert=1,\ \left\lVert x-y\right\rVert\geq\varepsilon\Bigr\}\qquad(0<\varepsilon\leq 2). A Banach space is called uniformly convex if \delta_{\mathcal{E}}(\varepsilon)>0 for every \varepsilon>0 ; see, for example, Benyamini–Lindenstrauss [ 3 ] . A Banach space \mathcal{E} is called superreflexive if it admits an equivalent uniformly convex norm; see, for example, [ 1 , 3 ] for background. The purpose of this note is to show that for uniformly bounded c_{0} -representations on superreflexive Banach spaces, every 1 -cocycle has sublinear growth along the word length, see Theorem 1 . In the Hilbert space setting, the corresponding class of unitary (or more generally, uniformly bounded) representations are called strongly mixing . Growth and rigidity properties of 1 -cocycles into unitary representations are intertwined with affine isometric actions, reduced cohomology, and harmonic cocycles; classical references include Guichardet [ 10 ] , Shalom [ 12 ] , and the monograph of Bekka–de la Harpe–Valette [ 2 ] . More recently, Cornulier–Tessera–Valette [ 6 ] analyzed the relation between sublinear growth and almost coboundaries, while Chifan–Sinclair [ 4 ] and Ozawa [ 11 ] obtained vanishing and weak sublinearity results for cocycles into weakly mixing representations of amenable groups; see also Gournay [ 9 ] for related mixing/cohomological phenomena. To the best of our knowledge, even the Hilbert space statement for strongly mixing representations (stated here as Corollary 2 ) does not seem to be explicitly recorded in the literature, despite the fact that quantitative and qualitative control of cocycles remains an active topic of current research. Our main result is the following theorem. It will be proved in the next section. Theorem 1 . Let G be a finitely generated group, \mathcal{E} be a superreflexive Banach space, \pi\colon G\to{\rm GL}(\mathcal{E}) be a uniformly bounded c_{0} -representation, and b\colon G\to\mathcal{E} be a 1 -cocycle. Then \limsup_{|g|\to\infty}\frac{\left\lVert b(g)\right\rVert}{|g|}=0. Equivalently, every such 1 -cocycle has sublinear growth. Applications to the setting of Hilbert spaces are discussed in the third section, where we also show that the weaker notion of weak mixing does not force sublinear growth of 1 -cocycles, by constructing a weakly mixing representation of the free group F_{2} together with a cocycle of linear growth along a sequence, see Proposition 3 . A final section records a consequence of Theorem 1 for vanishing of reduced cohomology for finitely generated amenable groups admitting controlled Følner sequences, see Corollary 5 . This extends results of Tessera [ 13 ] , who proved the vanishing of reduced cohomology for the L^{p} -representation associated with a mixing measure preserving actions G\curvearrowright(X,\mu) , where \mu is an infinite measure and 1<p<\infty . 2. Proof of the main theorem Proof of Theorem 1 . At first, we use [ 1 , Proposition 2.3] to replace the given norm on \mathcal{E} by an equivalent uniformly convex norm that is preserved by \pi . Hence, we may assume that \mathcal{E} is uniformly convex and \pi is an isometric representation. Passing to the underlying real Banach space, we may assume that \mathcal{E} is real. Assume for contradiction that \alpha:=\limsup_{|g|\to\infty}\frac{\left\lVert b(g)\right\rVert}{|g|}>0. For g\neq e set q(g):=\frac{\left\lVert b(g)\right\rVert}{|g|}. Step 1. Fix \Delta:=\delta_{\mathcal{E}}(1/2)>0 and choose \eta>0 so small that 0<\eta<\frac{\alpha\Delta}{8}. By the definition of \limsup , there exists R\in\mathbb{N} such that |x|\geq R implies q(x)\leq\alpha+\eta. Equivalently, |x|\geq R\implies\left\lVert b(x)\right\rVert\leq(\alpha+\eta)|x|. Step 2. For n\geq 1 , let \alpha_{n}:=\max\{\left\lVert b(g)\right\rVert:|g|=n\}. Since spheres of radius n are finite, the maximum exists. The sequence (\alpha_{n})_{n\geq 1} is subadditive: if |g|=n+m and g=xy is obtained by cutting a geodesic word for g after n letters, then |x|=n , |y|=m , and \left\lVert b(g)\right\rVert\leq\left\lVert b(x)\right\rVert+\left\lVert b(y)\right\rVert\leq\alpha_{n}+\alpha_{m}. Hence Fekete’s lemma gives the existence of \lim_{n\to\infty}\frac{\alpha_{n}}{n}=\inf_{n\geq 1}\frac{\alpha_{n}}{n}. Since \alpha_{n}/n=\max_{|g|=n}q(g) , this limit equals \alpha . In particular, for every n there exists g_{n}\in G such that |g_{n}|=n\qquad\text{and}\qquad\left\lVert b(g_{n})\right\rVert=\alpha_{n}\geq\alpha n. Fix a geodesic word g_{n}=s_{1}^{(n)}\cdots s_{n}^{(n)} for each n . Step 3. For n large enough, partition the geodesic word for g_{n} into consecutive geodesic subwords g_{n}=h_{1}^{(n)}h_{2}^{(n)}\cdots h_{m_{n}}^{(n)} such that for every i , R\leq\ell_{i}^{(n)}:=|h_{i}^{(n)}|\leq 2R, and \sum_{i=1}^{m_{n}}\ell_{i}^{(n)}=n. Fix n and suppress the superscript (n) from the notation. For 1\leq i\leq m_{n} define the prefix t_{i}:=h_{1}\cdots h_{i-1}, with t_{1}=e , and set v_{i}:=\pi(t_{i})b(h_{i}). Repeatedly applying the cocycle identity gives b(g_{n})=\sum_{i=1}^{m_{n}}v_{i}. Since \pi(t_{i}) is an isometry, \left\lVert v_{i}\right\rVert=\left\lVert b(h_{i})\right\rVert. As |h_{i}|\geq R , estimate (1) yields \left\lVert v_{i}\right\rVert\leq(\alpha+\eta)\ell_{i}. Because of (2), b(g_{n})\neq 0 . Choose a norming functional j_{n}\in\mathcal{E}^{\ast} with \left\lVert j_{n}\right\rVert=1 and j_{n}(b(g_{n}))=\left\lVert b(g_{n})\right\rVert. Step 4. For each block define d_{i}:=(\alpha+\eta)\ell_{i}-j_{n}(v_{i}). Since j_{n}(v_{i})\leq\left\lVert v_{i}\right\rVert\leq(\alpha+\eta)\ell_{i} by (5), we have d_{i}\geq 0 . Summing and using (2), (3), and (4), we obtain \displaystyle\sum_{i=1}^{m_{n}}d_{i} \displaystyle=(\alpha+\eta)\sum_{i=1}^{m_{n}}\ell_{i}-\sum_{i=1}^{m_{n}}j_{n}(v_{i}) \displaystyle=(\alpha+\eta)n-j_{n}\Bigl(\sum_{i=1}^{m_{n}}v_{i}\Bigr) \displaystyle=(\alpha+\eta)n-j_{n}(b(g_{n})) \displaystyle=(\alpha+\eta)n-\left\lVert b(g_{n})\right\rVert \displaystyle\leq\eta n. Thus \sum_{i=1}^{m_{n}}d_{i}\leq\eta n. Call an index i good if d_{i}\leq 4\eta\,\ell_{i}. Let G_{n} be the set of good indices. Then \sum_{i\notin G_{n}}4\eta\,\ell_{i}<\sum_{i\notin G_{n}}d_{i}\leq\sum_{i=1}^{m_{n}}d_{i}\leq\eta n, so \sum_{i\notin G_{n}}\ell_{i}\leq\frac{1}{4}n. Therefore \sum_{i\in G_{n}}\ell_{i}\geq\frac{3}{4}n. Step 5. Fix i\in G_{n} . Since d_{i}\leq 4\eta\ell_{i} , we have j_{n}(v_{i})\geq(\alpha-3\eta)\ell_{i}. Combining (5) and (7), we get \left\lVert v_{i}\right\rVert\geq j_{n}(v_{i})\geq(\alpha-3\eta)\ell_{i}>0. Hence the normalized vector w_{i}:=\frac{v_{i}}{\left\lVert v_{i}\right\rVert}\in S_{\mathcal{E}} is well defined, and j_{n}(w_{i})=\frac{j_{n}(v_{i})}{\left\lVert v_{i}\right\rVert}\geq\frac{(\alpha-3\eta)\ell_{i}}{(\alpha+\eta)\ell_{i}}=1-\frac{4\eta}{\alpha+\eta}>1-\Delta. Now fix i,j\in G_{n} . Since \left\lVert w_{i}\right\rVert=\left\lVert w_{j}\right\rVert=1 and \left\lVert j_{n}\right\rVert=1 , we get \Bigl\lVert\frac{w_{i}+w_{j}}{2}\Bigr\rVert\geq j_{n}\Bigl(\frac{w_{i}+w_{j}}{2}\Bigr)=\frac{j_{n}(w_{i})+j_{n}(w_{j})}{2}>1-\Delta. By the definition of \Delta=\delta_{\mathcal{E}}(1/2) , this forces \left\lVert w_{i}-w_{j}\right\rVert<\frac{1}{2} , hence \left\lVert w_{i}-w_{j}\right\rVert<\frac{1}{2}. Step 6. Let \mathcal{A}_{R}:=\{h\in G:R\leq|h|\leq 2R\}. This is a finite set; write M_{R}:=|\mathcal{A}_{R}| . Since each good block has length at most 2R , estimate (6) implies 2R\,|G_{n}|\geq\sum_{i\in G_{n}}\ell_{i}\geq\frac{3}{4}n, hence |G_{n}|\geq\frac{3}{8R}n. Among the good blocks, some element of \mathcal{A}_{R} occurs at least |G_{n}|/M_{R} times. Therefore there exist h^{(n)}\in\mathcal{A}_{R} and a subset I_{n}\subset G_{n} such that h_{i}=h^{(n)}\quad(i\in I_{n}), and |I_{n}|\geq\frac{3}{8RM_{R}}n. Passing to a subsequence, we may assume that h^{(n)}=h is independent of n . Set X_{n}:=\{t_{i}:i\in I_{n}\}. Since the prefixes t_{i} are distinct, we have |X_{n}|=|I_{n}|\geq\frac{3}{8RM_{R}}n, so |X_{n}|\to\infty . For t\in X_{n} , the corresponding vector is \pi(t)b(h)=v_{i} for the unique i\in I_{n} with t=t_{i} . By (8), \left\lVert b(h)\right\rVert=\left\lVert v_{i}\right\rVert\geq(\alpha-3\eta)|h|>0. Moreover, if t,t^{\prime}\in X_{n} correspond to