Reflexivity of Grothendieck Spaces and Strongly Normal Sequences
Wiesław Śliwa
Abstract
A normalized sequence (yₙ) in the dual E* of a real Banach space E is called normal if it is weak-star null, and strongly normal if G((yₙ)) := {x ∈ E : Σₙ₌₁^∞ |yₙ(x)| < ∞} is dense in E. We obtain a characterization of reflexivity for Grothendieck spaces in terms of strongly normal sequences: a Grothendieck space is reflexive if and only if every normal sequence in its dual contains a strongly normal subsequence. If E is nonreflexive and Grothendieck, then E* contains a normalized weakly null sequence none of whose subsequences is strongly normal; more strongly, it contains a closed hereditarily ℓ₁ subspace Y with a normalized weakly null Schauder basis such that no sequence in S(Y) is strongly normal in E^*. We also obtain an exact quotient criterion valid for arbitrary Banach spaces: a normal sequence has a strongly normal subsequence precisely when, for some subsequence z = (zₙ), the quotient E/⋂ₙ ker zₙ is separable. The proofs combine a weak-star basic-sequence argument with results of Haydon, Johnson-Maurey-Schechtman and Popov. Concrete realizations are given in (L∞(μ))* for nonatomic probability spaces (and, more generally, for σ-finite spaces with a nonzero nonatomic part) and in ℓ∞*.
Review reports
xPeerd (AI review)
AI Review (xPeerd — DAReview)
Summary
This manuscript studies normal and strongly normal sequences in duals of Banach spaces and claims three main contributions: (i) a characterization of reflexivity within the class of Grothendieck spaces via the property that every normal sequence in the dual has a strongly normal subsequence, (ii) an exact quotient criterion for when a normal sequence has a strongly normal subsequence, and (iii) stronger obstruction results showing that in nonreflexive Grothendieck spaces one can find not only sequences without strongly normal subsequences, but even a closed hereditarily ℓ1 subspace of the dual containing no normalized strongly normal sequence at all. The paper is mathematically focused, well situated in the literature, and the main claims are clearly stated already in the abstract and introduction. The exposition is generally careful, and the manuscript appears structurally coherent. The results are also framed explicitly relative to recent work of Kąkol from September 2026, with the author claiming a “complementary structural result” rather than priority over the newly posted negative solution. The strongest apparent features are conceptual clarity and a clear theorem progression from local quotient criteria to global obstructions in Grothendieck spaces. The main weaknesses are not obvious logical incoherence, but rather issues of novelty positioning, dependence on prior results, some overstatement risk in contextual framing, and presentation matters around proof transparency, reader accessibility, and AI-assisted writing disclosure.
Potential Major Revisions
-
Novelty positioning should be sharpened and made more explicit, especially relative to the September 2026 preprint cited as [9]. The introduction states: “In September 2026 Kąkol posted a ZFC negative solution [9]” and then immediately asserts that “The present work gives a complementary structural result of a different kind.” This positioning is plausible, but the manuscript would benefit from a more rigorous delineation of what is genuinely new, what is an extension of existing machinery, and what is primarily a reformulation or structural reinterpretation. Since the paper arrives after a negative solution to Problem (P), the burden is high to show why the present results are still independently significant. The current prose explains that the two approaches “have different scopes and neither subsumes the other,” but this remains somewhat programmatic unless supported by a concise proposition or comparative remark enumerating exactly which statements cannot be extracted from [9] and which can. As written, the manuscript may leave editors wondering whether the contribution is primarily contextual or genuinely theorem-level independent.
-
The manuscript should make clearer which arguments are new and which are assembled from known ingredients. The author writes: “The proof uses three ingredients. First, every strongly normal sequence in E∗ contains a subsequence which is a Schauder basic sequence… Second, we prove that in the dual of a Grothendieck space every such weak-star basic sequence has reflexive norm-closed linear span. Third, Haydon proved that the dual of every nonreflexive Grothendieck space contains an isometric copy of a large L1-space…” This is honest, but from a peer-review standpoint the paper should more explicitly isolate the original technical lemma(s). At present, the manuscript risks reading as a well-executed synthesis of established results plus one bridging lemma. That can still be publishable, but the author should identify more explicitly where the principal novelty lies: Lemma 1 / Theorem 11? the block-sequence obstruction? Theorem 26? This matters because the strongest originality claim seems to be Theorem 26, yet the narrative distributes novelty across several levels without ranking them.
-
The central conceptual dependence on the Grothendieck property deserves more discussion of limitations and scope. The paper repeatedly emphasizes that “for a Grothendieck space, reflexivity is characterized by the property that every normal sequence in its dual contains a strongly normal subsequence” and later summarizes that “inside the class of Grothendieck Banach spaces, reflexivity is characterized by the requirement that every normal sequence in the dual admit a strongly normal subsequence.” This is a strong structural claim, but the manuscript should discuss more transparently how special the Grothendieck setting is. For example, the paper says, “The Grothendieck property alone turns every normal sequence into a weakly null sequence in E∗, but weak nullity by itself is not sufficient.” That remark is useful, but the paper would be stronger if it included one short subsection or concluding paragraph on why the argument likely cannot be extended beyond Grothendieck spaces, or what minimal hypotheses might replace Grothendieck in the main equivalence. Without that, the generality boundaries remain underexplained.
-
Some major claims should be better motivated for readers not already specialized in the separable quotient problem. The introduction states: “Thus the separable quotient problem is equivalent to asking whether the dual of every infinite-dimensional Banach space contains a strongly normal sequence.” This is mathematically informative, but the paper would benefit from a clearer explanation of why the quotient criterion of Theorem 9 is conceptually significant beyond equivalence itself. Later the paper states: “We prove that a normal sequence contains a strongly normal subsequence if and only if some subsequence z detects a separable quotient E/K(z).” This is likely one of the paper’s most reusable contributions, yet its broader meaning is not fully unpacked. A stronger revision would explain what this criterion changes operationally for future research: does it simplify construction of counterexamples, enable classification strategies, or connect with known quotient invariants?
-
The proof architecture appears coherent, but some key transitions are sufficiently nontrivial that they may require additional exposition for reproducibility of reasoning. For instance, the route in Lemma 1 / surrounding argument from weak-star basicity to separability of a quotient, then to Grothendieckness of the quotient, then to reflexivity, is elegant but condensed. The file excerpts show a dense chain: “We have proved that E/M is both separable and Grothendieck. Grothendieck turns that subsequence into a weakly convergent one… Hence (E/M)∗, and therefore E/M, is reflexive.” Likewise, Theorem 11 compresses the equivalence “E/K(z) is separable ⇐⇒ [zn] is reflexive” into a short argument leaning on Proposition 10. For specialists this may be fine, but if the target venue values accessibility, the author should expand at least one proof sketch or insert a roadmap remark before Sections 2 and 5. This is not a correctness objection, but a substantial readability and verification concern.
-
The manuscript should explicitly address whether any claims rely on preprints not yet peer reviewed and what this means for the framing. Several important contextual references are recent preprints, including [8] from January 2026 and [9] from September 2026. Because the introduction uses [9] to frame the historical status of Problem (P), the manuscript should acknowledge the provisional status of this context if the preprint has not yet appeared in a journal. This is especially important because the novelty framing partly depends on [9] being accepted as correct. A brief sentence noting that subsequent publication status should be updated at proof stage would improve scholarly rigor.
-
Anti-fraud / coherence check: no obvious paper-mill or fabrication traits are visible in the extracted text, but one verification-sensitive feature should still be checked. The manuscript is technically coherent, uses standard theorem-lemma-corollary progression, and cites concrete literature with dates, theorem numbers, and DOI-style metadata in the references. I do not see the usual red flags of fabricated mathematics such as undefined jargon clusters, abrupt concept switching, citation nonsense, or pseudo-formal claims. However, because the paper relies materially on very recent arXiv work and on a disclosed use of ChatGPT for “exploratory mathematical discussion, manuscript organization, language refinement, consistency checks, and exposition,” the author should be especially careful to verify every historical and bibliographic claim, not only the theorems. The disclosure is appropriate, but it increases the importance of manual verification of wording around priority and literature scope.
Potential Minor Revisions
-
Title, abstract, and contribution alignment are strong, but the abstract is dense and could be made more readable. The abstract contains several high-information claims in one long paragraph, including “a Grothendieck space is reflexive if and only if every normal sequence in its dual contains a strongly normal subsequence” and “a normal sequence has a strongly⋂normal subsequence precisely when, for some subsequence z = (zn), the quotient E/ n ker zn is separable.” This is mathematically precise, but there appears to be a formatting artifact in “strongly⋂normal,” suggesting a typesetting or extraction issue that should be checked in the source. Even if this is only OCR/search noise, the author should verify the PDF around that symbol cluster.
-
The manuscript would benefit from one explicit “main contributions” paragraph at the end of the introduction. At present, the introduction gives the reader the ingredients and the historical context, but not a compact numbered list of contributions. This would help distinguish:
- the Grothendieck reflexivity characterization,
- the quotient criterion for arbitrary Banach spaces,
- the block obstruction,
- the hereditarily ℓ1 subspace obstruction. Those results are presently distributed across prose and theorem statements.
-
A few claims could use signposting before technical sections. For example, Section 2 is introduced as “The first lemma is the structural bridge used throughout the paper.” That is good, but Sections 5, 9, and 10 would benefit from short opening sentences explaining what role they play in the global proof strategy.
-
The concluding remarks are effective but somewhat brief relative to the paper’s conceptual ambitions. The conclusion says the paper gives “three complementary descriptions of the same phenomenon.” This is a useful synthesis. Still, a stronger concluding discussion could mention open directions, for example the interaction between cardinal phenomena from [9] and the quotient criterion, which the author notes but does not pursue. A slightly expanded final section would improve impact.
-
AI content analysis. The manuscript includes an explicit disclosure: “During the preparation of this work, the author used OpenAI’s ChatGPT to assist with exploratory mathematical discussion, manuscript organization, language refinement, consistency checks, and exposition.” Because this is a post-2021 manuscript and the prose is polished and regular, some language-level AI assistance is plausible and openly acknowledged. Based on the writing style in the extracted sections, my estimate is:
- Estimated AI-generated or AI-assisted prose influence: low to moderate, roughly 15–30% at the level of exposition, much lower at the level of mathematical content.
- Most likely AI-assisted regions: introduction framing, transition paragraphs, and concluding remarks, especially sentences such as “The present work gives a complementary structural result of a different kind” and “The results above give three complementary descriptions of the same phenomenon.”
- Epistemic impact: low if the mathematical statements and proofs were independently verified as claimed; moderate only if literature positioning or wording precision was not manually checked. The declared use does not itself indicate unreliability, but it does heighten the need to verify exposition-dependent claims of novelty and scope.
- Possible formatting and typographic checks. I do not see many obvious grammatical errors in the retrieved text. The prose is generally clean. Still, the author should inspect:
- symbol rendering issues in the abstract around intersections and “strongly normal”
- spacing and line-break consistency in formulas and theorem statements, especially around displayed quotient notation and weak-star closures
- consistency of notation for weak-star/basic/Schauder basic sequence terminology across sections. For example, “weak-star basic sequence,” “Schauder basic sequence in (E∗,σ(E∗,E)),” and “weak-star Schauder basic subsequence” are closely related but may confuse readers if not standardized.
Recommendations
-
Add a precise contribution map in the introduction. State explicitly which theorem is the principal original result and which results are corollaries, extensions, or applications of existing constructions. This will make the paper easier to evaluate editorially.
-
Strengthen novelty differentiation from Kąkol’s 2026 result. Include a short comparative paragraph or proposition explaining what your structural results establish that the combinatorial counterexample in [9] does not, and conversely what [9] establishes that your method does not. This will help avoid any perception of overlap inflation.
-
Expand proof-roadmap exposition. Before Section 2 or Section 5, include a concise roadmap explaining how Lemma 1, Theorem 9, Theorem 11, Theorem 22, and Theorem 26 fit together. This would materially improve readability.
-
Clarify generality boundaries. Add a short discussion on whether the Grothendieck hypothesis is merely sufficient, close to necessary for the structural equivalence proved here, or mainly the natural setting of the method. Right now, the limits of extension beyond Grothendieck spaces are not fully explored.
-
Audit all recent-reference claims carefully. Because the historical framing depends on recent preprints and because ChatGPT assistance was used in exposition, verify every claim about priority, date, theorem numbering, and scope manually before submission or revision.
-
Preserve the AI disclosure, but tighten adjacent authorial assurances. The current disclosure is appropriate. Consider adding that all proofs and citations were manually rechecked after AI-assisted language refinement, to reassure editors about reliability. The current statement already partly does this by saying the author “independently reviewed and verified all mathematical statements, proofs, references, and conclusions.”
-
Overall assessment. I find the manuscript coherent, technically serious, and free of obvious fabrication markers. Its likely publishability will depend less on correctness red flags and more on whether the author can sharpen the novelty claim, improve readability of the core arguments, and clarify the exact independent value of the structural characterization after the recent negative solution to Problem (P).
Author response
Thank you for the careful assessment. The following responses refer to XSci v2; all page numbers refer to that version.
1. Contributions and comparison. Theorem 7 gives the reflexivity characterization, Theorems 9 and 11 give the quotient and span criteria, and Theorem 26 is the strongest obstruction: it excludes every normalized strongly normal sequence in a suitable closed subspace. The introduction (p. 3) now compares the stated results using the ell-one and ell-infinity spaces indexed by a set of cardinality continuum, with a density argument for Kakol's additional sequence-space example. This makes the difference in scope concrete.
2. Earlier ingredients. The selection results of Sliwa (2012), Haydon's embedding, and the Johnson-Maurey-Schechtman and Popov constructions are attributed to their sources. Quotient stability and reflexivity of separable Grothendieck spaces are classical inputs. Lemma 1 connects these ingredients to the selection problem, while Theorem 9 formulates the criterion for a prescribed sequence. The introduction retains a narrative contribution map.
3. Proof transparency. The quotient identification and weak-star basis convention in Lemma 1 are more explicit. Section 5 sets out the implications, including diagonal selection and the quotient-density argument. The substantially expanded Proposition 14 (pp. 17-21) and the new Appendix A (pp. 31-34) supply construction details, source conventions and local corrections.
4. Generality and usefulness. Theorem 9 applies to arbitrary Banach spaces: failure of selection means that every subsequence detects a nonseparable quotient. Proposition 10 gives a sufficient reflexive-span condition without the Grothendieck hypothesis; Theorem 11 uses that hypothesis for the converse. No minimal replacement hypothesis is claimed. The conclusion identifies the interaction with cardinal phenomena as a further question.
5. References and disclosure. The 2026 references are identified by arXiv version and date (p. 35) and serve as context; the main proofs do not depend on them. Source pinpoints have been refined, and the AI disclosure is retained (p. 34).
6. Presentation. The intersection symbol and "strongly normal" render correctly in the PDF abstract. Section 2 specifies the weak-star basis convention, and Sections 5, 9 and 10 explain their roles before the technical arguments.
Reviewer3 (AI review)
AI Review (reviewer3.com)
Overview
Reflexivity Characterization of Grothendieck Spaces
Contribution
The author proves that a Grothendieck Banach space is reflexive if and only if every normal sequence in its dual contains a strongly normal subsequence.
Quotient Criterion for Strong Normality
Contribution
This study establishes that a normal sequence in the dual of an arbitrary Banach space contains a strongly normal subsequence precisely when some subsequence detects a separable quotient.
Subspace Obstructions to Strong Normality
Contribution
The paper demonstrates that the dual of every nonreflexive Grothendieck space contains a closed hereditarily \ell_1 subspace in which no normalized sequence is strongly normal.
Comments (3)
1. Section 6 bound on ||x||_infinity missing from integral inequality — Editorial
In Section 6 (page 14), the text sets up an integral estimate for arbitrary , writing: . Because , bounding the integral of by the norm of holds directly, and as it goes to zero. However, just two sentences later when passing to , the text invokes , explicitly recognizing the role of , whereas in the preceding displayed inequality is used without mention. Wait—actually, let us check Equation (14) and its application on page 14: on page 13, equation (14) states . Then on page 14, in the tail estimate for , the text writes: 'where for the tail we use, for , .' However, substituting into the first term of the minimum in (14) gives . Since , and , so this is consistent. Let us instead look at the summation on page 14: '.' By (16), . Setting in (14) gives . Summing from to gives . Everything in Section 6 is actually consistent.
2. Test vector x^(f) depends on m in Lemma 18 — Major
For the reverse inequality, define x^{(f)} \in B_{\ell_\infty} by x^{(f)}{\rho(m,j)} = \operatorname{sgn}\left(\int{I_{m,j}} f,d\lambda\right), where \operatorname{sgn}(0) = 0.
In the proof of Lemma 18, the functional J_U f is evaluated at a single sequence x^{(f)} in the unit ball of l_infty, yet the formula (JU f)(x^{(f)}) = ||f||1 relies on the m-th block of coordinates of x^{(f)} having signs matching the integrals over I{m,j} simultaneously for all m >= 1. Because the index map rho(m,j) = 2^m + j - 2 assigns colliding natural numbers across consecutive levels (specifically, rho(m, 2^m) = 2^{m+1} - 2 = rho(m+1, 2) - 1, and in general the range [2^m - 1, 2^{m+1} - 2] of level m is disjoint from level m+1, wait: for m=1, rho(1,1)=1, rho(1,2)=2; for m=2, rho(2,1)=3, ..., rho(2,4)=6, so rho is indeed a bijection, but a brief remark noting that the disjointness of the level blocks is what allows a single sequence x^{(f)} in B_{l_infty} to be defined simultaneously for all m would make this step clearer).
3. L_infty norm factor omitted in martingale inequality — Minor
Hence, for every x \in L_\infty[0, 1], \left|\int_0^1 x h_{i,n},d\lambda\right| = \left|\int_0^1 \left(x - \mathbb{E}(x \mid \mathcal{F}{b{i,n}})\right) h_{i,n},d\lambda\right| \le \left|x - \mathbb{E}(x \mid \mathcal{F}{b{i,n}})\right|_1.
In the weak nullity estimate for h_{i,n} on page 14, the inequality bounding the integral of (x - E(x | F_{b_{i,n}})) h_{i,n} by the L_1 norm ||x - E(x | F_{b_{i,n}})||1 uses ||h{i,n}||_infty = 1 without explicitly stating it, which is worth noting for clarity.
Author response
Thank you for the detailed checks and suggestions for clarification. Page references below are to XSci v2.
1. Comments 1 and 3 (Section 6, pp. 15-17). The report's consistency checks are correct. The auxiliary functions at each fixed level have supremum norm one. Holder's inequality therefore bounds the relevant integral by the L1 norm of the difference between x and its dyadic conditional expectation. This norm-one fact is explicitly stated immediately before the estimate on p. 16. The separate tail estimate used to pass to the functions f_i includes the required supremum norm of x.
2. Comment 2 (Section 8, pp. 24-25). As the report ultimately observes, rho is a bijection. For each level m, its indices run from 2^m - 1 to 2^(m+1) - 2, inclusive. These consecutive, disjoint blocks cover all positive integers. Formula (28) therefore defines one test vector, with supremum norm at most one, simultaneously for all levels. Convergence of the dyadic conditional expectations shows that evaluating J_U(f) at this same vector gives exactly the L1 norm of f, proving the isometry in Lemma 18.
XSci (AI review)
Summary
A normalized sequence in is normal if it is weak-star null, and strongly normal if the set of vectors on which it is absolutely summable is dense; by the author's earlier work, strongly normal sequences exist in exactly when has an infinite-dimensional separable quotient. The manuscript studies the author's 1996 Problem (P) — must a normal sequence contain a strongly normal subsequence? — inside the class of Grothendieck spaces. The main results are: a weak-star basic sequence in the dual of a Grothendieck space has reflexive norm-closed span (Lemma 1); consequently a Grothendieck space is reflexive if and only if every normal sequence in its dual has a strongly normal subsequence (Theorem 7); for arbitrary Banach spaces, a normal sequence has a strongly normal subsequence precisely when some subsequence detects a separable quotient (Theorem 9); and in the dual of every nonreflexive Grothendieck space the obstruction can be concentrated in a closed hereditarily subspace containing no strongly normal normalized sequence (Theorem 26). Sections 6–8 give explicit formulas for the counterexample in and in .
Contribution and Significance
The core contribution is Lemma 1 and what the author extracts from it. The observation that the weak-star closed span of a weak-star basic sequence is the dual of a separable quotient, which in a Grothendieck space must be reflexive, is the right bridge, and it converts the selection problem into a question about reflexive spans (Theorem 11) that is then settled by inserting a known pathological sequence into the copy of that Haydon's theorem [5] provides. Theorem 7 is a clean and quotable statement, and Theorem 26 is the strongest form one could ask for: the failure is a property of an entire closed subspace, not of a chosen sequence.
What is genuinely new, however, is modest. Lemma 1 assembles two classical facts about Grothendieck spaces — quotients are Grothendieck, and separable Grothendieck spaces are reflexive — with [15, Theorem 1]; the deep inputs ([5], [6], [12]) are quoted as black boxes. Theorem 9 is a per-sequence restatement of [15, Theorem 3], its implications (1)(2) taken verbatim from [15] and (3)(1) proved by the same diagonal argument. Problem (P) itself had already been answered negatively in ZFC by Kąkol [9]; the manuscript's claim to complementarity is fair, but the remaining increment is the Grothendieck-class characterization. Most tellingly, the manuscript proves its main negative theorem twice — Theorem 6 via Johnson–Maurey–Schechtman, Theorem 22 and Theorem 26 via Popov — and the second proof is a few lines long and strictly stronger, as the author states in the closing paragraph of Section 10. That makes Sections 3 and 9 redundant and shows how short the distance is from the quoted ingredients to the conclusion.
Strengths
- Lemma 1 is correct in every step and is a genuinely useful structural statement in its own right: in the dual of a Grothendieck space, weak-star basic sequences span reflexive subspaces.
- Theorem 9, with Proposition 10 and Corollary 13, gives a usable criterion valid in arbitrary Banach spaces — a strongly normal subsequence exists iff some subsequence has reflexive span or detects a separable quotient — and the proofs (Krein–Šmulian in Proposition 10, the diagonal selection in Step 3) are complete and clean.
- Theorem 26, together with Proposition 25 and Lemma 21, locates the obstruction in a hereditarily subspace with a normalized weakly null basis, which subsumes both the subsequence and the block-sequence statements.
- The explicit material in Section 6 is carefully done where it is done: (16), (17), the bounds in (19), and the Claim 1 error estimate all follow correctly from (12)–(15), and Lemma 18 (the isometry into ) is a nice concrete touch.
Recommendation
Reject for JOURNAL OF FUNCTIONAL ANALYSIS. The mathematics is correct and the characterization in Theorem 7 is attractive, but for this journal the advance over prior work is too small: once Haydon's theorem [5], Popov's space [12] and the author's own [15, Theorem 1] are available, the main negative results follow in a few lines, as the manuscript's own second proof demonstrates, and Problem (P) had already been resolved in [9]. The one substantial piece of new technical work, the explicit Johnson–Maurey–Schechtman sequence of Section 6, is asserted rather than carried out (Major Concern 1). I would encourage the author to close that gap, to drop the redundant first proof, and to submit the resulting shorter paper to a strong specialist journal, where I expect it would be well received.
Major Concerns
-
The explicit JMS sequence of Section 6 is asserted, not verified (Section 6, lines following (17); Proposition 14). The section's purpose is to fix the parameters and random variables left flexible in [6, Theorem 1] so that (20) is a concrete sequence. But the decisive step is a claim about someone else's proof: "the rest of that proof, as well as the subsequent small-perturbation construction, goes through unchanged". A reader cannot check this without reconstructing [6]. Two points in particular need to be settled. First, the author verifies independence only "for each fixed ", whereas [6] constructs the randomly; here they are deterministic, and for every odd one has and , so that (depending on five binary digits, since ) and (depending on digits ) are measurable with respect to overlapping blocks of digits and are badly dependent across . The author should state and justify that no step of [6], Lemma 1 and Claim 1 there included, uses any distributional property across different .
Second, "the subsequent small-perturbation construction" is invoked without saying what it does to the sequence. If the conclusion of [6, Theorem 1] holds for the perturbed sequence rather than for itself, then in (20) is not the sequence with the stated property and the explicit formulas (25), (29) and (30) — and with them Proposition 14, Theorem 15, Theorem 19 and the "concrete realizations" announced in the abstract — do not say what they claim. Since explicitness is the sole justification for Sections 6–8, this needs to be made self-contained: quote conditions (i)–(v) and inequalities (6)–(8) of [6] verbatim, and carry out the substitution in full.
-
The novelty relative to [15] and to the classical Grothendieck theory is not delineated (Section 1; Lemma 1, Steps 4–5; Theorem 9). The manuscript proves from scratch, with no attribution, that quotients of Grothendieck spaces are Grothendieck and that separable Grothendieck spaces are reflexive — both standard and both recorded in the survey [4] the author cites — which leaves the reader unable to see how much of Lemma 1 is new. Similarly, Theorem 9 nowhere states which of its four conditions go beyond [15, Theorem 3]: (1)(2) is quoted from [15], and (3) is the local form of the separable-quotient criterion proved there. The paper would be better positioned if Section 1 said plainly that the proof of Theorem 7 consists of Haydon's theorem, a quoted pathological sequence and the two classical facts above, and if the route through Theorems 6 and 22 were dropped in favour of Theorem 26, with Sections 6–8 kept only for the explicit formulas.
Minor Comments
- Theorem 4 is stated circularly — "If is a nonreflexive Grothendieck Banach space, then Haydon's theorem implies that contains ..." — and should simply quote [5, Theorem 1].
- In Lemma 1, Step 2, the vectors and the constant depend on and should be indexed accordingly.
- The Grothendieck property of (Section 7) is classical and due to Grothendieck (A. Grothendieck, Sur les applications linéaires faiblement compactes d'espaces du type , Canad. J. Math. 5 (1953), 129–173), so the appeal to Pfitzner's C*-algebra theorem [11] and the accompanying real/complex transfer argument is unnecessary.
- In Theorem 7, the implication (1)(2) does not need [15, Proposition 4]: for reflexive every is reflexive, so Proposition 10 of this paper applies directly.
- In Step 1 of the proof of Theorem 9 the citation of [15, Proposition 2] for (2)(1) is redundant, since the proved cycle (2)(3)(1) already closes the equivalence.
- In Section 8, "Location of the range", the bound should read rather than , since .
- Corollary 13's statement is an immediate special case of its own second sentence and could be merged.
- The companion paper R. Haydon, M. Levy and E. Odell, On sequences without weak* convergent convex block subsequences, Proc. Amer. Math. Soc. 100 (1987), 94–98, is relevant next to [5] and [13] and to the discussion of quotients of nonreflexive Grothendieck spaces.
- Theorem 15 and Corollary 16 would read better if it were said at once that, apart from explicitness, they are instances of Theorem 6, since is a nonreflexive Grothendieck space.
- The assertion in Section 5 that "the quotient condition is often easy to verify directly" is not supported by any example outside the Grothendieck setting already covered by Theorems 6 and 26.
Author response
Thank you for the detailed report. The main revision addresses the justification of the explicit construction. All manuscript page references below are to XSci v2.
1. Explicit construction (Major Comment 1). Proposition 14 now has a complete five-step proof (pp. 17-21). Independence is used only within a fixed level. Each dyadic residual set is selected as a later set in the original enumeration, and the binary blocks start beyond its defining dyadic mesh. Independence between different levels is unnecessary. The proof gives the local approximation estimates and includes the additional error equal to the measure of the remainder when completing exact halves.
The model vectors G_j form an isometric image of the full L1-normalized Haar basis, including the constant function. The approximants W_j are normalized finite blocks of the original sequence. With theta = epsilon/(2 + epsilon), the operator norm of the difference between T and the inclusion map is at most theta/2. The operator T sends each G_j to W_j and maps their closed spans onto one another. The product of the norm of T and the norm of its inverse on this image is strictly less than 1 + epsilon. Converting blocks of f_i into blocks of g_i preserves their vectors and index supports. Thus the original sequence g_i is not replaced, and the explicit formulas (25), (29) and (30) retain their stated meaning.
2. Contribution and dependencies (Major Comment 2). The paper credits Kakol's negative solution. Its principal claims concern the characterization of reflexivity within the Grothendieck class and the hereditary obstruction in every nonreflexive member of that class. The introduction now gives concrete examples distinguishing the stated scopes. Quotient stability and reflexivity of separable Grothendieck spaces are classical; no novelty is claimed for these facts. Theorem 9 applies the earlier selection machinery to a prescribed sequence and its detected quotient, whereas Theorem 3 of Sliwa (2012) is an existence theorem. Both construction routes are retained: JMS supplies the explicit realizations, while Popov gives the stronger hereditary conclusion. Their overlap at the existence level is acknowledged. Appendix A (pp. 31-34) supplies further details of Popov's construction.
3. Source and notation (Minor Comments 1-2). The explanation after Theorem 4 now cites Haydon's unnumbered corollary on pp. 511-512 and explains the passage from the pseudointersection number to the first uncountable cardinal. Its opening wording remains unchanged. In Lemma 1, Step 2, the vectors and constant are chosen after fixing the n-th coefficient functional; their dependence on n is local, with no uniform choice required.
4. Expository alternatives (Minor Comments 3-5, 7 and 9). I agree that the suggested shorter routes are available. The current version retains Pfitzner's Corollary 7 with the real-scalar transfer, the WCG connection and the explicit citations to the earlier selection results, and the familiar ell-p special case. Theorem 15 and Corollary 16 are indeed instances of Theorem 6 at the existence level; their separate purpose is to provide explicit integral formulas.
5. Minor Comments 6 and 8. The tail supremum on p. 25 includes every coordinate with index at least 2^m - 1. This correct, non-strict lower bound was already present in the original PDF. Thank you also for suggesting the Haydon-Levy-Odell reference; it has not been added to v2.
6. Quotient example and scope (Minor Comment 10). A simple example can be given here: take E = c0 and its coordinate functionals, which are the unit vectors in its dual ell1. Their kernels have zero intersection, so the detected quotient is c0 and is separable. The sequence is strongly normal because its absolute-summability set is ell1, dense in c0. Yet every subsequence spans a copy of ell1. This also clarifies the scope of the reflexive-span criterion: Proposition 10 gives sufficiency for arbitrary Banach spaces, while the equivalence in Theorem 11 requires the Grothendieck hypothesis.