betaFeedback

WorksMathematics

A remark on the independence step in Shkredov's approach to Zaremba's conjecture

Abstract

The proof of Proposition 35 in Shkredov's preprint arXiv:2603.14116v2 applies a repulsion inequality (Corollary 34) to three denominators x, y, z, which must be C-independent for the vector C of formula (116) of that paper. We show that, in the situation of that proof, this independence cannot hold: the identity (75) of the paper, d(y,z)x + d(z,x)y + d(x,y)z = 0, is a nonzero integer relation whose coefficients lie in the box defined by C as soon as two of the three inequalities needed for the contradiction in that proof are satisfied. The same holds for the triple (x, y_0, z) to which Lemma 32 is applied. We also record that the deduction drawn from Proposition 35 in Section 5.2 does not address the second alternative of that proposition. These remarks concern the proofs only: they do not contradict the statements of the theorems of that paper, nor of arXiv:2605.02518v2, whose Section 4 follows the same framework. Nothing in this note has been checked by a professional mathematician or by any third party.

Setting

We use the notation of [1] (version 2). Let qq be large, t=q1/2−εt=q^{1/2-\varepsilon}, N=q/t2N=q/t^2 and H=N9/20H=N^{9/20}. If an integer w=K(r1,…,rs)w=K(r_1,\ldots,r_s) is the denominator of a convergent of a fraction, write w^=K(r2,…,rs)\widehat{w}=K(r_2,\ldots,r_s) for the corresponding numerator, and put

d(x,y)=xy^−yx^citep‘[1,(70)]‘.d(x,y)=x\widehat{y}-y\widehat{x}\quad{citep}`[1, (70)]`.

For all such x,y,zx,y,z one has the identity

d(y,z)x+d(z,x)y+d(x,y)z=0(1)d(y,z)x+d(z,x)y+d(x,y)z=0 \tag*{(1)}

[1 (75)].

By [1 Definition 24], elements x1,x2,x3x_1,x_2,x_3 of Z/qZ\mathbb{Z}/q\mathbb{Z} are C⃗\vec{C}-independent if ∑αjxj≡0(modq)\sum\alpha_jx_j\equiv0\pmod q with integers ∣αj∣≤Cj|\alpha_j|\le C_j forces α1=α2=α3=0\alpha_1=\alpha_2=\alpha_3=0. A relation ∑αjxj=0\sum\alpha_jx_j=0 in Z\mathbb{Z} is in particular such a congruence. The conclusion of [1 Lemma 32] is stated for equations αx+βy+γz=0\alpha x+\beta y+\gamma z=0 in integers.

The step in question. In the proof of [1 Proposition 35] one has points a<b<ca<b<c of an interval II, with I∗=c−a≤∣I∣I_* = c-a\le|I|, and the following integers:

  • x=qs(a)x=q_s(a), a convergent denominator of a/qa/q;

  • y0=y0(b)y_0=y_0(b) and z=z(c)z=z(c), distinct M~\widetilde{M}-critical denominators of b/qb/q and c/qc/q supplied by [1 Lemma 32], with max⁡{y0,z}∈[tH,q]\max\{y_0,z\}\in[tH,\sqrt{q}]; the proof treats the case max⁡{y0,z}=y0\max\{y_0,z\}=y_0;

  • y≥y0y\ge y_0, a convergent denominator y=yn(b)≤qy=y_n(b)\le\sqrt{q} of b/qb/q with cn+1(b)y>qc_{n+1}(b)y>\sqrt{q}, and the number M~≤cn+1(b)\widetilde{M}\le c_{n+1}(b) defined by M~y=q\widetilde{M}y=\sqrt{q}.

Thus z<y0≤yz<y_0\le y. The bounds (121)–(122) of [1] are obtained through the numerators x^,y^,z^\widehat{x},\widehat{y},\widehat{z} of these convergents and read

∣b^x−qx^∣≤X1:=2I∗x,∣b^y−qy^∣≤X2:=qM~y=q,∣b^z−qz^∣≤X3:=2I∗z.(2)|\widehat{b}x-q\widehat{x}|\le X_1:=2I_*x,\qquad|\widehat{b}y-q\widehat{y}|\le X_2:=\frac{q}{\widetilde{M}y}=\sqrt{q},\qquad|\widehat{b}z-q\widehat{z}|\le X_3:=2I_*z. \tag*{(2)}

The vector C⃗\vec{C} is the one given by [1 (116)] with k=3k=3:

C1=2(12X2X3tX1)1/2,C2=2(12X1X3tX2)1/2,C3=2(12X1X2tX3)1/2.C_1=2\left(\frac{12X_2X_3}{tX_1}\right)^{1/2},\qquad C_2=2\left(\frac{12X_1X_3}{tX_2}\right)^{1/2},\qquad C_3=2\left(\frac{12X_1X_2}{tX_3}\right)^{1/2}.

The second part of Corollary 34 of [1] is applied in (123) to (x,y,z)(x,y,z) and the point bb, with this C⃗\vec{C}; its hypothesis includes the C⃗\vec{C}-independence of (x,y,z)(x,y,z). It yields C1x+C2y+C3z≥q/(4Mt)C_1x+C_2y+C_3z \ge q/(4Mt), and the proof obtains a contradiction by establishing

C1x<q12Mt,C2y<q12Mt,C3z<q12Mt,(3)C_1x < \frac{q}{12Mt}, \qquad C_2y < \frac{q}{12Mt}, \qquad C_3z < \frac{q}{12Mt}, \tag*{(3)}

through its inequalities (124)–(127). With (2) one computes

C12=48q ztx,C22=192I∗2xztq,C32=48q xtz,(4)C_1^2=\frac{48\sqrt{q}\,z}{tx}, \qquad C_2^2=\frac{192I_*^2xz}{t\sqrt{q}}, \qquad C_3^2=\frac{48\sqrt{q}\,x}{tz}, \tag*{(4)}

so that the first two inequalities of (3) are equivalent to

txz≤q3/26912M2andI∗2xy2zt<q5/227648M2.(5)txz\le\frac{q^{3/2}}{6912M^2}\qquad\text{and}\qquad I_*^2xy^2zt<\frac{q^{5/2}}{27648M^2}. \tag*{(5)}

The obstruction

Lemma 2.1. Let b,qb,q be integers and let x,yx,y be convergent denominators with numerators x^,y^\hat{x},\hat{y}. Then

qd(x,y)=y(bx−qx^)−x(by−qy^),hence∣d(x,y)∣≤x∣by−qy^∣+y∣bx−qx^∣q.qd(x,y)=y(bx-q\hat{x})-x(by-q\hat{y}), \qquad\text{hence}\qquad|d(x,y)|\le\frac{x|by-q\hat{y}|+y|bx-q\hat{x}|}{q}.

Proof. The terms in bb cancel, and what remains is q(xy^−yx^)q(x\hat{y}-y\hat{x}).

Proposition 2.2. Let x,y,zx,y,z be convergent denominators with y≠zy\ne z, let M≥1M\ge1, and assume (2) and the first two inequalities of (3). Put

(α,β,γ)=(d(y,z),d(z,x),d(x,y)).(\alpha,\beta,\gamma)=(d(y,z),d(z,x),d(x,y)).

Then (α,β,γ)(\alpha,\beta,\gamma) is a nonzero vector of integers with αx+βy+γz=0\alpha x+\beta y+\gamma z=0, and

∣α∣≤C1,∣β∣≤C2,∣γ∣≤C3.|\alpha|\le C_1,\qquad|\beta|\le C_2,\qquad|\gamma|\le C_3.

In particular (x,y,z)(x,y,z) is not C⃗\vec{C}-independent, and Corollary 34 of [1] does not apply to it.

Proof. The relation is (1). If α=d(y,z)=0\alpha=d(y,z)=0, then y^/y=z^/z\hat{y}/y=\hat{z}/z with both fractions in lowest terms, so y=zy=z, which is excluded; hence α≠0\alpha\ne0. By Lemma 2.1 and (2),

∣α∣≤z(q+2I∗y)q,∣β∣≤4I∗xzq,∣γ∣≤x(q+2I∗y)q.|\alpha|\le\frac{z(\sqrt{q}+2I_*y)}{q},\qquad|\beta|\le\frac{4I_*xz}{q},\qquad|\gamma|\le\frac{x(\sqrt{q}+2I_*y)}{q}.

Comparing squares with (4):

  • the bound for ∣β∣|\beta| is at most C2C_2 if and only if txz≤12q3/2txz\le12q^{3/2};

  • the bounds for ∣α∣|\alpha| and ∣γ∣|\gamma| are at most C1C_1 and C3C_3 respectively if and only if txz(q+2I∗y)2≤48q5/2txz(\sqrt{q}+2I_*y)^2\le48q^{5/2}.

The first condition follows from the first inequality of (5). For the second, (q+2I∗y)2≤2q+8I∗2y2(\sqrt{q}+2I_*y)^2\le2q+8I_*^2y^2, so (5) gives

txz(q+2I∗y)2<2q5/26912M2+8q5/227648M2<q5/2.txz(\sqrt{q}+2I_*y)^2<\frac{2q^{5/2}}{6912M^2}+\frac{8q^{5/2}}{27648M^2}<q^{5/2}.

Proposition 2.2 uses nothing beyond the bounds (2) and two of the inequalities (3) that the proof of [1 Proposition 35] needs for its contradiction. Hence, whenever that contradiction is available, the hypothesis of Corollary 34 fails for (x,y,z)(x,y,z).

Proposition 2.3. In the setting of Section 1, assume moreover z<y0≤yz<y_0\le y, y0≥tHy_0\ge tH, x≤Hqx\le H\sqrt{q}, M~≥1\widetilde{M}\ge1 and ∣by0−qy~0∣≤q/(M~y0)|by_0-q\widetilde{y}_0|\le q/(\widetilde{M}y_0). Then

(α0,β,γ0)=(d(y0,z),d(z,x),d(x,y0))(\alpha_0,\beta,\gamma_0)=(d(y_0,z),d(z,x),d(x,y_0))

is a nonzero vector of integers with α0x+βy0+γ0z=0\alpha_0x+\beta y_0+\gamma_0z=0 and ∣α0∣≤C1|\alpha_0|\le C_1, ∣β∣≤C2|\beta|\le C_2, ∣γ0∣≤C3|\gamma_0|\le C_3, for the same vector C⃗\vec{C}.

Proof. As before α0≠0\alpha_0\ne0 since y0≠zy_0\ne z, and the bound for β\beta is unchanged. By Lemma 2.1,

∣α0∣≤2I∗y0zq+zM~y0,∣γ0∣≤2I∗y0xq+xM~y0.|\alpha_0|\le\frac{2I^*y_0z}{q}+\frac{z}{\widetilde{M}y_0},\qquad|\gamma_0|\le\frac{2I^*y_0x}{q}+\frac{x}{\widetilde{M}y_0}.

In each bound the first term is at most C1/2C_1/2, respectively C3/2C_3/2, if and only if I∗xy02zt≤3q5/2I^*xy_0^2zt\le3q^{5/2}, which follows from (5) and y0≤yy_0\le y. The second term is at most C1/2C_1/2, respectively C3/2C_3/2, if and only if xzt≤12M~2y02qxzt\le12\widetilde{M}^2y_0^2\sqrt{q}, and indeed xzt≤Hq⋅y0⋅t≤qy02xzt\le H\sqrt{q}\cdot y_0\cdot t\le\sqrt{q}y_0^2. □

Remark 2.4. In [1] the hypotheses of Proposition 2.3 come from the statement of Lemma 32 (y0=max⁡{y0,z}≥tHy_0=\max\{y_0,z\}\ge tH, and y0y_0 is M~\widetilde{M}-critical) and from the range of xx: the critical denominators of aa considered there are at most T=qN2/7T=\sqrt{q}N^{2/7}, and x=qs(a)x=q_s(a) lies at most m2m^2 steps beyond them with partial quotients bounded by M∗≤N10−10M_*\le N^{10^{-10}}, so that x≤Hqx\le H\sqrt{q} for large qq. Thus the conclusion of Lemma 32, in the form in which it is used in the proof of Proposition 35, cannot hold in this range of parameters.

The remark following (77) in [1] anticipates small relations coming from (1) when a,b,ca,b,c are close, and the points are separated for this reason. Propositions 2.2 and 2.3 show that this separation does not suffice here: the coefficients of (1) are controlled by the residues modulo qq at the single point bb, and the inequalities that produce the contradiction are those that place the coefficients in the box.

The deduction drawn from Proposition 35

Proposition 35 of [1] concludes with an alternative: either II does not satisfy the δ\delta-Assumption, or some a∈I∩ZM−1(t)a\in I\cap Z_M^{-1}(t) has a large partial quotient (at least M∗M_*) at a denominator in [qN/H,qN][\sqrt{qN}/H,\sqrt{qN}]. In §5.2, Proposition 35 is invoked to conclude that subintervals of the relevant length “cannot satisfy δ\delta-Assumption”. This requires the second alternative to fail for every a∈I∩ZM−1(t)a\in I\cap Z_M^{-1}(t). The sentence preceding the deduction gives, for points of Z∗∩Z∗−1Z_*\cap Z_*^{-1}, an upper bound on the denominators x(a)x(a); it does not treat the partial quotients of the points of I∩ZM−1(t)I\cap Z_M^{-1}(t) at denominators in [qN/H,qN][\sqrt{qN}/H,\sqrt{qN}]. The text before Proposition 35 remarks that one should work with the smaller set (ZM(t)∩ZM∗(tH))∩(ZM(t)∩ZM∗(tH))−1(Z_M(t)\cap Z_{M_*}(tH))\cap(Z_M(t)\cap Z_{M_*}(tH))^{-1}; we did not find this restriction carried out in the statement or proof of Proposition 35, nor an argument excluding the second alternative in §5.2. We record this as a point that we could not complete from the text, independently of Section 2.

Scope

  1. Section 2 is a statement about the proof of [1 Proposition 35] as written in version 2: the triples to which Lemma 32 and Corollary 34 are applied there carry the relation (1) with coefficients in the box of C⃗\vec{C}. The results of [1] whose proofs pass through Proposition 35 (§5.2, and §5.4 which uses its conclusion) depend on a replacement for this step.

  1. The introduction of [2] states that its §4 outlines Shkredov’s argument and that its expansion theorem is implemented in the framework of the proof of Theorem 8 of [1]. To that extent Theorem 1.2 of [2] depends on the same step. We have not examined [2] beyond this.

  1. These remarks do not contradict the statements of the theorems of [1] or [2]. Results that do not use this step, such as the bound M=O(log⁡q/log⁡log⁡q)M=O(\log q/\log\log q) of [3], are not concerned.

(4) Status. Lemma 2.1 and Propositions 2.2 and 2.3 are proved above. Their application to [1] rests on our reading of formulas (70), (75), (116), (121)–(127), of Definition 24, Lemma 32 and Corollary 34 of that paper.

(5) No external check. Nothing in this note has been checked by a professional mathematician or by any third party: neither the proofs, nor the reading of [1], nor the ancillary script. The only checks are those carried out by the author with the AI model named below.

Ancillary file

anc/cramer_identity.py checks in exact integer arithmetic: Lemma 2.1 and identity (1) on convergents of random fractions (2993 configurations with q=1012+39q = 10^{12} + 39); and the inequalities ∣α∣≤C1|\alpha| \le C_1, ∣β∣≤C2|\beta| \le C_2, ∣γ∣≤C3|\gamma| \le C_3 of Proposition 2.2 on 5000 random parameter sets satisfying (5), together with those of Proposition 2.3 on the 4957 of these sets that satisfy xzt≤qy02xzt \le\sqrt{q}y_0^2. Its output is in anc/cramer_identity-output.log. These checks support the algebra; they are not part of the proofs.

Use of AI

This note was prepared with the assistance of an AI model (Claude, Anthropic), which carried out the reading of [1], the computations and the drafting under the author’s direction. The author takes responsibility for the content.

Independent researcher, France. ORCID 0009-0009-1526-1178

References

  1. [1]I. D. Shkredov, On some results of Korobov and Larcher and Zaremba’s conjecture, arXiv:2603.14116v2 (24 June 2026).arxiv.org/abs/2603.14116
  2. [2]X. Z. Zhang, Expansion in SL₂(Z/qZ) and Zaremba’s conjecture, arXiv:2605.02518v2 (8 May 2026).
  3. [3]N. G. Moschevitin, B. Murphy, I. D. Shkredov, On Korobov bound concerning Zaremba’s conjecture, arXiv:2212.14646.

Paper details

vlnc:W_2gzrsy5j
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Anthropic Claude Opus

View statement
Looking for related papers…
Contents