Probability measure annihilating all finite dimensional subspaces

Nizar El Idrissi, Hicham Zoubeir · 2025

Read Probability measure annihilating all finite dimensional subspaces by Nizar El Idrissi, Hicham Zoubeir (2025) 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-2512-21664.txt.

Introduction

Nizar El Idrissi and Hicham Zoubeir build probability measures on infinite-dimensional Banach spaces with a striking property: every finite-dimensional subspace receives measure zero. Such measures matter where there is no analogue of ordinary volume. Gaussian measures and infinite-product builds were already known to have it.

The authors add a new construction based on prime numbers. Their method adds to the Gaussian and product approaches and is fairly elementary. They argue such measures are in fact generic among measures not supported on any finite-dimensional subspace. Along the way, they prove that any infinite-dimensional Banach space contains an uncountable family of independent vectors, as many as the continuum itself. The note widens the toolbox for building such measures, a setting where intuition from finite dimensions often fails.

Catalog record

CREATORNizar El Idrissi, Hicham Zoubeir
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE12 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2512-21664

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Probability measure annihilating all finite dimensional subspaces Nizar El Idrissi, Hicham Zoubeir Abstract We construct a probability measure annihilating all finite dimensional subspaces on an arbitrary infinite dimensional Banach space. 0 0 footnotetext: 2020 Mathematics Subject Classification. 46B09; 28A05; 28A09; 03E10; 15A03; 11N05; 13A05 0 0 footnotetext: Key words and phrases. Banach space; probability measure; measurability; cardinality of the continuum; linear independence; prime number theorem; UFD (unique factorization domain) 1 Introduction In [ IZ26 ] , following results in [ CH19 ] and [ IK23 ] , we proved Theorem 1.1 . Let (\Omega,\mathfrak{F},P) be a probability space and H be a real Hilbert space of infinite dimension. Let (e_{n})_{n\in\mathbb{N}} be a sequence of random variables defined on (\Omega,\mathfrak{F},P) and with values in H . Let Q be a complete \sigma -finite probability measure on H such that any finite dimensional vector subspace A of H satisfies Q(A)=0 . Suppose that for all k\in\mathbb{N} , \left((e_{0},\cdots,e_{k})_{*}(P)\right) is absolutely continuous with respect to Q^{\otimes(k+1)} , so that the Radon-Nykodym derivative \frac{d\left((e_{0},\cdots,e_{k})_{*}(P)\right)}{dQ^{\otimes(k+1)}} exists, which we denote by p^{(k)}:H^{k+1}\to[0,\infty) . Then (e_{n})_{n\in\mathbb{N}} is free almost surely. Even though the references [ VTC12 ] and [ Kuk20 ] were given in our article [ IZ26 ] and were assumed to contain some examples, we probably skimmed over the question of the existence of a probability measures on an infinite dimensional Banach space annihilating all finite dimensional vector subspaces. In this article, we definitely show the existence of such probability measures on any infinite dimensional Banach space: Theorem 1.2 . Let E be an infinite dimensional Banach space over \mathbb{R} or \mathbb{C} . Then there exists a probability measure P on E with its Borel \sigma -algebra annihilating all finite dimensional vector subspaces. 2 Proof of the theorem Let E be an infinite dimensional Banach space over \mathbb{R} or \mathbb{C} and denote by \mathcal{P} the set of prime numbers. The following proof is motivated by the fact that there is no atomless probability measure on a countable set. Since (0,1)\subset\mathbb{R} admits such a probability measure \lambda (the Lebesgue measure for instance), we will define the probability measure P on E as a pushforward of \lambda by an injective and measurable map x:(0,1)\to E . The first and second parts of the proof are inspired by the note [ Lac73 ] . However, instead of an arbitrary choice of (N_{t})_{(0,1)} such that \forall t\in(0,1):N_{t}\text{ is infinite and }\forall 0<t\neq t^{\prime}<1:N_{t}\cap N_{t^{\prime}}\text{ is finite}, (1) we provide a specific construction of (N_{t})_{(0,1)} satisfying not only ( 1 ) but also some measurability conditions. The third part of the proof is the smallest one and concludes the proof cunningly. First part: For all t\in(0,1) and j\in\mathbb{N}^{*} such that 2\leq t^{-j} , consider e_{j}(t):=\max\{p\in\mathcal{P}/p\leq t^{-j}\} and if t^{-j}<2 e_{j}(t):=2 Further, define f_{j}(t):=\Pi_{i=1}^{j}e_{i}(t). Clearly, (e_{j}(t)) and (f_{j}(t)) are nondecreasing in j and nonincreasing in t . Moreover, e_{j}(t) and f_{j}(t) go to +\infty as t^{-j} goes to +\infty since there are infinitely many primes. Furthermore, (f_{j}(t)) is increasing in j since all prime numbers are greater or equal than 2. For all t\in(0,1) , consider N_{t}:=\{f_{j}(t)/j\in\mathbb{N}^{*}\}\subseteq\mathbb{N}^{*}. Clearly N_{t} is infinite for all t\in(0,1) . Let’s show that N_{t}\cap N_{t^{\prime}} is finite for all t`>t . Let m\in N_{t}\cap N_{t^{\prime}} so that there exist u,v\in\mathbb{N}^{*} such that m=f_{u}(t)=f_{v}(t^{\prime}). Since f_{u}(t) and f_{v}(t^{\prime}) are equal and are a product of u and v prime numbers respectively, we necessarily have u=v by the unique factorization property in \mathbb{N}^{*} . For any prime p\in\mathcal{P} , denote by p_{\text{succ}}\in\mathcal{P} the smallest prime number larger than p . A consequence of the prime number theorem ensures that \frac{p_{\text{succ}}-p}{p}\to 0 as p\to+\infty . Therefore, as v goes to +\infty , we have t^{\prime v}((e_{v}(t^{\prime}))_{\text{succ}}-e_{v}(t^{\prime}))=t^{\prime v}o(e_{v}(t^{\prime}))=t^{\prime v}o(t^{\prime-v})=o(1). Hence, for v large enough, we have (\frac{t^{\prime}}{t})^{v}-1>t^{\prime v}((e_{v}(t^{\prime}))_{\text{succ}}-e_{v}(t^{\prime})) which in turn implies t^{-v}-t^{\prime-v}>(e_{v}(t^{\prime}))_{\text{succ}}-e_{v}(t^{\prime}) and so t^{-v}>(e_{v}(t^{\prime}))_{\text{succ}} and therefore e_{v}(t)>e_{v}(t^{\prime}). So for v large enough, f_{v}(t^{\prime})<f_{v}(t) , a contradiction. Since (f_{j}(t)) is nondecreasing in j , then the above discussion shows that N_{t}\cap N_{t^{\prime}}=\{m\in\mathbb{N}^{*}/\exists u,v:f_{u}(t)=f_{v}(t^{\prime})\} is finite. Second part: Using the Hahn-Banach theorem multiple times and a variant of the Gram-Schmidt process, we can easily prove that there exists a sequence (x_{j},x_{j}^{*})_{j\in\mathbb{N}^{*}} such that (x_{j},x_{j}^{*})\in E\times E^{*} for all j\in\mathbb{N}^{*} , such that x_{u} has norm 1 for all u\in\mathbb{N}^{*} , x_{v}^{*}(x_{u})=0 for all u\neq v\in\mathbb{N}^{*} and x_{u}^{*}(x_{u})\neq 0 for all u\in\mathbb{N}^{*} . For all t\in(0,1) , define x_{t}=\sum_{j\in N_{t}}\frac{x_{j}}{2^{j}}. The construction of (N_{t})_{(0,1)} satisfying ( 1 ) was for instance given in [ Bud71 ] . However, the measurability of x:(0,1)\to E and Range(x) was not established. The construction of the (N_{t})_{(0,1)} that was provided in part 1 ensures these measurability properties, as the discussions here and in the first appendix show. First, observe as in [ Lac73 ] that (x_{t})_{(0,1)} is a linearly independent family in E having the cardinality of the continuum and that x:(0,1)\to x_{t}\in E is injective. Denote S:=\{p^{-\frac{1}{i}}/p\in\mathcal{P},i\in\mathbb{N}^{*}\} For any p\in\mathcal{P} and i\in\mathbb{N}^{*} , we have (e_{i})^{-1}(\{p\})=((p_{\text{succ}})^{-\frac{1}{i}},p^{-\frac{1}{i}}] In particular, for any t^{*}\in(0,1)\setminus S , we have \forall i\in\mathbb{N}^{*}:(e_{i})^{-1}(\{e_{i}(t^{*})\})=((e_{i}(t^{*}))_{\text{succ}}^{-\frac{1}{i}},e_{i}(t^{*})^{-\frac{1}{i}}) and so for all i\in\mathbb{N}^{*} , e_{i} is continuous as a function from (0,1) to \mathbb{N} in this case. In general, for all i\in\mathbb{N}^{*} , e_{i} is measurable as a function from (0,1) to \mathbb{N} , and the same goes for the f_{i} . The fact that x:(0,1)\to E is measurable is a straightforward consequence of the fact that the f_{j}(t) are measurable in t\in(0,1) for all j\in\mathbb{N} . Third part: For any measurable subset M of E , set P(M):=\lambda(x^{-1}(M)). Clearly, P is a probability measure on E . Moreover, if M is a finite dimensional subspace of E , the set M\cap Range(x) is finite and so P(M)=\lambda(x^{-1}(M\cap Range(x)))=0 since x is injective and \lambda is atomless. Therefore, the proof is finished ∎. Appendix A Measurability of Range(x) In this appendix, we show the additional result that Range(x) is a measurable subset of E . This implies that there is an uncountable measurable family of independent vectors having the cardinality of the continuum in any infinite dimensional Banach space, an important result in itself. So Let’s show that Range(x) is a measurable subset of E . Certainly, \overline{Range(x)} is a measurable subset of E since it is closed, so it suffices to show that \overline{Range(x)}\setminus Range(x) is countable. Consider the following cases: • Let t^{*}\in(0,1)\setminus S and let’s show that X_{t^{*}}:=\{l\in E/(\forall m\in\mathbb{N}^{*})(\forall\eta>0)(\exists t\in(t^{*}-\eta,t^{*}+\eta)):\lVert x_{t}-l\rVert\leq 1/m\} is at most countable. Indeed, we have from the continuity of x:(0,1)\to E at t^{*}\in(0,1)\setminus S that X_{t^{*}}=\{x_{t^{*}}\} . • Let’s show that X_{0}:=\{l\in E/(\forall m\in\mathbb{N}^{*})(\forall\eta>0)(\exists t\in(0,\eta)):\lVert x_{t}-l\rVert\leq 1/m\} is at most countable. Since \lVert x_{t}\rVert=\lVert\sum_{j\in\mathbb{N}^{*}}\frac{x_{f_{j}(t)}}{2^{f_{j}(t)}}\rVert\leq\sum_{u\geq f_{1}(t)}\frac{1}{2^{u}}=\frac{1}{2^{f_{1}(t)-1}}\to 0 as t\to 0^{+} , this shows that X_{0}=\{0\} . • Let’s show that X_{1}:=\{l\in E/(\forall m\in\mathbb{N}^{*})(\forall\eta>0)(\exists t\in(1-\eta,1)):\lVert x_{t}-l\rVert\leq 1/m\} is at most countable. We have x_{j}^{*}(x_{t})=\sum_{u\in\mathbb{N}^{*}}\frac{x_{j}^{*}(x_{f_{u}(t)})}{2^{f_{u}(t)}}=\begin{cases}\frac{x_{j}^{*}(x_{j})}{2^{j}}\text{ if }(\exists u\in\mathbb{N}^{*})f_{u}(t)=j\\ 0\text{ otherwise}\end{cases} In particular, \lVert\frac{x_{j}^{*}(x_{t})}{x_{j}^{*}(x_{j})}x_{j}\rVert\leq\frac{1}{2^{j}} , which is summable. For all u\in\mathbb{N}^{*} , the sequence of natural numbers f_{u}(t) converges to 2^{u} as t\to 1^{-} , so it is equal to 2^{u} for t close enough to 1^{-} . Therefore x_{t}=\sum_{u\in\mathbb{N}}\frac{x_{f_{u}(t)}}{2^{f_{u}(t)}}\to\sum_{u\in\mathbb{N}^{*}}\frac{x_{2^{u}}}{2^{2^{u}}} as t\to 1^{-} by interchanging the limit with the sum (the limit on the right obviously exists). This shows that X_{1}=\{\sum_{u\in\mathbb{N}^{*}}\frac{x_{2^{u}}}{2^{2^{u}}}\} . • Let t^{*}\in S and let’s show that X_{t^{*}}^{-}:=\{l\in

Read the complete text