REVIEW 1 major objections 4 minor 11 references
The structure of sets with cube-avoiding sumsets
T0 review · 1 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read Dense summands avoiding a cube in a finite abelian group power are almost entirely captured by a bounded common coordinate set.
desk verdict A genuinely new bounded-dimensional dichotomy for cube-avoiding sumsets, with a sound proof that hinges on an external correlation theorem the authors need to state more carefully. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
Two tools carry the argument. The first is a simultaneous regularity lemma (Lemma 2.2): given accuracy parameters, any finite family of subsets of $G^n$ has a nonempty coordinate set $I$ of size bounded in terms of the parameters alone such that most coordinate slices of every set are $(r,\beta)$-pseudorandom, meaning restrictions to at most $r$ coordinates change density by at most $\beta$. The second is a correlation theorem for product-space models (restated as Theorems 2.5 and 3.5): dense pseudorandom Boolean functions on independent identically distributed coordinates whose joint distribution has correlation strictly below $1$ must have positive expected product. Applied to the distribution supported on tuples with sum in $Z_0$, this yields the key proposition that dense pseudorandom summands contain a positive fraction of tuples whose sum lands in $Z_0^n$. The contradiction argument then shows the dense pseudorandom slices over $I$ cannot have sum in $Z_0^I$, forcing the cylinders to avoid it.
What would settle it
Construct, for some finite abelian $G$ and some $Z_0$ not contained in a strict coset, a sequence of dense $E,F\subset G^n$ with $(E+F)\cap Z_0^n=\emptyset$ but with no nonempty $I\subset[n]$ of size below the claimed $C(G,\varepsilon)$ such that both sets are $\varepsilon$-close to cylinders over $I$; Proposition 2.3 says such examples cannot exist, so the first such example would refute the theorem. A cheaper falsifier is to check the imported correlation theorem directly: exhibit dense pseudorandom $f,g$ on i.i.d. coordinates with $\rho(P)<1$ and $\mathbb{E}fg$ arbitrarily small, which would contradict Theorem 2.5.
Extended reading notes
Core claim
The paper's central claim is Theorem 1.3: for $d\ge 2$, every finite abelian group $G$, every $Z_0\subseteq G$ not contained in a strict coset, and every $\varepsilon>0$, there is a constant $C=C(d,G,\varepsilon)$ such that whenever $E_1,\dots,E_d\subseteq G^n$ satisfy $(E_1+\cdots+E_d)\cap Z_0^n=\emptyset$, there is a nonempty $I\subseteq[n]$ with $|I|<C$ and subsets $E'_j\subseteq G^I$ such that $|E_j\setminus(E'_j\times G^{I^c})|\le \varepsilon|G^n|$ for all $j$ and $(E'_1+\cdots+E'_d)\cap Z_0^I=\emptyset$. In words, dense summands that jointly avoid the cube are $\varepsilon$-close to sets depending only on a common bounded family of coordinates; the theorem's content is that this bounded family can be chosen before seeing $n$.
Load-bearing premise
The proof imports a theorem saying that dense pseudorandom Boolean functions on independent coordinates with correlation below one must have positive expected product; the paper restates this result rather than proving it, and if the restatement misses a hypothesis (such as a restriction on the dimension relative to the pseudorandomness parameters) the structure theorem would lose its main support.
Editorial extensions
If this is right
- For any fixed $G$ and $Z_0$, the family of dense $d$-tuples avoiding $Z_0^n$ is, up to $\varepsilon$-error, parameterised by finitely many coordinate sets $I$ and low-dimensional avoiders in $G^I$; the classification is independent of $n$.
- When one summand is sparse the theorem degenerates to a trivial bound, and otherwise all summands must share the same structured coordinate set; in particular no dense example can split structure across disjoint coordinate sets while avoiding $Z_0^n$.
- The condition on $Z_0$ is necessary: if $Z_0$ lies in a proper coset of a subgroup, Examples 3.1 and 3.2 produce high-dimensional avoiders with no bounded common coordinate structure.
- For the prime two-summand case, if $E+F$ avoids $\{0,1\}^n$ then $E$ and $F$ are $\varepsilon$-close to $E'\times\mathbb{Z}_p^{I^c}$ and $F'\times\mathbb{Z}_p^{I^c}$ for a common nonempty $I$ of size bounded by $C(p,\varepsilon)$.
Reading between the lines
- Beyond the paper: the same regularity-plus-correlation scheme should apply to any tensor-product constraint $f^{\otimes n}(E_1,\dots,E_d)\cap Z_0^n=\emptyset$ for which the uniform distribution on the solution set of $f(x_1,\dots,x_d)\in Z_0$ has correlation $<1$; the paper raises Latin squares as a candidate, and the bottleneck would be verifying the correlation condition.
- Beyond the paper: if the correlation theorem can be made effective with polynomial dependence, the same proof would give a bound on $|I|$ polynomial in $\log(1/\varepsilon)$, bearing on the paper's Question 4.1, which asks specifically for $|I|=O(\log(\varepsilon^{-1}))$.
- Beyond the paper: the theorem implies a form of finitary stability for cube-avoiding dense configurations at every fixed accuracy; combining it with a classification of the low-dimensional avoiders in $G^I$ would yield a complete approximate description of all dense avoiders in arbitrary dimension.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proves a structural theorem for dense subsets of G^n (G a finite abelian group) whose sumset avoids a power Z_0^n of a forbidden set Z_0. If Z_0 is not contained in any strict coset, then for any fixed number d >= 2 of summands and any epsilon > 0, any family E_1,...,E_d in G^n with empty (E_1+...+E_d) ∩ Z_0^n is epsilon-close to a family depending on a common set I of coordinates with 0 < |I| < C(d,G,epsilon), where C is independent of n. The proof combines a simultaneous regularity lemma (Lemma 2.2 and Lemma 3.3) with a pseudorandom-summands proposition (Proposition 2.3 and Proposition 3.4) that is deduced from an external correlation theorem of Hazła, Holenstein, and Mossel (Theorems 2.5 and 3.5). The paper also gives examples showing that the assumption on Z_0 is necessary, proves the two-summand case over Z_p and finite abelian groups, and concludes with open problems on optimal bounds and generalisations to latin squares.
Significance. If correct, this is a strong and clean structural result in high-dimensional additive combinatorics: the conclusion is uniform in n, the allowed coordinate set is common to all summands, and the non-degeneracy condition on Z_0 is shown to be necessary. The internal reductions are elegant and the energy bookkeeping in Lemma 2.2 is coherent. The paper is honest about its dependence on an external theorem, and the main risk is the faithfulness of that import rather than an internal error. The d-variable extension to several summands is a valuable contribution, but it rests entirely on the multi-variable form of the quoted correlation theorem.
major comments (1)
- [Section 2.3, Theorem 2.5; Section 3.2, Theorem 3.5] The proof of Theorems 1.1–1.3 rests entirely on the import of [7, Theorem 7.1], stated as Theorems 2.5 and 3.5. The manuscript says these are 'special cases' but does not verify that the hypotheses match the source. In particular: (i) Definition 2.1 defines pseudorandomness with respect to the uniform density on X^n, whereas the theorem in [7] may define it with respect to the product of the marginals of the distribution P; in the present applications the marginals are uniform, so this is repairable, but the statement as written is not literally the cited theorem. (ii) The quantifier order in Theorem 2.5 asserts constants r, beta, c that are independent of the finite alphabet Omega and of n; if [7] allows dependence on Omega or on n, the applications still go through because G is fixed and the small-n case is trivial, but the quoted statement needs qualification to be faithful. (iii) Theorem 3.5 is asserted to be the d-variable special case of [7, Theorem 7.1], but no justification is given that the cited theorem covers d >= 3. Proposition 3.4, and hence Theorem 1.3, depends on this d-variable version. The authors should either provide a proof of Theorem 3.5 or give a precise statement of [7, Theorem 7.1] showing that it indeed implies the claimed multi-variable version. As it stands, this is a load-bearing gap in the proof of the main theorem.
minor comments (4)
- [Lemma 2.2, proof] The proof uses the symbol p for the size of X without defining it; the energy-increment bound and the growth bound should be stated in terms of |X|. Also, the bound on |I_{s+1} \ I_s| appears as 'p|I_s|r' in the typeset text; this should presumably be p^{|I_s|} r or |X|^{|I_s|} r.
- [Lemma 2.2, proof] The argument to ensure that I is non-empty is incomplete: if no slice is non-pseudorandom at the empty coordinate set, then the first iteration adds no coordinates and I remains empty. This is easily fixed by adding an arbitrary coordinate to I_0 at the start.
- [Section 3.2, definition of rho(P)] The symbol U_j is used both for a single coordinate and for the vector of all other coordinates in the definition of rho(P). This overloading is confusing and should be disambiguated.
- [Example 3.2] The final sentence appears to be missing the conclusion '(E_1 + ... + E_d) ∩ Z_0^n = ∅'; the equality and the empty-set symbol are lost in the typesetting.
Circularity Check
No circular reduction: the central derivation is a genuine new theorem; the only load-bearing external input is the independent correlation theorem of Hazła–Holenstein–Mossel, and the adapted regularity lemma is re-proved in the paper.
full rationale
I walked the derivation chain: Theorem 1.1 follows from Lemma 2.2 and Proposition 2.3; Proposition 2.3 follows from Theorem 2.5, which is explicitly stated as a special case of the external result [7, Theorem 7.1] by Hazła, Holenstein and Mossel. The same external theorem, in the form of Theorem 3.5, supports Proposition 3.4 and hence Theorem 1.3. There is no step in which a parameter is fitted to the very quantity being predicted: the constants β, c and r are produced by the external correlation theorem from the input data (d, G, Z0, ε), and the pseudorandomness condition is verified inside the proof rather than assumed as the conclusion. The regularity lemma is adapted from [9, Lemma 3.2], but the paper includes a full proof via an energy-increment argument, so the result does not reduce to the citation. The companion-paper self-citation [8] is motivational only and is not used in any proof. The only genuine risk is whether the quoted special cases of [7, Theorem 7.1] faithfully capture all hypotheses of that theorem; if they do not, the theorems would lack support, but that would be a correctness or fidelity issue, not a circularity of the derivation. Accordingly, the paper shows no self-definitional step, no fitted input renamed as a prediction, and no load-bearing self-citation chain.
Assumptions & free parameters
assumptions (5)
- domain assumption Z0 is not contained in any strict coset of G
- standard math Theorem 2.5, a special case of [7, Theorem 7.1]
- standard math Theorem 3.5, the multi-summand version of [7, Theorem 7.1]
- standard math Basic finite abelian group structure (subgroups, cosets, quotients)
- standard math Basic probabilistic and energy-increment facts (variance, Cauchy-Schwarz)
Cite this review
Pith. "Pith review of The structure of sets with cube-avoiding sumsets." pith.science (2026). https://pith.science/paper/L7A3IGPM
@misc{pith2026241114145,
author = {Pith},
title = {Pith review of: The structure of sets with cube-avoiding sumsets},
year = {2026},
howpublished = {\url{https://pith.science/paper/L7A3IGPM}},
note = {Machine review of arXiv:2411.14145}
}
abstract
We prove that if $d \ge 2$ is an integer, $G$ is a finite abelian group, $Z_0$ is a subset of $G$ not contained in any strict coset in $G$, and $E_1,\dots,E_d$ are dense subsets of $G^n$ such that the sumset $E_1+\dots+E_d$ avoids $Z_0^n$ then $E_1, \dots, E_d$ essentially have bounded dimension. More precisely, they are almost entirely contained in sets $E_1' \times G^{I^c}, \dots, E_d' \times G^{I^c}$, where the size of $I \subset [n]$ is non-zero and independent of $n$, and $E_1',\dots,E_d'$ are subsets of $G^{I}$ such that the sumset $E_1'+\dots+E_d'$ avoids $Z_0^I$.
Reference graph
Works this paper leans on
-
[9]
P. Keevash, N. Lifshitz, E. Long, D. Minzer, Forbidden intersection for codes , Jour. Lon- don. Math. Soc. 108 (2023), 2037-2083
work page 2023
-
[7]
J. Haz ˛ła, T. Holenstein, E. Mossel, Product space models of correlation: Between noise stability and additive combinatorics , Discrete Anal. 19 (2018)
work page 2018
-
[1]
N. Alon, N. Linial, R. Meshulam, Additive bases of vector spaces over prime fields , J. Combin. Theory Ser. A 57 (1991), 203-210
work page 1991
- [2]
-
[3]
E. Breuillard, B. J. Green, T. Tao, The structure of approximate groups , Publ. Math. IHES 116 (2012), 115-221
work page 2012
-
[4]
H. Furstenberg, Ergodic behavior of diagonal measures and a theorem of Szeme rédi on arithmetic progressions, J. Anal. Math. 31 (1977), 204-256
work page 1977
-
[5]
W. T. Gowers, B. J. Green, F. Manners, T. Tao, Marton ’s Conjecture in abelian groups with bounded torsion , arXiv:2404.02244 (2024)
arXiv 2024
-
[6]
B. J. Green, On Sárközy’s theorem for shifted primes , J. Amer. Math. Soc. 37 (2024), 1121-1201
work page 2024
Show all 11 references
-
[8]
Karam, P
T. Karam, P. Keevash, Extremal expansions by cubes in the torus , in preparation
-
[10]
Sárközy, On difference sets of sequences of integers
A. Sárközy, On difference sets of sequences of integers. I , Acta Math. Acad. Sci. Hungar. 31 (1978), 125-149
1978
-
[11]
Sárközy, On difference sets of sequences of integers
A. Sárközy, On difference sets of sequences of integers. III , Acta Math. Acad. Sci. Hungar. 31 (1978), 355–386. 12
1978
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.