gpt-6.2-sol high: cef-configurations
Reviewer record, not the publication verdict. This preserves a completed model response, including unaccepted proposals and its stated source limitations. The curated report and repair note specify which conclusions were adopted. Seven initial source-label failures were reconciled in explicit derivatives; native statuses and original access declarations remain in the provenance record.
Completed visible assessment, preserved as a readable check record. It is a model assessment, not a formal certificate. This document contains no private reasoning stream.
Within the assigned scope, the main rational character and colored-configuration ranges survive. The arbitrary-space version of the power proposition, the reverse filtration implication in Lemma 6.3.2, and the unqualified negative injectivity bounds are false as printed. Section 6.4 has a genuine gap in the general-ring proof: reducing a spectral-sequence subquotient modulo a maximal ideal does not justify the dimension inequality. The replacement coefficient argument avoids that step and retains the finite-generation conclusion. The all-n integral generator assertion in Theorem 1.9(iii) is a false theorem claim, with a compact-core configuration-space counterexample. This audit uses the recorded 2013-07-14 received-date cutoff. Publication years and versions above are bibliographic evidence, not claims of inspection; a citation whose exact contents or hypothesis match could not be checked is treated conditionally. No post-cutoff result is used to close an obligation. Later knowledge cannot be excluded from model training; the new arguments are separated from the printed proofs and do not rely on a later FI-module theorem. The journal metadata record is not a comparison with the journal full text.
C1
Checked statement: Sections 6.1–6.2: functoriality of powers and ordered configurations, the hypotheses of the finite-type assertion for powers, and the geometric, coefficient, and naturality inputs to the rational Leray argument.
Returned assessment: claim_false
The verdict concerns the literal arbitrary-space proposition \ref{pr:cohoMbullet}, not the rational configuration-space theorem \ref{theorem:coho:config2}. For an injection f:S→T, restriction sends a T-configuration to an S-configuration, preserves distinctness, and composes contravariantly; restriction on powers gives the commuting square with the inclusions. Cohomology reverses variance again. The formula in \ref{eq:MbulletFIsharp} also composes: a coordinate not surviving both partial injections is assigned the basepoint. For a nonempty path-connected space with degreewise finite field cohomology, the power argument works. Over a field, the corresponding homology is degreewise finite, so the cohomological cross product is an isomorphism for finite powers. A degree-p tensor has at most p positive-degree coordinates; the other coordinates are units supplied by the FI structure maps. Koszul signs under relabeling do not affect generation. In particular the argument works for the manifolds in \ref{theorem:coho:config2}. In the Leray argument for configurations, small product neighborhoods of a point in M^S can be chosen as products of disjoint coordinate balls, one ball for each block of coincident labels. Their intersections with the configuration locus are products of configurations in balls and are path-connected for d≥2. This gives the required constant R^0, not a general property of Leray. The positive-degree generators suffice: the (0,0) term is the unit, so it need not be included in the generating submodule in \ref{thm:fisubalgebra}. Once the stated E2 algebra description is established naturally under forgetting labels, the finite-type argument is sound: each bidegree of E∞ is a subquotient of a finitely generated rational FI-module and the abutment filtration has only i+1 possible graded pieces in total degree i. Finite generation passes up a short exact sequence by lifting generators of the quotient and adjoining generators of the submodule.
Premises used
- Internal: \ref{thm:fisubalgebra} requires a positive-grading generating sub-FI-module; \ref{th:noetherian} applies degreewise over the Noetherian ring Q. The latter theorem is used as an internal prerequisite, not independently refereed here.
- To: available_from=1996 (year precision); date_evidence=CEF bibliography, Topology 35, 1057–1067; date_basis=journal publication; version=1996 journal article. The input attributed to it is that for the inclusion of ordered configurations in the powers of an oriented d-manifold, d≥2, the rational E2 algebra is generated by the base cohomology and the pair linking classes of bidegree (0,d−1), naturally for coordinate projections. The article is not supplied; its precise manifold and change-of-coefficients statements are not verified from that citation.
- Classical foundations used below: field Künneth for singular homology of products, field universal coefficients, integral homology Künneth for free singular chain complexes, and cohomology universal coefficients for free integral singular chains. available_from=2002 (year precision); date_evidence=bibliographic record of Hatcher, Algebraic Topology, Cambridge University Press; date_basis=book publication; version=2002 edition. Leray and its edge/filtration construction use the classical first-quadrant derived-functor spectral sequence for sections of a direct image and the comparison of constant-sheaf with singular cohomology on paracompact locally contractible manifolds; dated foundational reference: Weibel, An Introduction to Homological Algebra, 1994 edition, as listed at We. These are bibliographic identifications, not inspections of the external texts.
Missing inputs
- The text of To establishing exactly which manifolds and coefficient rings are covered, and whether the needed naturality is stated there. The local argument below gives a separate route to the input; it does not verify what To says.
Scope changes
- The path-connectedness repair changes the general topological-space proposition, but imposes no additional restriction on a connected manifold.
- For the local Leray calculation, the manifold is taken without boundary. For a manifold with collar, pushing the boundary inward gives the corresponding homotopy equivalence on each configuration space by an isotopy through embeddings, compatible with forgetting labels.
Located concern
\ref{pr:cohoMbullet}, §6.1, lines 1834–1879; proof sentence ‘The assumption that M is connected guarantees that H^0(M;k)≃k’
Connected is not enough for the proposition about arbitrary topological spaces with singular cohomology.
Take the closed topologist’s sine curve, the graph of sin(1/x) for 0<x≤1 together with the vertical limit segment. It is connected as the closure of a connected graph. It has two path components: a path joining the segment to the graph would have first coordinate tending continuously to 0 from positive values and hence traverse arbitrarily small coordinates where the second coordinate is alternately 1 and −1, contradicting continuity at departure from the segment. Both components are contractible. Every singular simplex lies in one component, so H^0(M;k)=k² and H^j(M;k)=0 for j>0. The power M^n has 2^n path components. Thus H^0(M^n;k) has dimension 2^n, whereas a finite set of FI generators in degrees m_j spans at most Σ_j(n){m_j} dimensions at n: there are only (n){m_j} injections out of degree m_j.
Require nonempty path-connectedness, or directly assume H^0(M;k)=k along with the stated degreewise finiteness. The support argument in the audit then proves the proposition.
The arbitrary-space proposition is false as printed. Its use for a connected manifold is unaffected.
Located concern
Proof of \ref{theorem:coho:config2}, §6.2, lines 1889–1903; first two proof paragraphs
The general identification of the Leray row E2^{p,0} with H^p(Y;Q) is false, and the displayed square in the next paragraph has the inclusion indices reversed.
For the map from two points to one point, R^0f_*Q=Q², so even E2^{0,0} differs from H^0(Y;Q). In the configuration case an injection T→S induces Conf_S→Conf_T and M^S→M^T, so the equality is i_T∘Conf(f)=M^bullet(f)∘i_S. The neighborhoods described in the audit give R^0(i_S)_*Q=Q: any two ordered configurations in a ball of dimension at least two can be joined by first choosing intermediate distinct positions, then moving one point at a time along paths avoiding the finitely many other points.
State the general row as H^p(Y;R^0f_*Q), verify the constant sheaf in the application, and use the correctly typed square. The map of Leray sequences then goes from T to S, as required for an FI-module.
A false general assertion and a mistyped identity in the proof, not counterexamples to the configuration theorem.
Located concern
Proof of \ref{theorem:coho:config2}, §6.2, lines 1889–1903; compare the opening of the proof of \ref{th:config} and the field paragraph of \ref{thm:ConfMFIsharpfg}
The rational generator description alone does not verify the uniform support decomposition used in §6.3 or an integral and arbitrary-field version used in §6.4. In particular one cannot deduce these just by changing coefficients in a rational isomorphism.
The FI maps and the existence of integral link classes matter in those applications. A sequence of termwise decompositions as symmetric-group representations would not, by itself, identify the maps T in \ref{def:Phia}. The pair class is naturally labeled and changes by the degree (−1)^d of the antipodal map when the pair is reversed, giving exactly the two S₂ actions stated in §6.2.
New local argument for the needed input. In a coordinate ball, forgetting the last point is locally trivial: move each point of a small variation of the base configuration by a chart homeomorphism supported in its own disjoint smaller ball. The fiber is a ball punctured at the other labels. Excision in the pair consisting of a ball and the complement of the punctures gives one degree-(d−1) linking class per puncture and no other positive-degree cohomology; the classes lift by the pair direction maps. The bundle spectral sequence therefore gives, inductively, a free basis of products of pair classes, with each larger endpoint used at most once (fix a total order of the labels). Indeed the global lifts give permanent generators for the fiber row, the base classes are permanent, and products give lifts of the resulting associated-graded basis. This works over Z and by the same argument over any coefficient ring. The basis monomials are forests. Their spans split by the partition into connected components, independently of the order: a relabeled forest can be reduced to the displayed basis preserving the partition. If edges to a common larger endpoint repeat, pull back the three-label relation expressing that product as a linear combination of the other two trees on those labels. That relation follows from the two-element basis in degree 2(d−1) on three labels; each replacement reduces the sum of larger endpoints and preserves the connected components. Now for a partition π let Δ_π be the closed partial diagonal imposing its equalities, and L_π the free group of its forest classes in degree (|S|−|π|)(d−1). The sheaf R^{q(d−1)}(i_S)_*k splits as the sum of the constant modules L_π⊗k pushed forward from Δ_π for |S|−|π|=q. To check this, use the local product of disjoint coordinate balls: only forests with edges inside the equal-coordinate blocks occur. The local classes glue as sheaf germs since on a linking sphere the change of an oriented chart has degree +1; products of the pair germs give the partition summands. Away from an edge diagonal its linking germ is zero. The indicated map of sheaves is an isomorphism by the forest basis on each stalk. Under a projection of powers the class is pulled back with the extra labels as singleton blocks. Taking H^p gives the direct sum of H^p(Δ_π;k)⊗L_π, using that closed pushforward is exact and computes this cohomology on Δ_π. This is a separate derivation, not a claim about the text of To or Ch.
It supplies a geometric route to the E2 support and coefficient inputs used in C2 and C3, without importing a later configuration-space theorem.
C2
Checked statement: Lemma 6.3.2, Theorem 6.3.1, the n≥2i character and dimension range, and the n≥i+|μ| colored configuration-space range.
Returned assessment: claim_false
Lemma 6.3.2(i) is correct. After applying exact Φ_a, a cycle whose translate is a boundary can be corrected by a lift from X when T on X is onto; injectivity on Y then shows the corrected cycle is zero. Conversely, lift a target cycle in Y when T on Y is onto; injectivity on Z shows its lift is a cycle. These arguments give exactly max(B,C) and max(D,E). For a filtration only the direction from graded pieces to the filtered object is needed, and that direction follows from the diagram of short exact Φ_a sequences: injectivity is preserved by extensions, as is surjectivity. With that direction and nonnegative injectivity bounds the induction works. Write D=d−1 and use A(p,q)=max(0,p+2q+1−D). For the first differential the incoming term, when present, has surjectivity bound p+2q+1−D, and otherwise has bound 0. The outgoing term has injectivity degree 0. At page kD+1 the incoming term, if present, has surjectivity bound p+2q+3−2D≤p+2q+1−D; if absent it has bound 0. Thus the corrected (a_k) propagates. For q>0 and k>q the outgoing target is zero. For 2≤k≤q its injectivity bound is p+2q+(k−1)(D−2), already nonnegative, so the printed (b_k) is unchanged. For fixed (p,q) there are finitely many possible incoming or outgoing pages, uniformly in n. On the limit page the q>0 surjectivity bound is p+Dq+2−D≤i, and the q=0 term is a quotient of the bottom row, with surjectivity bound p=i. Weight is at most p+2q≤i. This proves the corrected injectivity bound and the unchanged stability degree ≤i, conditional on the E2 input and the internal weight computation. For the E2 support decomposition one can use C1's local calculation: mark the union of nonsingleton blocks, of size at most 2q, together with the singleton coordinates carrying positive homological degree in Künneth. There are at most p of the latter. The projection maps and insertion of a fixed basepoint split off the complementary singleton coordinates. Over Q take duals of the homological decompositions, which have only finitely many support summands for each fixed S even if a summand is infinite-dimensional. The pieces are transported by bijections of their supports, giving the required direct sum of M(W_j), not merely an objectwise decomposition. For the character conclusion, finite-dimensional cohomology in the application is needed; the finite-type theorem supplies it under the preceding introduction's compactness/finite-cohomology context. For the colored result, let a=|μ|. The finite free relabeling action with deck group S_{n−a}×S_μ identifies rational quotient cohomology with its invariant subspace by transfer (equivalently, average lifts of singular simplices). Averaging also identifies S_{n−a}-invariants with coinvariants equivariantly for S_a. Apply stability at Φ_a degree n−a≥i and then take S_μ-invariants. This proves the asserted vector-space isomorphism, even without arguing from dimensions. The last line's reference to S_|μ|-invariants should read S_μ-invariants.
Premises used
- Internal: exactness of \ref{def:Phia} for rings containing Q; the quotient-surjectivity direction of \ref{lemma:injsurjdescend}; the computation \ref{eq:PhiaMW} for \ref{pr:stabdegprojective}. The latter is an orbit decomposition of subsets, and each existing summand maps identically to the same summand one degree later.
- Internal, conditional outside the assigned scope: \ref{pr:Mmuweight} bounds the weight of M(W_j) by j over a characteristic-zero field; \ref{thm:persinomial} gives the uniform character-polynomial range n≥stability degree+weight for finitely generated FI-modules over such a field.
- Ch: available_from=2012 (year precision); date_evidence=CEF bibliography, Invent. Math. 188, 465–504; date_basis=journal publication; version=2012 journal article, §3.3. The input invoked by CEF is a natural decomposition of E2^{p,q(d−1)} for rational cohomology of ordered configurations on a connected oriented d-manifold into free-on-FB summands supported in sizes ≤p+2q. The external section is unsupplied. The FI naturality or any coefficient extension should not be inferred from the bibliography; the audit gives a separate support argument.
- Mac: available_from=1995 (year precision); date_evidence=CEF bibliography; date_basis=book publication; version=Symmetric Functions and Hall Polynomials, second edition, Example I.7.14, as invoked by \ref{thm:persinomial}. Its precise input is the characteristic-zero character polynomial of the padded irreducible V(λ)_n in the range n≥|λ|+λ₁. The book is not inspected, and the internal polynomiality theorem is used conditionally.
Missing inputs
- External text of Ch §3.3 for the printed citation, and external texts used by the internal character-polynomial theorem if that theorem is to be audited independently. The finite-dimensional character corollary should explicitly retain the finite-cohomology context.
Scope changes
- The corrected injectivity bound is weaker only where the printed bound is negative. It has the same operational assertion there if a negative range is understood as testing all nonnegative indices.
- No claim is made that the numerical stable ranges are optimal for each manifold.
Located concern
\ref{lemma:homologyfacts}(ii), line 1928
The ‘only if’ direction is false, even for a two-step filtration over Q.
Let V=M(0) and let F⊂V be 0 on the empty set and Q on every nonempty finite set, with the inherited maps. V has injectivity and surjectivity degrees 0. At a=0 the map in Φ_0(F) at n=0 is 0→Q, not onto; the map in Φ_0(V/F) at n=0 is Q→0, not injective. Thus neither graded piece satisfies the pair (0,0).
Replace ‘if and only if’ by ‘if’. For any short exact sequence, a diagram chase after exact Φ_a gives the two extension implications separately. Iterate along the filtration.
The lemma as stated is false, but the abutment argument uses only the true direction.
Located concern
\ref{th:config}, lines 1918–1921, and its proof, lines 1938–1981
The injectivity bound can be negative under a definition requiring the degree to be nonnegative. The same problem occurs in (a_k) and in the base case's treatment of out-of-quadrant terms.
Take i=0 and M=R³. Its ordered configurations are nonempty and connected, so H^0=M(0) and inj-deg=0. The theorem asserts inj-deg≤−1. Even the zero FI-module cannot have negative injectivity or surjectivity degree under \ref{def:injectivitydegree}.
Use max(0,i+2−d) in the theorem, max(0,p+2q+1−D) in (a_k), and give missing terms degree 0. The corrected induction is checked above. Alternatively, explicitly define a negative bound as shorthand for the required maps being injective at every nonnegative index, rather than an inequality on the degree.
The literal quantitative theorem is false for low i. The stability degree ≤i, weight ≤i, and both claimed ranges survive the repair.
Located concern
\ref{thm:confncharpoly}, introduction immediately preceding lines 430–440; deduction in §6.3 just after the proof of \ref{th:config}
The Betti polynomial need not have degree exactly i. Also the introductory ‘If dim M≥3’ must be read in its retained finite-dimensional context for traces to be defined.
For M=S⁴, under the preceding compactness hypothesis, total degree 3 on the E2 page consists only of the pair row, but the compact example is not needed to settle exactness: take the interior of the 4-ball, where the base is contractible and degree 3 has dimension binom(n,2), not degree 3. For a compact counterexample without any differential computation, take M=S⁴ and i=1: the E2 page has no terms of total degree 1, so the Betti number is identically zero. If finite cohomology is dropped altogether, M=(R² minus a closed discrete infinite set)×R^{d−2} is a connected oriented d-manifold with infinite-dimensional H^1, as the separate puncture loops detect; already H^1(Conf_1(M);Q) is not a finite-dimensional representation. For d≥3 the degree-one bottom row also survives for every n.
Write ‘degree at most i’ and retain the preceding compactness/finite rational cohomology assumption for this character statement. No extra finiteness is necessary for the corrected weight and stability-degree inequalities themselves.
No failure of the n≥2i range for the stated finite-type application, or of the n≥i+|μ| range in \ref{co:vakilwood}.
C3
Checked statement: Proposition 6.4.2 and Theorem 6.4.3: the homotopy FI-sharp extension of the forgetful maps and finite generation in degrees ≤i or ≤2i over the stated coefficient rings.
Returned assessment: repairable_gap
The extension and finite-generation conclusions can be retained, but the printed composition argument needs a collision-free homotopy, and the residue-field step is invalid. New geometric repair: write the collar coordinate as x∈(0,c), with x=0 at the missing boundary, and let h_u, 0≤u≤e<c, compress the collar by x↦u+(1−u/c)x, leaving the complement fixed. Take Φ=h_e and put new points in 0<x<e. In a composite, the surviving original points have undergone zero, one or two compressions, according to which operations added points. If there are two compressions, all surviving added points are constant and lie in the collar 0<x<h_e(e): first-stage constants that survive the second restriction lie in e<x<h_e(e), and second-stage constants lie in 0<x<e. Move this whole labeled constant configuration inside the larger collar to the chosen constants in 0<x<e, holding the original points fixed outside the larger collar. The configuration space of that connected collar is path-connected. Then vary h_e∘h_u from u=e to 0, keeping the constants in 0<x<e. With one compression all constants are already in the smaller collar and can be moved there; with no compression there are no constants. If the composite adds no surviving labels, isotope away any compressions with no constants to avoid. Surviving original labels are exactly ψ(B∩C), with the indicated composite relabeling. This proves functoriality in the homotopy category and agrees with forgetful maps on co-FI. Passing to cohomology and using inversion of partial injections gives FI-sharp with the intended covariant FI restriction. For finite generation over a field, the E2 input in C1 gives a dimension bound without requiring Leray to be a spectral sequence of FI-sharp modules. For fixed (p,q) in bidegree (p,q(d−1)), its supports have size at most p+2q. For a given support there are boundedly many partition types and basis classes, and the cohomology on the active coordinates in fixed degree is bounded uniformly over all fields. To see uniformity, only products with boundedly many factors contribute; integral homology of M is finitely generated, so field universal coefficients bound dimensions uniformly by the sizes of integral generating sets. Taking field subquotients and the finite abutment filtration gives dim_F H^i≤C_i(1+n)^B, where B=i for d≥3 and B=2i for d=2. There is no tensoring of subquotients here. New algebraic repair for a general Noetherian k: for each n let A_j=H_j(Conf_n(M);Z), a finitely generated abelian group. Write r_j for its free rank and t_{p,j} for the number of cyclic p-primary summands. The integral-chain cohomology universal coefficient sequence with coefficients in F_p gives t_{p,i−1}, t_{p,i}≤dim_{F_p}H^i(Conf_n(M);F_p), and rational cohomology bounds r_i. In degree zero set A_{−1}=0. Let I be the ideal of k consisting of elements killed by some nonzero integer. Since I is finitely generated over k, one nonzero integer N annihilates I. In Hom_Z(A_i,k), each free summand contributes one copy of k, and each p^a-cyclic summand contributes k[p^a]. The latter vanishes for p∤N; for p|N it equals k[p^{min(a,v_p(N))}] by Bézout. These form a fixed finite list of finitely generated ideals, independent of n. Thus Hom_Z(A_i,k) has O((1+n)^B) generators. In Ext¹_Z(A_{i−1},k) use invariant factors of the finite torsion subgroup: there are max_p t_{p,i−1} such factors, each contributing a cyclic k-module k/mk, so Ext has the same bound. The universal coefficient sequence is k-linear; lift generators of Hom and adjoin generators of Ext. This proves the desired bound for H^i with k coefficients, including k=Z, without using finite Krull dimension. Finally, write H^i≅M(W) by FI-sharp classification. For any j, W_j is a direct summand at every n≥j, hence is finitely generated by the growth bound at sufficiently large n. If W_j≠0, a maximal ideal m with W_j/mW_j≠0 exists by finite generation and Nakayama. The fiber then has at least binom(n,j) dimensions from that summand alone, so j≤B. This proves the claimed finite generation in the specified FI degrees, not just qualitative finite generation.
Premises used
- For the homotopy construction the collar is the one already assumed in the printed proof; path-connectedness of configurations in the collar follows by moving finitely many points one at a time along paths avoiding the others, since the collar is connected and has dimension ≥2.
- Integral finiteness used in the new argument: the interior of a compact manifold, or that interior with finitely many further points deleted, has finite domination. Delete small disjoint coordinate balls around the punctures and use collars or radial retractions to obtain a compact core. A compact smooth core has finite CW type; for a topological core the standard ENR route is finite domination: embed the compact manifold in Euclidean space using finitely many charts and bump functions, take a neighborhood retraction, and a finite cubical complex inside that neighborhood containing the core. The compact core is a retract of that finite complex. Classical ENR input: a locally compact locally contractible finite-dimensional subset of Euclidean space that is locally closed is a neighborhood retract. available_from=1967 (year precision); date_evidence=bibliographic record of Borsuk, Theory of Retracts; date_basis=book publication; version=1967 edition, not externally inspected.
- To propagate finiteness from the core to Conf_n, the forget-one map is a locally trivial bundle with fiber M minus finitely many points by the local disjoint-chart isotopies in C1. Induct on n for homology with any locally constant coefficients having finitely generated abelian fibers. The base case follows from finite domination (pull the coefficient system back to the finite complex); on the bundle spectral sequence the homology of the fiber with restricted coefficients is finitely generated by the same argument. Its groups form a local system on the base, to which induction applies. The first-quadrant finite filtration gives finiteness in each degree of the total space. Classical input: homology Serre spectral sequence of a locally trivial bundle over a path-connected base of CW type, with local coefficients on the total space and E² given by base homology with the corresponding fiber-homology systems. available_from=2001 (year precision); date_evidence=bibliographic record of McCleary, A User’s Guide to Spectral Sequences; date_basis=book publication; version=second edition, 2001, not externally inspected. The homology and cohomology universal coefficient inputs have the dated bibliographic record given in C1.
- Forster: available_from=1964 (year precision); date_evidence=CEF bibliography, Math. Z. 84, 80–87; date_basis=journal publication; version=1964 journal article, Satz 1 as cited. The relevant generator bound is for a finitely generated module over a Noetherian ring of finite Krull dimension δ, with all maximal-ideal fibers of dimension ≤N: at most N+δ generators. The article is unsupplied; the repair does not use this result.
- Internal: \ref{th:charFIsharp} and \ref{eq:MWdecomp} give the support decomposition over any commutative k. In the growth-to-generation argument of \ref{thm:polyFIsharp}, the finite generation of W_j can be checked before choosing a detecting maximal ideal, by the large-n direct summand argument above. This removes any need for detection of arbitrary non-finitely-generated modules by residue fields.
Missing inputs
- The printed proof needs a justification of the field and integral E2 bound beyond the rational computation cited from Ch. The separate local calculation and coefficient argument above identify the needed replacement; they do not establish an integral claim about Ch.
- The exact text of Forster's Satz 1 is not supplied. Even the usual finitely generated form of the result does not justify the residue-field step preceding its application.
Scope changes
- The replacement proof uses Noetherianity but not finite Krull dimension. This is a property of the new argument, not an extension asserted by the printed proof.
- The FI-sharp conclusion concerns homotopy classes of maps, not the existence of a strict topological FI-sharp functor with the displayed representatives.
Located concern
\ref{pr:CMhTop}, lines 2028–2039, especially line 2038
The composite may add two groups of constants in different collar layers. The text identifies the direct composite's constants with the wrong q, and an arbitrary isotopy of Φ does not by itself keep the moving points disjoint from those constants.
When both operations add points and all the first additions survive the intermediate restriction, the first group ends up in Φ(R), while the second lies in R. The direct composite places all additions in R. The block of constants in the iterated map must therefore be moved, not just the original coordinates re-isotoped. If an operation adds no point, its effect on original coordinates is the identity rather than Φ.
Use the two nested collar layers and the explicit ordered homotopy in the audit; use q indexed by ψ(B∩C)⊂U for the direct composite.
Repairs the homotopy functor proof without changing its hypotheses.
Located concern
\ref{thm:ConfMFIsharpfg}, lines 2050–2055
The phrase ‘from the proof of Theorem \ref{th:config}’ does not establish that the E2 computation holds over arbitrary fields or Z. Nor does the homotopy FI-sharp structure on configurations make their Leray sequence an FI-sharp sequence.
The calculation cited there is rational, whereas the field growth argument uses actual k-dimensions and the integer argument needs actual generator bounds. Computing the pages over Q cannot bound mod-p dimensions in the presence of torsion. The manuscript correctly notices the homotopy-invariance problem for the Leray sequence; a separate bound on the page entries is enough.
Use the ring-coefficient local calculation in C1, at least over fields, and count supports as in C2. Apply only the objectwise field dimension inequality and the abutment filtration; use FI-sharp only on the abutment.
Supplies the missing coefficient justification and retains both exponents.
Located concern
\ref{thm:ConfMFIsharpfg}, lines 2057–2061
Tensoring with a residue field does not preserve subobjects or subquotients in the sense required by the printed dimension comparison. The stated version of the Forster input also omits finite generation.
For the same class of coefficient rings as in the theorem, take k=Q[x,y], m=(x,y), and the submodule m^r⊂k. Its residue fiber at m has dimension r+1 by the degree-r monomials, while the fiber of the ambient module has dimension 1; the induced map on fibers is zero for r>0. Thus ‘since F is a field’ does not give the asserted inequality for fibers of E∞ subquotients. Also Q as a Z-module has zero maximal-ideal fibers and is not finitely generated, so the unqualified generator bound as restated in the proof is false.
Use the new universal-coefficient and bounded-integer-torsion argument in the audit, followed by FI-sharp support detection. This does not tensor any spectral sequence subquotient with a residue field.
A false intermediate step in the general-ring proof, not a counterexample to Theorem 6.4.3.
C4
Checked statement: Theorem 1.9 items (i)–(iv): rational and mod-p Betti numbers, the minimum number of integral generators, and the minimum number of generators of the p-torsion part, including degree bounds and equality for all n≥0.
Returned assessment: claim_false
With the connected oriented d≥2 hypotheses of §6.4, put B=i for d≥3 and B=2i for d=2. By C3 and classification the degree-i cohomology with each of the three coefficient choices is a direct sum over 0≤j≤B of binom(n,j) copies of a fixed finitely generated coefficient module W_j. For (i) and (ii), take the corresponding field dimensions; the resulting polynomials hold at every n≥0, of degree at most B. For (iv), fix p and use W_j for the integral coefficients. The number of generators of its p-primary torsion equals the number of cyclic p-power summands, which is also the dimension of the subgroup killed by p. This number is additive on direct sums, so (iv) has the same all-n conclusion and degree bound, whether ‘p-torsion part’ means the primary subgroup or the subgroup killed by p. These invariants are different: in cohomological notation the universal coefficient formula is b_i(F_p)=b_i(Q)+t_p(H^i(Z))+t_p(H^{i+1}(Z)), not just rational rank plus the torsion in degree i. For (iii), an abelian group's minimum number of generators is its free rank plus the maximum, over primes, of the numbers of cyclic primary summands. The maximum is necessary by reduction modulo p and sufficient by combining cyclic summands of relatively prime order. Only finitely many primes occur in the finitely many W_j. Consequently (iii) is the free-rank polynomial plus the maximum of finitely many torsion polynomials, with 0 included for the torsion-free case. It is eventually polynomial of degree at most B since the sign of a difference of two polynomials is eventually constant. It need not be polynomial for all n.
Premises used
- Internal: \ref{th:charFIsharp} and \ref{eq:MWdecomp} over Z, Q and F_p, with finite generation in the bounded FI degrees of \ref{thm:ConfMFIsharpfg}, established by the replacement argument in C3.
- For the counterexample below: integral Künneth and cohomology universal coefficients, with dated bibliographic record given in C1. The closed oriented 3-manifolds RP³ and L(3,1) have simply connected S³ covers with cyclic deck groups of order 2 and 3 respectively, hence H_1=Z/2 and Z/3. Their H_2 is zero by integral Poincaré duality and H^1=Hom(H_1,Z)=0, and H_3=Z by orientation. Poincaré duality in this use is H_2≅H^1 for a closed connected oriented 3-manifold; available_from=2002 (year precision) for the cited exposition, date_evidence=Hatcher, Algebraic Topology, Cambridge University Press bibliographic record, date_basis=book publication, version=2002 edition, not externally inspected.
Missing inputs
Scope changes
- Items (i), (ii) and (iv): retain all n≥0 in the §6.4 setting, with upper degree bounds. Item (iii): retain the upper degree bound only for the eventual polynomial; the maximum formula holds for every n≥0.
- The counterexample disproves the general statement of (iii). The audit does not infer that a switch of dominant prime occurs for orientable surfaces.
Located concern
\ref{thm:intro:openbetti}(iii), lines 430–440; last paragraph of §6.4, line 2064
The minimum number of integral generators is not additive on the FI-sharp direct-sum decomposition. The all-n conclusion for (iii) fails for actual configuration spaces, not just for abstract FI-sharp modules.
Let M be the interior of the boundary connected sum of two copies of RP³×D³ and one copy of L(3,1)×S¹×D². This is a connected oriented compact-core 6-manifold with nonempty boundary, and the core is homotopy equivalent to the wedge of the three closed factors RP³, RP³ and L(3,1)×S¹: the connecting handles join distinct components along disks, which are contractible. Passing to the interior preserves this homotopy type by the collar. Thus H_1(M;Z)=Z⊕(Z/2)²⊕Z/3 and H_2(M;Z)=Z/3. The degree-3 comparison H^3(M^n;Z)→H^3(Conf_n(M);Z) is an isomorphism for every n: local configurations have no positive cohomology below degree 5 by the ball calculation in C1, so the Leray bottom row is the only contribution and no differential can hit it in degree 3. Integral Künneth in homological degree 2 of M^n gives one H_2(M) per factor and one H_1(M)⊗H_1(M) per unordered pair of factors; Tor terms in this degree vanish because H_0 is free. Per pair there are eight 2-cyclic summands (two from each free/torsion cross term and four from the 2/2 product), and three 3-cyclic summands (one from each free/torsion cross term and one from the 3/3 product). The torsion of H^3 therefore has counts t₂=8 binom(n,2) and t₃=n+3 binom(n,2), by universal coefficients. The rational rank is 3n+binom(n,3): M has rational Betti numbers 1,1,0,3 in degrees 0 to 3. At n=1 the maximum torsion count is t₃=1 and t₂=0; for every n≥2 it is t₂ since 5 binom(n,2)>n. A polynomial agreeing with the minimum number of generators for all n≥2 must be 3n+binom(n,3)+8 binom(n,2); its value at 1 is 3 rather than the actual 4. In particular the obstruction is present even with i>0 and with all the geometric hypotheses in §6.4.
State eventual polynomiality for (iii) and, for an all-n formula, give the rank polynomial plus the maximum of the finitely many p-primary count polynomials. An all-n polynomial follows under an extra condition, for example that a single prime realizes the maximum for every n.
Theorem 1.9(iii) is false as stated. Items (i), (ii), and (iv) do not use this nonadditive invariant.
Located concern
\ref{thm:intro:openbetti}, introductory parenthetical in lines 430–440
The degrees are upper bounds, not exact degrees.
Take M=R⁴, the interior of D⁴, and i=3. Its base powers are contractible and the first positive Leray row is the free pair row in degree 3. It follows over Z and every field that H^3 is free of rank binom(n,2) in the respective coefficients. Items (i) and (ii), and the integral generator count in this example, thus have degree 2 rather than 3; the p-torsion count is identically zero. The bound also cannot be read as exact for (iv) in dimension 2: H^0 has no p-torsion, for instance for the disk.
Use ‘degree at most i’ or ‘degree at most 2i’ wherever polynomiality is asserted.
The support bound yields the corrected degree statements for the all-n polynomials and for the eventual polynomial in (iii).
Located concern
\ref{thm:intro:openbetti}, hypothesis line in 430–440, compared with \ref{thm:ConfMFIsharpfg}, lines 2044–2048
The introduction does not repeat the connectedness and dimension restrictions of the theorem used to deduce it.
If read independently to allow disconnected compact cores, take two disjoint d-balls, d≥2. Each assignment of the n labels to the two components gives one path component of the ordered configuration space, so b₀=2^n over any field and the integral generator count in degree zero is 2^n. If dimension 1 is also allowed, the interior of an interval has n! configuration components.
Explicitly carry forward the connected oriented d≥2 setting of §6.4. The main counterexample to (iii) still applies in that setting.
On the contextual reading the examples outside that setting are not objections; on a literal unrestricted reading, even the all-n Betti polynomial assertions are false.
Outside the audit
- The journal full text and any differences from the supplied arXiv v4 version.
- The complete proofs in the external references, including the precise coverage of To and the formulation and functoriality in Ch §3.3, and the external proof of Forster's theorem.
- The auxiliary Dold–Thom, infinite-loop-space and algebraic K-theory identifications in \ref{remark:FIspaces}; the partial-injection construction itself was checked.
- Whole-paper theorems outside the assigned scope, including the general Noetherian and character-polynomial theorems insofar as they supply conditional internal prerequisites.
- Preserved source-access limitation: To
- Preserved source-access limitation: Ch
- Preserved source-access limitation: Forster
- Preserved source-access limitation: FH
- Preserved source-access limitation: Mac
- Preserved source-access limitation: Duke Math. J. full text, DOI 10.1215/00127094-3120274
Additional supporting archive items are identified in the original report. View the manuscript hashes and repository snapshot.