Isaac’s Condition (TU) is false

Ralf Stephan

arXiv: 2610.21899v1 · PDF · 15 October 2026
Categories: math.NT
Licence: CC0 1.0 The site shows the text.

Abstract

In 2006 Isaac reduced the simple normality to base 2 of \sqrt s, for s not a perfect square, to a property of dyadic expansions that he called Condition (TU): the upper frequency of the digit 1 in \sqrt\nu, along any subsequence, depends only on the tail of the expansion of \nu. We show that Condition (TU) is false, already for the full sequence. We give two proofs. The first is elementary and turns Isaac’s own reduction against the condition. The second uses the theorem of Kaufman and Mosquera–Shmerkin on the Fourier decay of smooth images of self-similar measures, and gives counterexamples for almost every point of a Cantor measure. All results are machine-verified in Lean 4; the second proof uses the two cited theorems as hypotheses.

Items 26

Theorem

Theorem 1.1

NTXS

Condition (TU) is false. More precisely, there are real numbers \nu,\nu'\in(0,1) with \nu irrational and \nu'-\nu\in\mathbb{Z}[\tfrac12] such that

\limsup_{N\to\infty}f_N(\sqrt\nu)\ne\limsup_{N\to\infty}f_N(\sqrt{\nu'}).

Proof of Theorem 1.1

Assume \textup{(TU)}_1 and take x from the construction. Put \nu=x^2 and \nu'=x^2+1/4. Both lie in (0,1), \nu is irrational by Lemma 3.4(1), and \nu'-\nu=1/4. By Lemma 3.4(2), \limsup f_N(\sqrt\nu)=0, while \limsup f_N(\sqrt{\nu'})\ge2/5 by Lemma 3.4(3). This contradicts \textup{(TU)}_1.

Uses Lemma 3.4

Proof of Theorem 1.1

Let g(\omega)=\sqrt{\omega^2/4+1/4}. Then g is smooth with g''=1/(16g^3)>0. By Theorems 4.1 and 4.2 and Lemma 4.4(1), g(\omega) is simply normal to base 2 for \mu-almost every \omega. Since \mu has no atoms, also \omega^2 is irrational for \mu-almost every \omega: the exceptional set is contained in the countable set \{\pm\sqrt q:q\in\mathbb{Q}\}. Fix \omega with both properties. Then 0<\omega\le1/7. Put \nu=\omega^2/4 and \nu'=\nu+1/4=g(\omega)^2. Both lie in (0,1), \nu is irrational, and \nu'-\nu\in\mathbb{Z}[\tfrac12]. Since X_{k+1}(\omega/2)=X_k(\omega), Lemma 4.4(2) gives f_N(\omega/2)\le1/3+2/N\le2/5 for N\ge30. Hence

\limsup f_N(\sqrt\nu)\le\tfrac25<\tfrac12=\lim f_N(\sqrt{\nu'}).\qedhere

Uses Theorem 4.1, Theorem 4.2, Lemma 4.4

Remark

Remark 1.2

3PBG

(TU) does hold for Lebesgue-almost every \nu. By Borel’s theorem and since \nu\mapsto\sqrt{\nu+d} maps null sets to null sets, for almost every \nu the numbers \sqrt{\nu+d} are simply normal for all d\in\mathbb{Z}[\tfrac12] at once. So whether (TU) holds “almost everywhere” depends on the reference measure, and the second proof shows that this cannot be repaired by passing to a measure-theoretic form of (TU).

Definition

Definition 2.1 (Isaac)

20YB

Condition (TU) is the statement: for all \nu,\nu'\in(0,1) with \nu irrational and the same tail, and for every strictly increasing sequence (n_i) of natural numbers,

\limsup_{i\to\infty}f_{n_i}(\sqrt\nu)=\limsup_{i\to\infty}f_{n_i}(\sqrt{\nu'}).

We write \textup{(TU)}_1 for the same statement with only n_i=i.

Lemma

Lemma 2.2

ZPRU

Let a,b be real and s,K\in\mathbb{N}.

If X_{k+s}(a)=X_k(b) for all k\ge K, then f_N(a)-f_N(b)\to0.
If X_{k+s}(a)=1-X_k(b) for all k\ge K, then f_N(a)+f_N(b)\to1.

Used by Proof

Proof of Lemma 2.2

In both cases |S_N(a)-S_N(b)|, respectively |S_N(a)+S_N(b)-N|, is at most K+2s for every N.

Lemma

Lemma 2.3

6PLV

Let z,w be real and m\ge1 with X_m(z)=0 and \lfloor2^mz\rfloor\le\lfloor2^mw\rfloor\le\lfloor2^mz\rfloor+1. Then \lfloor2^kw\rfloor=\lfloor2^kz\rfloor for all k<m; in particular z and w have the same first m-1 digits. Consequently, if X_n(z)=0, then S_N(z+2^{-n})\le S_N(z)+1 for all N.

Used by Proof

Proof of Lemma 2.3

Since \lfloor2^mz\rfloor is even, the two values at level m have the same quotient by 2, and one descends from k to k-1 using \lfloor2^{k}t\rfloor=2\lfloor2^{k-1}t\rfloor+X_{k}(t). For the second statement, z+2^{-n} agrees with z in the first n-1 digits, has digit 1 at n, and agrees with z after n.

Lemma

Lemma 2.4

JVS8

Let z be real and M\ge1. There is m>M with X_m(z)=0 and f_{m-1}(z)\ge f_M(z).

Used by Proof

Proof of Lemma 2.4

No expansion of the form X_k=1 for all k>M occurs with our digit convention, so there is a first m>M with X_m(z)=0. The digits between M and m are all 1, so the frequency does not decrease from M to m-1.

Proposition

Proposition 3.1

53BW

Assume \textup{(TU)}_1. Let y be irrational with 1/2<y<1 and y^2\in\mathbb{Z}[\tfrac12]. Then f_N(y)\ge2/5 for infinitely many N.

Used by Proof

Proof of Proposition 3.1

Put \omega_1=y/2-1/4 and \omega_2=1-y/8; both lie in (0,1). Then \omega_2^2-\omega_1^2=15/16-15y^2/64\in\mathbb{Z}[\tfrac12], and \omega_1^2 is irrational because y is. Moreover X_{k+1}(\omega_1)=X_k(y) for k\ge2, and X_{k+3}(\omega_2)=1-X_k(y) for all k, since y is irrational. If f_N(y)<2/5 for all large N, Lemma 2.2 gives \limsup f_N(\omega_1)\le9/20 and \limsup f_N(\omega_2)\ge11/20, contradicting \textup{(TU)}_1 for the pair (\omega_1^2,\omega_2^2).

Uses Lemma 2.2

Lemma

Lemma 3.2

VX06

Let q\ge2 and let A<2^{q-1} be odd. Then A^2+4^q is not a perfect square. Consequently, if n\ge3 and A<2^{n-2} is odd, then y(A/2^n) is irrational. It lies in (1/2,1) and has dyadic square.

Used by Proof

Proof of Lemma 3.2

If A^2+4^q=Z^2, then (Z-A)(Z+A)=2^{2q}, so Z-A=2^\alpha, Z+A=2^\beta with \alpha<\beta. As A is odd, \alpha=1, whence A=2^{2q-2}-1\ge 2^{q-1}, a contradiction. For the second statement, y(A/2^n)=\sqrt{A^2+4^{n-1}}/2^n; and A/2^n<1/4.

Lemma

Lemma 3.3

HWSI

Assume \textup{(TU)}_1. Let n\ge3, let A<2^{n-2} be odd and let B\in\mathbb{N}. Then there is m\ge B such that X_m(y(A/2^n))=0 and f_{m-1}(y(A/2^n))\ge2/5.

Proof of Lemma 3.3

By Proposition 3.1 and Lemma 3.2, there is some M>B with f_M(y(A/2^n))\ge2/5; now apply Lemma 2.4.

Uses Proposition 3.1, Lemma 3.2, Lemma 2.4

Lemma

Lemma 3.4

I1U8

In the construction above:

x^2 is a Liouville number; in particular it is irrational;
S_N(x)^2\le N+1 for all N; in particular f_N(x)\to0;
f_N(y(x))\ge2/5 for infinitely many N.

Used by Proof

Proof of Lemma 3.4

(1) We have 0<x^2-x_k^2\le\frac34\,2^{-n_{k+1}}<(4^{n_k})^{-k} because n_{k+1}\ge2kn_k, and x_k^2=A_k^2/4^{n_k}.

(2) Given N, let j be least with N<n_{j+1}-1. By Lemma 2.3, applied at m=n_{j+1}-1, the numbers x and x_j have the same first N digits, and S_N(x_j)\le j+1 by the second part of Lemma 2.3. If j\ge1, then N\ge n_j-1\ge3\cdot4^j-1\ge(j+1)^2-1.

(3) Let m=n_{j+1}. By the choice of n_{j+1}, X_m(y_j)=0 and f_{m-1}(y_j)\ge2/5. Since y(x)+y_j>1,

0\le y(x)-y_j\le x^2-x_j^2\le\tfrac34\,2^{-m},

so \lfloor2^my_j\rfloor\le\lfloor2^my(x)\rfloor\le\lfloor2^my_j\rfloor+1. By Lemma 2.3, y(x) and y_j have the same first m-1 digits, so f_{m-1}(y(x))\ge2/5 for every j.

Uses Lemma 2.3

Remark

Remark 3.5

PG02

One can replace Proposition 3.1 by [Isa06, Thm. 2]: under (TU), y_j=\sqrt{4^{n_j-1}+A_j^2}/2^{n_j} is a shift of the square root of a non-square, hence simply normal. Proposition 3.1 avoids this and the passage through \{\sqrt s\}. The gap condition n_{j+1}\ge4(j+1)n_j+4 is stronger than needed for (2) and (3); it is chosen so that x^2 is a Liouville number.

Theorem

Theorem 4.1 (Mosquera–Shmerkin)

73UW

Let F\in C^2(\mathbb{R}) with F''>0, and let \mu=\mu^p_{a,t} be a homogeneous self-similar measure with all p_i>0 and without atoms. Then there are \sigma>0 and C>0 such that

|\widehat{F\mu}(u)|\le C|u|^{-\sigma}\qquad(|u|\ge1).

Used by Proof

Theorem

Theorem 4.2 (Davenport–Erdős–LeVeque)

2IJF

Let \mu be a Borel probability measure on \mathbb{R} with |\widehat\mu(u)|\le C|u|^{-\sigma} for |u|\ge1, where \sigma>0. Then \mu-almost every z is simply normal to base 2, that is, f_N(z)\to1/2.

Used by Proof

Remark

Remark 4.3

VAP7

In [MS18, Thm. 3.1] it is only assumed that \mu is not a single atom; we assume that \mu has no atoms, which is all we need. The theorem of [DEL63] is stated for Lebesgue measure on an interval: if \sum_NN^{-1}I(N)<\infty for every m\ne0, where I(N) is the mean square of the Weyl sum N^{-1}\sum_{n\le N}e(ms_n(x)), then s_n(x) is uniformly distributed modulo 1 for almost all x. Its proof [DEL63, §2] uses only that the measure is finite, so it holds for any Borel probability measure \mu. For s_n(x)=2^nx one has

I(N)=N^{-2}\sum_{n,n'\le N}\widehat\mu\bigl(m(2^n-2^{n'})\bigr)\ll N^{-1}

under power decay, so \mu-almost every z is normal to base 2, as noted in [MS18, §1]. We record only simple normality to base 2.

Lemma

Lemma 4.4

BBW5

The map \pi is injective and \mu has no atoms.
For every \iota, X_k(\pi(\iota))=0 unless 3\mid k. Hence S_N(\pi(\iota))\le\lfloor N/3\rfloor for all N.

Used by Proof

Proof of Lemma 4.4

For every m, 8^m\pi(\iota)=\sum_{k<m}\iota_k8^{m-1-k}+R with 0\le R\le1/7. Hence \lfloor8^m\pi(\iota)\rfloor determines \iota_0,\dots,\iota_{m-1}, and \lfloor2^{3m+r}\pi(\iota)\rfloor is even for r=1,2. Each point of \{0,1\}^\mathbb{N} has product measure at most 2^{-n} for every n.

Remark

Remark 4.5

6UCC

(i) The measure \mu is an affine image of the Bernoulli convolution \nu_{1/8}, the distribution of \sum_{n\ge1}\pm8^{-n}. According to [MS18, §1], Kaufman [Kau84] proved \dim_F(F\nu_a)>0 for all a\in(0,1/2) and all C^2 diffeomorphisms F with F''>0, so for this measure Theorem 4.1 is already due to him; Theorem 4.1 extends and quantifies his argument.

(ii) The proof shows more: (TU) fails at (\omega^2/4,\omega^2/4+1/4) for \mu-almost every \omega, a set of Hausdorff dimension 1/3. The same argument works for the Bernoulli measure of parameter p\ne1/2 on the dyadic digits. That measure gives a set of dimension close to 1 for p close to 1/2. It requires the strong law of large numbers in place of Lemma 4.4(2).

Source and licence

The text on this page is from arXiv:2610.21899v1 by Ralf Stephan, under the licence CC0 1.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/2610/2610.21899.jsonl.