Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated

The Sidon Decomposition Problem in Abelian Groups of Bounded Torsion

Full paper in Markdown.


Mark Lewko, “The Sidon Decomposition Problem in Abelian Groups of Bounded Torsion,” arXiv:2606.06669v1 (2026).

Main result

Let GG be a compact abelian group and let Γ=G^\Gamma=\widehat G be its discrete dual. Lewko proves that if Γ\Gamma has bounded torsion---there is an NN such that Nγ=0N\gamma=0 for every γ∈Γ\gamma\in\Gamma---then a set Λ⊆Γ∖{0}\Lambda\subseteq\Gamma\setminus\{0\} is Sidon if and only if it is a finite union of quasi-independent sets. Here quasi-independence means that no nonzero finitely supported coefficient vector in {−1,0,1}Λ\{-1,0,1\}^{\Lambda} sums to zero. This is Theorem 1.2 (Section 1, source lines 17--31).

The quantitative core is the following finite prime-power statement. Let q=psq=p^s and A⊆(Z/qZ)nA\subseteq(\mathbb Z/q\mathbb Z)^n. If there is a δ>0\delta>0 such that every B⊆AB\subseteq A contains a quasi-independent Q⊆BQ\subseteq B with ∣Q∣≥δ∣B∣|Q|\geq\delta|B|, then

A=⨆i=1kAi,k≤⌈slog⁡2pδ⌉,A=\bigsqcup_{i=1}^k A_i, \qquad k\leq\left\lceil\frac{s\log_2 p}{\delta}\right\rceil,

with every AiA_i quasi-independent. This is Theorem 1.3 (Section 1, source lines 35--41), proved in Section 5. The bound is uniform in nn and ∣A∣|A|. Pisier's arithmetic characterization supplies the proportional hypothesis for finite subsets of a Sidon set (Theorem 1.1, source lines 21--25), and a compactness argument turns the uniform finite coloring into an infinite decomposition (Section 6, source lines 293--317).

Mechanisms potentially useful for E0774

For subsets of N⊂Z\mathbb N\subset\mathbb Z, dissociation in E0774 is the same signed-relation condition as quasi-independence: canceling the intersection of two unequal finite subsets converts an equality of subset sums into a nonzero {−1,0,1}\{-1,0,1\} relation, and conversely.

Lewko's proof packages those relations as follows:

  1. For each T⊆AT\subseteq A, collect the signed zero-relations supported on TT. Proportional quasi-independence gives at least 2δ∣T∣2^{\delta|T|} distinct subset sums and hence growth of the subgroup generated by TT (Section 3, source lines 75--85).
  2. In exponent psp^s, compare that growth with the finite coefficient space (Z/psZ)T(\mathbb Z/p^s\mathbb Z)^T. Kernel counting and reduction modulo pp show that the reduced signed relations supported on TT span a space of dimension at most
(1−δslog⁡2p)∣T∣.\left(1-\frac{\delta}{s\log_2 p}\right)|T|.

This is Corollary 3.3 (source lines 87--113). 3. A support-partition theorem then turns any uniform local dimension gap into a coloring: if a collection C⊆FX∖{0}\mathcal C\subseteq\mathbb F^X\setminus\{0\} satisfies dim⁡span⁡{c∈C:supp⁡c⊆Y}≤(k−1)∣Y∣/k\dim\operatorname{span}\{c\in\mathcal C:\operatorname{supp}c\subseteq Y\} \leq (k-1)|Y|/k for every Y⊆XY\subseteq X, then XX can be partitioned into kk classes, none containing the support of a member of C\mathcal C. This is Theorem 4.3 (source lines 171--201), derived from Rado--Horn. 4. Because every nonzero signed coefficient ±1\pm1 stays nonzero modulo pp, the support-avoiding coloring rules out the original signed relations and makes every color class quasi-independent (Section 5, source lines 203--229).

Why the theorem does not settle E0774

The paper explicitly says that the analogous problem for Γ=Z\Gamma=\mathbb Z is not addressed (Section 1, source line 43). The exclusion is structural, not terminological: Z\mathbb Z has no bounded exponent, so it has neither the finite coefficient space (Z/psZ)T(\mathbb Z/p^s\mathbb Z)^T nor a fixed prime-power coordinate factor to which Bourgain's projection theorem can reduce the problem.

In particular, the decisive count ∣ker⁡ΦT∣ ∣⟨T⟩∣=q∣T∣|\ker\Phi_T|\,|\langle T\rangle|=q^{|T|} takes place between finite groups. For a nonempty T⊂ZT\subset\mathbb Z, the generated subgroup ⟨T⟩\langle T\rangle is infinite, so distinct subset sums do not yield the same finite-cardinality bound on the relation kernel. Reducing the integers modulo a prime does not repair this uniformly: it introduces modular zero-relations that need not be integer zero-relations, while no single modulus is supplied by the E0774 hypothesis. Thus the Rado--Horn partition mechanism remains relevant, but the bounded-torsion argument that establishes its dimension hypothesis is exactly the missing transfer step for E0774.