Pith. sign in

REVIEW 2 major objections 5 minor 34 references

VC-dimension of generalized progressions in some nonabelian groups

T0 review · 2 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read This paper proves finite, explicit upper bounds on the VC-dimension of generalized progressions in the Heisenberg group (at most 267) and in free groups (at most 3k-1).

desk verdict Two genuinely new VC-dimension bounds for generalized progressions in nonabelian groups; the free group theorem is solid, and the Heisenberg finiteness is sound, but the explicit 267/140 constants rest on a tight reconstruction of a Karpinski–Macintyre bound. read the letter →

arxiv 2505.21789 v1 pith:AHJWNI33 submitted 2025-05-27 math.GR math.COmath.LO

classification math.GRmath.COmath.LO MSC 20F6520F1820E0503C45
keywords VC-dimensiongeneralizedprogressionsHeisenberggroupfreegroupsCayleygraphNIPformulassemialgebraicfamiliesapproximate
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Generalized progressions generalize multidimensional arithmetic progressions to arbitrary groups: fix generators $a_1,\dots,a_k$ and bounds $N_1,\dots,N_k$, and take all group elements representable by words in the generators and their inverses in which $a_i$ and $a_i^{-1}$ together appear at most $N_i$ times. The paper asks whether the family of all left translates of all such progressions has bounded VC-dimension, the standard measure of how many different subsets a family can cut out of a finite set. It proves two concrete finiteness results: in the integer Heisenberg group with its standard two generators the VC-dimension is at most 267, and in the free group on $k$ generators it is at most $3k-1$. These are the first finite upper bounds for generalized progressions in nonabelian groups, and they matter because the VC-dimension of such progressions is the open step separating structure theorems for approximate groups from a fully general theory.

What carries the argument

The argument stands on three mechanisms. First, Theorem 4.9 gives an explicit arithmetic description of the Heisenberg progression $P(N_1,N_2)$: a point $(a,b,c)$ belongs to it exactly when one of four sign cases holds, each bounding $c$ between floor-function expressions built from $a,b,N_1,N_2$. This description makes the whole family uniformly quantifier-free definable in the structure $(\mathbb{Z},+,\cdot,<,0,1,E)$, and the effective semialgebraic shatter bound of Theorem 4.11, applied to a formula assembled from 14 polynomials of degree at most 2 in the parameter variables and with a constant $d(2d-1)^{\ell-1}$ traced to a bound on the number of connected components of a real algebraic variety, converts that definability into the explicit constants 267 and 140. Second, for free groups the Cayley graph is a tree, and the pseudometrics $d_i(g,x)$ that count occurrences of generator $i$ in the reduced word for $g^{-1}x$ make each progression exactly a set $\{x : d_i(g,x) \leq N_i \text{ for all } i\}$, which is connected in the tree; the proof then uses properties of minimal trees, a lemma on connected intersections, and dominating sequences to show that no set of $3k$ points can be shattered. Third, closure of bounded-VC families under Boolean combinations and a coset-counting lemma for finite-index subgroups are used to sharpen the Heisenberg constants.

What would settle it

A direct falsifier for Theorem 1.2 would be a set of 268 elements of the Heisenberg group whose every subset is cut out by a left translate of some $P(N_1,N_2)$; for Theorem 1.3, a set of $3k$ elements in $F_k$ shattered by translates of generalized progressions would do, and for $k=2$ shattering six points would refute the bound $3k-1=5$. A cheaper check of the numerical machinery is to recompute the claimed step where $4\log(648\sum_{i=0}^{5} 2^i\binom{14n}{i}) < n$ holds at $n=268$, and the companion inequality with $288\sum_{i=0}^{3} 2^i\binom{14n}{i}$ fails at $n=36$; a mismatch there would localize an arithmetic error in the constants.

Watch

Extended reading notes

Core claim

The central discovery is that the translate families $ mathcal{P}_H(A,B)$ and $ mathcal{P}_{F_k}(a_1,\dots,a_k)$ are tame in the VC sense, with explicit quantitative bounds. For the Heisenberg group, Theorem 1.2 gives $ operatorname{VC}( mathcal{P}_H(A,B)) \leq 267$, and the finer Theorem 4.12 gives $ operatorname{VC}_H(P(N_1,N_2)) \leq 140$ for a single progression; these numbers come from an exact coordinate description of $P(N_1,N_2)$ as a union of four integer-polynomial regions, a uniform quantifier-free definition in an NIP structure, and an effective bound on the shatter function of semialgebraic families. For the free group, Theorem 1.3 gives $ operatorname{VC}( mathcal{P}_{F_k}(a_1,\dots,a_k)) \leq 3k-1$, proved by viewing $F_k$ in its Cayley tree and showing that each progression is the intersection of $k$ connected sets defined by generator-counting pseudometrics; a combinatorial argument with dominating sequences rules out shattering any set of $3k$ points. The paper presents both results as a first step toward deciding whether all generalized progressions in arbitrary groups with fixed generators have rank-dependent bounded VC-dimension.

Load-bearing premise

Everything numerical in the Heisenberg theorem rests on the paper's version of the effective bound on how many different subsets a family of polynomial inequalities can cut out; if that reconstructed bound, including the constant $d(2d-1)^{\ell-1}$, is off, the numbers 267 and 140 would change, although the qualitative finiteness result would probably survive.

Editorial extensions

If this is right

  • In the Heisenberg group, no set of 268 elements can be shattered by the family of left translates of all generalized progressions generated by $A$ and $B$; each individual progression $P(N_1,N_2)$ is a VC-set with VC-dimension at most 140.
  • In the free group $F_k$, no set of $3k$ elements is shattered by translates of generalized progressions; for the integer line ($k=1$) this recovers the known fact that intervals have VC-dimension 2.
  • The Heisenberg case gives a positive answer to a first nontrivial instance of Question 3.10 (step 2, rank 2), and the free-group case is an extremal nonabelian test case for the broader Question 3.11.
  • The Heisenberg argument gives finiteness without any explicit constant through NIP definability, so the qualitative conclusion is stable and would survive even if the numerical bounds were improved or corrected.
  • The free-group proof identifies the bound as a purely tree-theoretic phenomenon, suggesting the same geometric tools apply to other groups with tree-like Cayley graphs and to related set systems such as metric balls.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The explicit constant 267 is almost certainly not the true value; the paper itself notes the bound is not optimal, and the same route would give a smaller constant if the semialgebraic shatter estimate or the number of polynomials in the defining formula were tightened.
  • The free-group proof suggests the sharp bound may be $2k$ rather than $3k-1$: the dominating-sequence argument already implies the cut-out set has size at most $2k$, so a suitable refinement of the branching lemma would close the gap.
  • A concrete way to test the transfer from free groups to arbitrary groups is to study how VC-dimension behaves under quotients: the paper shows the target family is isomorphic to a coset system in $F_k$, but also gives an example where enlarging a set by a subgroup sends finite VC-dimension to infinity.
  • A brute-force check of the semialgebraic bound on a small family with the same parameters (14 polynomials, degree 2, 5 parameters) would isolate whether the constants 267 and 140 are artifacts of the reconstruction or genuine limits of the method.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper studies the VC-dimension of the family of all left translates of all generalized progressions generated by a fixed finite set of generators, in two nonabelian settings. In the integer Heisenberg group H with standard generators A and B, it proves that the VC-dimension of the family P_H(A,B) is at most 267 (Theorem 1.2), and that each fixed progression has translate-family VC-dimension at most 140 (Theorem 4.12). In the free group F_k on k generators, it proves that the analogous family has VC-dimension at most 3k-1 (Theorem 1.3). The Heisenberg argument combines an explicit arithmetic description of generalized progressions (Theorem 4.9), a model-theoretic NIP argument for qualitative finiteness (Corollary 4.10), and an effective semialgebraic shatter-function bound attributed to Karpinski-Macintyre to obtain the explicit constants. The free group proof uses tree geometry, pseudometrics d_i, and a structural dichotomy for shattered sets to derive the 3k-1 bound, with a matching lower-bound example for k=2.

Significance. If correct, these are the first finite upper bounds on the VC-dimension of generalized-progression translate systems in nonabelian groups, answering a natural special case of the paper's motivating Question 3.11. The free group bound 3k-1 is clean, elementary in spirit, and supported by a concrete shattering construction for k=2, making it a solid contribution. The Heisenberg result is more conditional because the explicit constants depend on a reconstructed external theorem, but the qualitative finiteness (Corollary 4.10) and the detailed description of Heisenberg progressions (Theorem 4.9) are likely to be useful beyond the specific numerical bound. The paper is careful with definitions, includes a useful coset-decomposition lemma (Lemma 3.6), and the free group section is rigorous modulo minor typographical issues.

major comments (2)
  1. [Section 4, Theorem 4.11 and the following 'Explanation'] The explicit Heisenberg bounds 267 and 140 in Theorem 4.12 rest on a reconstructed version of the Karpinski-Macintyre bound rather than a verbatim quotation from [17]. The paper itself states that Theorem 4.11 'is not stated in [17] explicitly in this form', and the Explanation reconstructs the constant B = d(2d-1)^{ℓ-1} from a Milnor-type Betti-number bound. This is load-bearing: the numerical inequalities used in the proof of Theorem 4.12 are extremely tight (for example, the condition 4 log_2(648 Σ_{i=0}^5 2^i C(14·268,i)) < 268 has margin below 1 in base-2 log units, and the second bound at n=36 has a margin of about 0.5). If a faithful application of the Karpinski-Macintyre proof requires a different multiplier, such as an extra factor depending on s or on the number of polynomials defining the relevant variety, the explicit numbers 267 and 140 would fail even though the ineffective finiteness result would survive. The authors should either quote the exact theorem from [17] with the exact constants and hypotheses, or provide a complete and verifiable proof of Theorem 4.11, including the justification that the Milnor constant d(2d-1)^{ℓ-1} correctly bounds the number of connected components used in the Warren-type sum.
  2. [Section 4, proof of Theorem 4.12, step bounding πS(n) after equation (†)] The displayed bound πS(n) ≤ (648 Σ_{i=0}^5 2^i C(14n,i))^4 does not follow from Proposition 2.8 as claimed. The formula φ is a disjunction over (ε1,ε2)∈{0,1}^2 of conjunctions ρ̂ ∧ θ̂, and Proposition 2.8 bounds intersections, not unions. A union bound would give πS(n) ≤ 4 · (4 · 162 Σ_{i=0}^5 2^i C(14n,i)) = 2592 Σ_{i=0}^5 2^i C(14n,i), not a fourth power. The same issue occurs again when bounding πH_P(n) by (72 Σ_{i=0}^3 2^i C(14n,i))^4 in the second part of the theorem. The error is an overestimate, so the final numerical conclusions may still be true with the corrected (smaller) union bound, but the proof as written is logically invalid at this key step. The derivation should be corrected to a valid union bound, and the numerical inequalities should then be re-verified for n=268 and n=36 with the corrected expression.
minor comments (5)
  1. [Lemma 5.14] There is a typographical error in the inequality used in the proof: the text reads 'd_i(p, y) ≤ d_i(y, x_{1,i})' but the hypothesis of Lemma 5.10 and the choice of x_{1,i} require 'd_i(p, y) ≤ d_i(p, x_{1,i})' (i.e., the point y in X_1 should be compared with the maximizing point x_{1,i} for the distance from p). Please correct this.
  2. [Corollary 4.10] The claim that 'any formula of the form E(p(¯x, ¯y))' is NIP in the structure Z is asserted as 'a simple exercise' without proof. Since this is the basis for the qualitative finiteness result in the Heisenberg case, a brief argument (e.g., reducing modulo 2 to a Boolean combination of parity predicates) should be included.
  3. [Section 4, proof of Theorem 4.12] The statement 'One can check that the first inequality holds when n = 268' (and the analogous check for n = 36) would benefit from a reproducible verification, given the tight margins. Please include the actual values of the relevant logarithms, or a small table, so that the numerical bounds can be independently confirmed.
  4. [Section 5, Definition 5.8 paragraph] The phrase 'generalized progressions in F_k generated by a_1, . . . , a_n' should read 'a_1, . . . , a_k', since the free group has k fixed generators.
  5. [Section 3.3] There is a spelling typo in 'this sitatuation reduces to Question 3.10'; it should be 'situation'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: main bounds derived from definitions and independent external results; reconstructed Karpinski-Macintyre bound is external and not a fit to the conclusion.

full rationale

The derivation chain for Theorem 1.2 starts with an explicit description of P(N1,N2) in Theorem 4.9, proved from the word-reduction algorithm and Lemmas 4.6-4.8. The VC bound then follows by (i) quantifier-free definability in an NIP structure (Corollary 4.10) and (ii) Theorem 4.11, an external Karpinski-Macintyre bound whose proof is reconstructed in the 'Explanation' paragraph. The reconstruction is transparent about its source and its Milnor constant; even if that constant were wrong, that would be a correctness defect, not circularity, because the bound is not fitted to the numerical conclusion and is imported from an independent source. Theorem 1.3 is proved by elementary tree-geometric lemmas (5.3, 5.10, 5.11, 5.14, 5.16, 5.19) and does not invoke the authors' prior results at all. The self-citations [9,10] appear only as motivation for Question 3.10 and are not premises of Theorems 1.2 or 1.3. No parameter is fitted to data, no quantity is defined in terms of the target, and no uniqueness or ansatz is smuggled in via citation. Hence no significant circularity is present.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

No free parameters are fitted; the central claims depend only on standard mathematical background. No invented entities are introduced. The proofs build on Sauer-Shelah, the NIP property of the real field, the Karpinski-Macintyre/Milnor semialgebraic bounds, and tree properties of free group Cayley graphs.

assumptions (4)
  • standard math Sauer-Shelah lemma: if a set system has VC-dimension d then its shatter function is at most the sum over i from 0 to d of binomial(n,i).
    Used in Corollary 2.11 and Lemma 3.6 to bound shatter functions; cited from [12].
  • standard math The real ordered field (R,+,·,<) is NIP, and quantifier-free formulas transfer from Z to R via substructure embedding.
    Used in Corollary 4.10 and Theorem 4.12 to transfer shatter bounds from the integers to the reals; the real field NIP is standard.
  • standard math Karpinski-Macintyre theorem on VC-dimension of semialgebraic families and Milnor's bound on Betti numbers of real algebraic varieties.
    The explicit Heisenberg bounds in Theorem 4.12 depend on Theorem 4.11, which is an adaptation of [17] and [23]; the paper supplies an explanation rather than a verbatim quote.
  • standard math The Cayley graph of a free group with respect to a free generating set is a tree, and the reduced word for g^{-1}h gives the unique path metric.
    Foundation for Section 5: d_i counts letter occurrences along the unique reduced path, and connected subsets of a tree have a unique closest point.

how reviews work

0 comments
Cite this review

Pith. "Pith review of VC-dimension of generalized progressions in some nonabelian groups." pith.science (2026). https://pith.science/paper/AHJWNI33

@misc{pith2026250521789,
  author       = {Pith},
  title        = {Pith review of: VC-dimension of generalized progressions in some nonabelian groups},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AHJWNI33}},
  note         = {Machine review of arXiv:2505.21789}
}
abstract

We analyze generalized progressions in some nonabelian groups using a measure of complexity called VC-dimension, which was originally introduced in statistical learning theory by Vapnik and Chervonenkis. Here by a "generalized progression" in a group $G$, we mean a finite subset of $G$ built from a fixed set of generators in analogy to a (multidimensional) arithmetic progression of integers. These sets play an important role in additive combinatorics and, in particular, the study of approximate groups. Our two main results establish finite upper bounds on the VC-dimension of certain set systems of generalized progressions in finitely generated free groups and also the Heisenberg group over $\mathbb{Z}$.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

34 extracted references · 30 canonical work pages

  1. [17]

    Karpinski and A

    M. Karpinski and A. Macintyre, Polynomial bounds for VC dimension of sigmoidal and general Pfaffian neural networks , vol. 54, 1997, 1st Annual Dagstuhl Seminar on Neural Computing (1994), pp. 169–176

  2. [1]

    N. Alon, S. Dar, M. Parnas, and D. Ron, Testing of clustering , SIAM J. Discrete Math. 16 (2003), no. 3, 393–417

  3. [2]

    N. Alon, E. Fischer, M. Krivelevich, and M. Szegedy, Efficient testing of large graphs , Combinatorica 20 (2000), no. 4, 451–476

  4. [3]

    N. Alon, E. Fischer, and I. Newman, Efficient testing of bipartite graphs for forbidden induced sub- graphs, SIAM J. Comput. 37 (2007), no. 3, 959–976

  5. [4]

    Assouad, Densit´ e et dimension, Ann

    P. Assouad, Densit´ e et dimension, Ann. Inst. Fourier (Grenoble) 33 (1983), no. 3, 233–282

  6. [5]

    Blumer, A

    A. Blumer, A. Ehrenfeucht, D. Haussler, and M. K. Warmuth, Learnability and the Vapnik- Chervonenkis dimension , J. Assoc. Comput. Mach. 36 (1989), no. 4, 929–965

  7. [6]

    Breuillard, B

    E. Breuillard, B. Green, and T. Tao, The structure of approximate groups , Publ. Math. Inst. Hautes ´Etudes Sci. 116 (2012), 115–221

  8. [7]

    Chatzigeorgiou, Bounds on the Lambert function and their application to the outage analysis of user cooperation, IEEE Communications Letters 17 (2013), 1505–1508

    I. Chatzigeorgiou, Bounds on the Lambert function and their application to the outage analysis of user cooperation, IEEE Communications Letters 17 (2013), 1505–1508

Show all 34 references
  1. [8]

    Goldreich, S

    O. Goldreich, S. Goldwasser, and D. Ron, Property testing and its connection to learning and approx- imation, J. ACM 45 (1998), no. 4, 653–750

  2. [9]

    Conant and A

    G. Conant and A. Pillay, Approximate subgroups with bounded VC-dimension, Math. Ann. 388 (2024), no. 1, 1001–1043

  3. [10]

    Conant, A

    G. Conant, A. Pillay, and C. Terry, Structure and regularity for subsets of groups with finite VC- dimension, J. Eur. Math. Soc. (JEMS) 24 (2022), no. 2, 583–621

  4. [11]

    C. J. J. Despres, The Vapnik-Chervonenkis dimension of cubes in Rd, arXiv:1412.6612, 2014

  5. [12]

    R. M. Dudley, A course on empirical processes, ´Ecole d’´ et´ e de probabilit´ es de Saint-Flour, XII—1982, Lecture Notes in Math., vol. 1097, Springer, Berlin, 1984, pp. 1–142

  6. [13]

    Duret, Les corps faiblement alg´ ebriquement clos non s´ eparablement clos ont la propri´ et´ e d’ind´ ependence, Model theory of algebra and arithmetic (Proc

    J.-L. Duret, Les corps faiblement alg´ ebriquement clos non s´ eparablement clos ont la propri´ et´ e d’ind´ ependence, Model theory of algebra and arithmetic (Proc. Conf., Karpacz, 1979), Lecture Notes in Math., vol. 834, Springer, Berlin-New York, 1980, pp. 136–162

  7. [14]

    Gillibert, T

    P. Gillibert, T. Lachmann, and C. M¨ ullner, The VC-dimension of axis-parallel boxes on the torus , J. Complexity 68 (2022), Paper No. 101600, 14

  8. [15]

    G´ omez G´ omez and P

    A. G´ omez G´ omez and P. L. Kaufmann, On the Vapnik-Chervonenkis dimension of products of intervals in Rd, arXiv:2104.07136, 2021

  9. [16]

    Gurevich and P

    Y. Gurevich and P. H. Schmitt, The theory of ordered abelian groups does not have the independence property, Trans. Amer. Math. Soc. 284 (1984), no. 1, 171–182

  10. [18]

    M. C. Laskowski, Vapnik-Chervonenkis classes of definable sets, J. London Math. Soc. (2) 45 (1992), no. 2, 377–384

  11. [19]

    Lov´ asz and B

    L. Lov´ asz and B. Szegedy, Regularity partitions and the topology of graphons , An irregular mind, Bolyai Soc. Math. Stud., vol. 21, J´ anos Bolyai Math. Soc., Budapest, 2010, pp. 415–446. 18

  12. [20]

    Marker, Model theory , Graduate Texts in Mathematics, vol

    D. Marker, Model theory , Graduate Texts in Mathematics, vol. 217, Springer-Verlag, New York, 2002

  13. [21]

    Matouˇ sek, E

    J. Matouˇ sek, E. Welzl, and L. Wernisch,Discrepancy and approximations for bounded VC-dimension, Combinatorica 13 (1993), no. 4, 455–466

  14. [22]

    Meier, Groups, graphs and trees, London Mathematical Society Student Texts, vol

    J. Meier, Groups, graphs and trees, London Mathematical Society Student Texts, vol. 73, Cambridge University Press, Cambridge, 2008, An introduction to the geometry of infinite groups

  15. [23]

    Milnor, On the Betti numbers of real varieties , Proc

    J. Milnor, On the Betti numbers of real varieties , Proc. Amer. Math. Soc. 15 (1964), 275–280

  16. [24]

    Sela, Diophantine geometry over groups VIII: Stability , Ann

    Z. Sela, Diophantine geometry over groups VIII: Stability , Ann. of Math. (2) 177 (2013), no. 3, 787–868

  17. [25]

    Shelah, Classification theory and the number of nonisomorphic models, second ed., Studies in Logic and the Foundations of Mathematics, vol

    S. Shelah, Classification theory and the number of nonisomorphic models, second ed., Studies in Logic and the Foundations of Mathematics, vol. 92, North-Holland Publishing Co., Amsterdam, 1990

  18. [26]

    Simon, A guide to NIP theories , Lecture Notes in Logic, vol

    P. Simon, A guide to NIP theories , Lecture Notes in Logic, vol. 44, Association for Symbolic Logic, Chicago, IL; Cambridge Scientific Publishers, Cambridge, 2015

  19. [27]

    Sisask, Convolutions of sets with bounded VC-dimension are uniformly continuous, Discrete Anal

    O. Sisask, Convolutions of sets with bounded VC-dimension are uniformly continuous, Discrete Anal. (2021), Paper No. 1, 25

  20. [28]

    Tao, Product set estimates for non-commutative groups, Combinatorica 28 (2008), no

    T. Tao, Product set estimates for non-commutative groups, Combinatorica 28 (2008), no. 5, 547–594

  21. [29]

    Tao, The free nilpotent group , What’s new (blog), December 21, 2009, https://terrytao

    T. Tao, The free nilpotent group , What’s new (blog), December 21, 2009, https://terrytao. wordpress.com/2009/12/21/the-free-nilpotent-group/

  22. [30]

    L. G. Valiant, A theory of the learnable , Communications of the ACM 27 (1984), no. 11, 1134–1142

  23. [31]

    V. N. Vapnik and A. J. ˇCervonenkis, The uniform convergence of frequencies of the appearance of events to their probabilities , Dokl. Akad. Nauk SSSR 181 (1968), 781–783

  24. [32]

    Vidyasagar, Learning and generalization, second ed., Communications and Control Engineering Series, Springer-Verlag London, Ltd., London, 2003, With applications to neural networks

    M. Vidyasagar, Learning and generalization, second ed., Communications and Control Engineering Series, Springer-Verlag London, Ltd., London, 2003, With applications to neural networks

  25. [33]

    R. S. Wenocur and R. M. Dudley, Some special Vapnik-Chervonenkis classes , Discrete Math. 33 (1981), no. 3, 313–318

  26. [34]

    A. J. Wilkie, Model completeness results for expansions of the ordered field of real numbers by restricted Pfaffian functions and the exponential function , J. Amer. Math. Soc. 9 (1996), no. 4, 1051– 1094. 19

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.