Algebra, representation theory and topology

FI-modules and stability for representations of symmetric groups

Thomas Church, Jordan S. Ellenberg and Benson Farb

54 pages114 claims indexed66 referencesManuscript 2014-10-30

gpt-6.2-sol high: cef-verify-repairs

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.

Both proposed coefficient repairs withstand the audit with the explicit topology, finiteness, and naturality checks given here. They repair the general-ring proof without finite Krull dimension, not the false all-n assertion about the total number of integral generators. The connected-sum example proves that failure and shows why the replacement must be a maximum. In the rational proof, removing the false filtration converse and clipping injectivity degrees at zero preserves the range calculations under the stated rational inputs. These are present-day repairs, not claims about historical priority or about the contents of the journal version.

V1

Checked statement: CEX2: the exact all-n formula and eventual polynomial for the minimum number of integral generators, and both universal-coefficient repairs of the configuration-space FI-sharp generation theorem.

Returned assessment: repairable_gap

The residue-field step in the original proof fails. Both proposed replacements work once the coefficient-independent topological inputs and the naturality of the coefficient sequences are made explicit. Write D_i=i for d≥3 and D_i=2i for d=2. The repaired theorem holds over every commutative Noetherian ring with unit, without a Krull-dimension restriction. The support bound alone holds over arbitrary commutative rings with unit.

FI-sharp structure. Use a connected collar component with interior coordinate 0<t<1. For 0≤u<1, let h_u send t to u+(1−u)t on that collar and be the identity off it. First restrict and relabel along the defined part of a partial injection. If adding labels, apply h_u for a fixed u>0 and place the new labels at distinct constants with t<u; if not, take u=0. For two successive maps with parameters a,b, the surviving original labels are moved by h_b h_a=h_v, where v=b+(1−b)a. The surviving first-stage constants have depths between b and v, and the second-stage constants have depths below b. All constants in the composite therefore lie in the region vacated by h_v. If the direct composite adds labels, change v to its chosen positive parameter, rescaling the depths of the constants continuously, then move their ordered configuration to the chosen constants within the vacated collar. That configuration space is path connected: a connected manifold of dimension at least two remains path connected after removing finitely many points, and temporary positions allow the labels to be moved one by one. If no constants survive, just isotope h_v to the identity. This proves composition in the homotopy category, including the case in which points were added and later all forgotten. Inverting partial bijections converts contravariant cohomology into covariant FI-sharp cohomology; on total injections the resulting maps are pullbacks along forgetting. Homology is already covariant. No strict composition on spaces is needed.

Support and exactness. The supplied classification gives V_S=⊕{T⊆S} W_T, with a partial injection relabeling the summands in its domain and killing the others. One can see the exactness needed here directly. For t∈S set e_t=V(I{S−{t}}). These are commuting idempotents, and the mutually orthogonal projectors ∏{t∈T}(1−e_t)∏{t∉T}e_t sum to the identity. Inclusion and partial retraction identify the T-projector image with the full-support projector image on V_T, which is W_T: the other images on V_T come from proper subsets. These natural projectors split every pointwise short exact sequence into exact sequences of coefficient groups. Thus support vanishing above D is preserved by extensions and subquotients in FI-sharp. If V_n has a generator bound O(n^D), each W_r is finitely generated because it is a summand of V_n for every n≥r, so one can take n large. Any nonzero finitely generated W_r has a nonzero residue-field quotient, and the binom(n,r) copies force r≤D. Over a field, this lower bound is simply a dimension bound. Conversely, finitely generated W_r supported in r≤D give FI generators in those degrees, by the subset formula. Neither test tensors an arbitrary submodule inclusion into a residue field.

Field growth and integral finiteness. These can be established without the omitted Leray input. Put X_m=Conf_m(M). Forgetting the last label gives a locally trivial bundle X_{m+1}→X_m with fiber M−P for an ordered m-point set P. Local trivializations move the points in disjoint coordinate balls by supported ambient isotopies. Along a loop, their composites give transport extending to a homeomorphism of M isotopic to the identity. Thus transport acts trivially on H^(M;F) and H_(M;Z). The relative groups of (M,M−P) are zero outside degree d by local excision; in degree d they are copies of the coefficient group indexed by the punctures. These systems are constant because the punctures are labeled and the local orientations are preserved. The long exact pair sequence therefore gives a two-step filtration of fiber cohomology: an image from H^b(M;F), of dimension at most b_b(M;F), and a subspace of the degree-(b+1) relative group, of dimension at most m and present only for b=d−1. Both pieces are constant systems; the extension need not be constant.

Let c_{m,i}=dim_F H^i(X_m;F), with negative indices zero and c_{m,0}=1 by path connectedness. The coefficient long exact sequence on the base bounds the dimensions for the two-step system by the sum for the two pieces. The cohomological Serre spectral sequence then gives, for i>0, c_{m+1,i} ≤ c_{m,i}+Σ_{b=1}^i b_b(M;F)c_{m,i−b}+m c_{m,i−(d−1)}. The numbers b_b(M;F) are finite. The compact model is homotopy equivalent to its interior by pushing the boundary inward. In the smooth or PL case the compact model has finite CW type. For topological manifolds it is enough to use the classical compact ENR fact: in a retracting Euclidean neighborhood of the compact model, a finite union of sufficiently small closed cubes containing the model gives a finite CW domination. Either way, its integral homology is finitely generated in each degree. Inducting on i, the displayed recurrence bounds the increment by O((m+1)^(i−1)) for d≥3 and O((m+1)^(2i−1)) for d=2. Summing gives O((n+1)^D_i) for every field separately. There is no need for the constants to be uniform across fields.

For integral homology use the homological pair sequence of the same fiber. In degree b its two pieces are the cokernel of H_{b+1}(M;Z)→H_{b+1}(M,M−P;Z) and the kernel of H_b(M;Z)→H_b(M,M−P;Z). These are constant finitely generated systems. If the base has degreewise finitely generated homology, then it has degreewise finitely generated homology with each constant finitely generated abelian coefficient group by homological UCT. The long exact sequence of local coefficients and the homological Serre spectral sequence give degreewise finite generation for the total space. Induct from X_0, a point. In both Serre arguments the filtrations used for the estimates are finite in each total degree. The spectral sequences are used only for one n at a time, not as spectral sequences of FI-sharp modules.

Repair A, using homological support. Decompose H_j(X_•;Z) by subsets. Its coefficient groups are finitely generated abelian groups by the preceding paragraph. If the coefficient at r is nonzero, tensor detects it over Q when it has rank or over a suitable F_p when it has torsion. Choose this field once for that r. Homological UCT injects H_j(X_n;Z)⊗F into H_j(X_n;F), so the field homology has dimension at least binom(n,r) for every n≥r. Over a field cohomology is the linear dual of homology, and the finite-dimensional bound just proved forces r≤D_j. Now use the natural sequence of ordinary cohomological UCT over Z: 0 → Ext^1_Z(H_{i−1}(X_•;Z),k) → H^i(X_•;k) → Hom_Z(H_i(X_•;Z),k) → 0. It is k-linear by naturality in the coefficient group, and natural for maps in the homotopy category. On the ends use the contravariance of Ext and Hom in homology together with partial inversion. On a finite subset sum, the reverse of a partial relabeling and projection induces exactly the forward partial relabeling and projection on the new coefficient groups: retained components use the inverse isomorphism in the argument of Hom or Ext, and omitted components have zero maps. Thus the left end has support ≤D_{i−1} (it is zero for i=0) and the right end has support ≤D_i. Free cyclic summands contribute k to Hom and zero to Ext; Z/h contributes k[h]={x∈k:hx=0} to Hom and k/hk to Ext. Over Noetherian k these are finitely generated. Exact coefficient extraction and lifting generators through the sequence prove finite generation in support ≤D_i, without a natural splitting.

Repair B, using cohomological support. The coefficient sequence with Tor also works, but its use for singular cochains must be justified. For a fixed X_n, take a finite free presentation of each finitely generated H_j(X_n;Z), with free relation group. Place this two-term complex in degrees j+1 and j, lift its generators to cycles and its relations to bounding chains, and sum over j. This gives a degreewise finite free complex mapping by a homology isomorphism to the free singular chain complex. Its mapping cone is a free acyclic complex over Z; its boundary groups are free, so the short exact sequences onto the boundary groups split and give a contraction. The comparison therefore computes cohomology with any abelian coefficients. In particular the canonical map H^j(X_n;Z)⊗P→H^j(X_n;P) is an isomorphism for any free abelian P, even of infinite rank, as can be checked on this finite-in-each-degree complex. The complex need not be chosen functorially: it verifies a canonical comparison map. Applying the coefficient long exact sequence to a free presentation of the underlying abelian group of k now gives naturally 0 → H^i(X_•;Z)⊗k → H^i(X_•;k) → Tor_1^Z(H^{i+1}(X_•;Z),k) → 0. Naturality in spaces and coefficients (or using a functorial free presentation for this step) gives FI-sharp compatibility and k-linearity. Integral cohomology is finitely generated degreewise by ordinary UCT. The left injection, tested with Q or a suitable F_p, and binomial multiplicities prove integral cohomological support ≤D_i independently of Repair A. The field growth bound likewise gives field FI-sharp support ≤D_i. Write W'_r for the coefficient group of integral cohomology in degree i+1. Extract the coefficient at any r>D_i from the sequence with k=F_p. Its middle coefficient is zero, so Tor_1^Z(W'_r,F_p)=W'_r[p]=0 for every p. Thus W'_r is free and contributes no Tor for any k. This checks all excess supports, including both 2i+1 and 2i+2 in dimension two. The ends of the sequence have finitely generated coefficients for Noetherian k, as in Repair A. This proves the same theorem independently of the homological-support argument.

Coefficient-ring scope. Both sequences prove support vanishing for arbitrary commutative k. Finite generation does not follow for all such k. For instance, take the punctured L^7(3), for which H_1=Z/3, and take the square-zero extension k=Z⋉I with I an infinite-dimensional F_3-vector space. Then H^1(M;k)=Hom_Z(Z/3,k)=k[3]=I. The action on I factors through F_3, so this is not finitely generated even at configuration size one. Noetherianity is sufficient; the repairs only need finiteness of the annihilator modules for the relevant torsion orders.

Arithmetic conclusion. Let W_j be the integral cohomology coefficient group in degree i and support size j. Set R(n)=Σ_j rank(W_j) binom(n,j) and T_p(n)=Σ_j dim_Fp(W_j[p]) binom(n,j). These are all-n formulas for the free rank and the minimum generator count of the p-primary subgroup, respectively. Only finitely many primes occur, because there are finitely many nonzero, finitely generated W_j. A finitely generated abelian group of rank r needs at least r+dim_Fp(T[p]) generators for every p, by reducing modulo p, where T is its torsion subgroup. Conversely, list cyclic generators of each primary subgroup in slots and combine generators in the same slot by the Chinese remainder theorem; adjoining r free generators attains the lower bound. Hence the exact answer is R(n)+max_p T_p(n), taking the maximum as zero if there is no torsion. Pairwise differences of these finitely many polynomials have a fixed sign on a sufficiently large tail, so one polynomial realizes the maximum on the whole tail. The degree is at most D_i. Rank agrees with rational Betti number, and each field Betti number separately has an all-n binomial formula from its own FI-sharp decomposition. Mod-p cohomology need not be the reduction of integral cohomology; the Tor quotient explains why.

Premises used

  • The supplied FI-sharp partial-injection definition, additive classification, and configuration-manifold hypotheses. The collar composition and the coefficient arguments are repairs, not deductions from the printed composition sentence or from the printed change-of-scalars step.
  • Classical topology: collaring and compact-manifold finiteness, coordinate-ball isotopies, excision, local-coefficient long exact sequences, and first-quadrant Serre spectral sequences for the forgetful bundles.
  • Classical algebra: UCT over Z, free subgroups of free abelian groups, finite abelian group decomposition, and exact additive operations on finite sums. No Forster estimate or later FI-module result is used.

Missing inputs

  • The earlier configuration-space proof cited at lines 1933 and 2051 is omitted. Church §3.3 is cited in the supplied rational argument, but its text is not supplied. The excerpts do not independently establish the all-characteristic Leray E_2 support/dimension bound at line 2055 or the integral E_2 generator bound needed at line 2057. The new field and integral arguments replace, rather than verify or refute, those coefficient-specific inputs.
  • The text of Forster's cited Satz 1 and the journal pages named in the candidate are not supplied. The repaired proof does not rely on them.

Scope changes

  • Remove the finite Krull-dimension hypothesis, but do not extend finite generation unconditionally to non-Noetherian coefficient rings.
  • Replace all-n polynomiality of the total integral generator minimum by an exact all-n formula and eventual polynomiality.
  • Use degree at most i or 2i, not an asserted exact degree. No change in the allowed manifold dimensions is needed.

Located concern

CEF-repairs-source, original TeX lines 2057–2061

A subquotient over k need not become a subquotient of the ambient residue-field fiber. The proof also states a Forster-type implication without a finite-generation hypothesis.

For k=C[x,y] and m=(x,y), the included ideal m has fiber m/m² of dimension two while k has fiber of dimension one. For the missing finiteness hypothesis, Q/Z over Z has zero fibers at all maximal ideals but is not finitely generated.

Replace that step by either of the natural coefficient sequences above, after proving field growth and integral finiteness independently. Do not identify the Leray sequence over a residue field with the tensor of the sequence over k.

The printed general-ring inference is invalid. The repaired finite-generation theorem holds for every commutative Noetherian k, including rings of infinite Krull dimension.

Located concern

CEF-repairs-source, original TeX lines 2031–2039; candidate-repairs, CEX2 collar construction

An isotopy of the compression alone does not ensure that the moving points avoid the inserted constants. The original composition sentence also uses constants from the wrong target and does not account for cases with no insertions.

A retained first-stage constant can lie in the second compression's image region relative to the single-step parameter. It is not enough to declare both outputs constant on those labels and isotope the remaining coordinates without tracking separation.

Track the composite compression parameter and move the output constants only within its vacated region, as above; when there are no output constants, use the unrestricted compression isotopy.

The homotopy FI-sharp structure is valid, with a composition proof that applies to both homology and cohomology.

Located concern

CEF-repairs-source, original TeX lines 432–437 and 2064; candidate-repairs, CEX2 arithmetic conclusion

The integral total minimum is not additive across subset summands, and none of the growth arguments forces an exact polynomial degree.

The controlling prime can change with n, as V2 shows. Also, H^1 with integral coefficients is torsion free by UCT, so its fixed-prime generator polynomial is zero rather than necessarily of degree one or two.

Use the all-n rank-plus-maximum formula and state eventual polynomiality for the total minimum. Retain all-n polynomials for field Betti numbers and fixed-prime generator counts, with degree upper bounds.

The replacement formula and eventuality are valid; the original all-n total-generator assertion is false.

V2

Checked statement: CEX1's connected-sum counterexample and CEX3's filtration converse, negative-degree correction, page estimates, and stability-range consumers.

Returned assessment: claim_false

The connected sum is a genuine counterexample to the original all-n generator assertion. The filtration converse is also false. Under the usual nonnegative degree convention the negative numerical bound must be clipped, but this does not weaken either stability-range application. These are distinct failures of the uncorrected claims, not failures of the candidate's corrected conclusions.

Manifold and homology. Use N=L^7(3)#UTS^4#UTS^4 and M=N minus a point. The scalar order-three action on S^7 is free and orientation preserving. Its quotient is the Hopf circle bundle over CP^3 with the fiber transition functions cubed, so its Euler class is ±3h. In the Gysin sequence multiplication by this class between successive nonzero even cohomology groups of CP^3 is multiplication by ±3. It gives intermediate cohomology Z/3 in degrees 2,4,6 and zero otherwise. Integral oriented duality gives intermediate homology Z/3 in degrees 1,3,5. The unit tangent sphere bundle of S^4 has oriented fiber S^3, and its Euler number is 2 (for example the two zeros of the height gradient each have index +1). Gysin gives only H^4=Z/2 in intermediate cohomological degrees, hence only H_3=Z/2 in intermediate homology. Both summand types are smooth closed connected oriented 7-manifolds. Removing a ball preserves homology below degree six by excision and the pair sequence; the orientation class maps onto the local relative orientation group. Gluing along S^6 adds the summands' homology in degrees 1 through 5 by Mayer–Vietoris. Puncturing the sum preserves these groups. The punctured manifold is diffeomorphic, by radial reparametrization in a coordinate ball, to the interior of the compact complement of an open ball and has a compact model with boundary S^6. In particular it meets the connectedness, orientability, dimension, and boundary hypotheses. We have H_1(M)=Z/3, H_2(M)=0, H_3(M)=Z/3⊕(Z/2)^2, and rational homology vanishes in degrees 1 through 4, which is all the rational vanishing needed here.

Products and relative comparison. Put Y_m=M^m. The split integral homological Künneth sequence gives H_1(Y_m)=(Z/3)^m and H_2(Y_m)=(Z/3)^binom(m,2): in degree two the new nonzero tensor term is H_1(Y_m)⊗H_1(M), and all Tor terms vanish. In degree three, beyond the old H_3(Y_m), adjoining one factor adds H_3(M), H_2(Y_m)⊗H_1(M), and Tor_1^Z(H_1(Y_m),H_1(M)). The splitting as groups matters because a nonsplit extension of order-three groups need not have the same generator count. The new multiplicities are 2 for Z/2 and 1+binom(m,2)+m for Z/3. Summing gives H_3(Y_n)=(Z/2)^(2n) ⊕ (Z/3)^[n+binom(n,2)+binom(n,3)]. Rational Künneth gives H_4(Y_n;Q)=0, so Hom_Z(H_4(Y_n),Z)=0. Consequently integral UCT identifies H^4(Y_n;Z) with Ext^1_Z(H_3(Y_n),Z), which has the displayed cyclic factors.

The diagonal comparison does not require transversality to a singular union. Enumerate the pair diagonals and delete them in succession. At each stage the next diagonal restricted to the current open ambient manifold is a closed embedded submanifold of codimension seven. Excision to a tubular neighborhood and the relative homology of its normal disk and sphere fibers give zero relative homology in degrees less than seven (the Thom shift gives the same vanishing with a possible orientation system). This applies even though the restricted diagonal need not be compact. The exact sequence of each pair gives an isomorphism on H_j for j≤5, using vanishing in both degrees j and j+1. Equivalently the triples give vanishing for (Y_n,Conf_n(M)) in degrees below seven. For n=0,1 there is nothing to delete. Naturality of UCT now gives the required H^4 isomorphism; only H_3 and H_4 are needed. This supplies a relative comparison independent of the candidate's abbreviated general-position argument.

The group is finite, so its generator count is the maximum of its two primary counts. It is max(2n,(n^3+5n)/6). The cubic exceeds 2n for n≥3, since their difference is n(n^2−7)/6. At n=1,2 the actual counts are 2,4, while the cubic gives 1,3. A polynomial agreeing on the infinite tail is identically that cubic, so cannot give the all-n count. This does not contradict finite support or the all-n formulas for each prime.

Filtrations. Use Φ_a(V)n=(V{n+a})_{S_n}, where a letters are distinguished. Over a ring containing Q, averaging identifies coinvariants with invariants and gives invariant lifts in surjections, hence Φ_a is exact. In a short exact sequence of Φ_a values, injectivity of both outer transition maps implies injectivity in the middle by projecting a kernel element to the quotient and then testing it on the left. Surjectivity of both outer maps implies surjectivity in the middle by lifting first on the right and correcting on the left. This proves graded bounds ⇒ filtered bounds for each threshold separately. For the reverse take V to be constant Q and F to be zero below size N≥1 and equal to V from size N on. This is a sub-FI-module because injections never decrease size. The ambient transitions are identities for every a; for a=0 at n=N−1, F has transition 0→Q and V/F has transition Q→0. Thus the claimed converse fails for both kinds of bound, even with finite generation. The complex lemma in part (i) is valid: for injectivity on homology, lift a preboundary in X using surjectivity and subtract its boundary before applying injectivity on Y. For surjectivity, lift a target cycle in Y using surjectivity and apply injectivity on Z to see that the lift is a cycle. Exactness of Φ_a is what allows these tests to compute the transition on homology.

Rational page bounds, conditional on the stated rational Leray input. Set D=d−1≥2, h=p+2q+1−D, and h_+=max(0,h) for a term in bidegree (p,qD). The rows are spaced by D, so the only possible nonzero differentials have page numbers kD+1. At the first such page a present incoming term has surjectivity bound h; if outside the quadrant it is absent. The middle and any present outgoing term initially have injectivity bound zero. Part (i) gives injectivity at most h_+ and surjectivity at most p+2q after this page. For k≥2 keep the surjectivity bound s_k=p+2q+min(k−2,q−1)(D−2) for q>0. At page kD+1 a present incoming term has surjectivity at most p−kD−1+2(q+k)+(k−2)(D−2)=p+2q+3−2D≤h. Thus the new injectivity bound remains h_+. If 0<q<k the outgoing term is absent and s_k≤s_{k+1} suffices. If q≥k, the outgoing term has injectivity bound p+2q+(k−1)(D−2), already nonnegative, by applying h_+ to that bidegree. The maximum with the middle surjectivity bound is s_{k+1}. These estimates also handle q=k, where the outgoing term reaches the bottom row. In the bottom row q=0 there are no outgoing differentials, and its terminal term is a quotient of E_2^{p,0}, so its surjectivity bound is p.

Each fixed bidegree stabilizes by the first-quadrant bounds. When p+Dq=i the resulting injectivity bound is ≤max(0,i+2−d); the surjectivity bound is ≤i, since for q>0 the stabilized bound is p+2q+(q−1)(D−2)=i+2−D. Weight is ≤p+2q≤i on each graded piece by the rational initial bound and passage to subquotients. The finite filtration passes the injectivity and surjectivity bounds to the abutment by the valid direction of the lemma, and likewise passes the weight bound by characteristic-zero semisimplicity at each size. The negative number in the original literal degree inequality does not describe a least nonnegative degree: already H^0 is the constant Q module, of degree zero, whereas i+2−d<0 at i=0. If the expression is read merely as a lower limit on the allowed nonnegative transition indices, clipping it to zero changes none of those indices.

Publication-safe consequences. Given the stated rational spectral-sequence input and the cited weight-plus-stability-to-character theorem in its finite-dimensional setting, stability degree ≤i and weight ≤i still give the character-polynomial range n≥2i. The Betti polynomial obtained by evaluation at the identity has degree at most i, not necessarily i; for example b_1(Conf_n(S^3))=0 by the same successive-diagonal argument in codimension three. For colored configurations with a=|μ|, stability degree alone gives equivariant stabilization of the S_{n−a} coinvariants for n−a≥i. Averaging identifies these with invariants. The quotient by S_{n−a}×S_μ is the colored configuration space, and rational transfer for the finite covering identifies its cohomology with the invariants of that group. Taking S_μ-invariants therefore gives the claimed vector-space stability for n≥i+a; it is not necessary to take invariants under all of S_a, or to assert a map of colored spaces. Neither application uses the false filtration converse or a strictly negative degree.

Premises used

  • Classical oriented-bundle Gysin sequence, Poincaré duality, excision, Mayer–Vietoris, tubular neighborhoods and the Thom shift, split integral homological Künneth, and integral UCT.
  • For the rational induction: an FI-compatible first-quadrant Leray spectral sequence with the stated rows, initial rational bounds, and finite abutment filtration. These are used as inputs, not inferred from the homotopy FI-sharp structure.
  • For the consumers: characteristic-zero semisimplicity and rational finite-cover transfer; the character range additionally uses the omitted weight-and-stability theorem with its hypotheses. The literal negative-degree objection uses the usual least nonnegative degree convention.

Missing inputs

  • The earlier rational Leray decomposition and the propositions giving the initial stability and weight bounds are not supplied in full. The induction is checked relative to the bounds stated at lines 1935–1936.
  • The original definition referenced at line 1916, the general character-range theorem cited at line 1983, and the introductory character conclusion as printed outside the supplied ranges are not available for direct inspection. The transition-index conclusion and the dependence on the character theorem are stated explicitly.
  • No journal pages or underlying reports named inside the candidate are supplied for version comparison.

Scope changes

  • The connected-sum example and the original arXiv claims are adjudicated directly; no claim is made about which corrections appear in an unprovided journal version.
  • The range audit verifies the page induction and the colored application and specifies the external dependency for the general character range. It does not reprove the omitted rational Leray or character theorem.
  • The second, PL-thickening counterexample is not needed for V2 and is not adjudicated.

Located concern

CEF-repairs-source, original TeX lines 432, 436 and 2064; candidate-repairs, CEX1 connected-sum example

The asserted all-n polynomial for the total minimum is false under the manifold hypotheses.

For the punctured connected sum above, H^4 has primary counts 2n and (n^3+5n)/6. Their maximum differs from the eventual cubic at n=1 and n=2.

State the rank-plus-maximum formula for every n and replace all-n polynomiality of the total minimum by eventual polynomiality.

A theorem item must change; repairing only its proof is not enough. Integral FI-sharp bounded generation and fixed-prime polynomiality remain consistent with the example.

Located concern

CEF-repairs-source, original TeX lines 1927–1928 and 1977

The filtration lemma asserts a false converse; the complex lemma and the forward use in the theorem are valid.

A threshold submodule of the constant FI-module has transition 0→Q and its quotient has Q→0 at the threshold, while the full module has identity transitions.

State only that bounds for all graded pieces imply the same bounds for the filtered module.

The page-to-abutment step needs only that implication and does not lose any range.

Located concern

CEF-repairs-source, original TeX lines 1920, 1938–1952 and 1954–1980

The injectivity-degree inequality omits the maximum with zero. A term outside the quadrant should not be assigned a negative degree.

For i=0 and d=3 the printed bound is −1, but H^0 has least nonnegative degree zero. At the first page the middle injectivity bound is already zero, independently of whether there is an incoming term.

Use h_+ on the pages, omit nonexistent terms, and state the abutment injectivity bound as max(0,i+2−d).

The numerical degree assertion changes in low degrees, but stability degree ≤i, weight ≤i, and the two range consequences are unchanged given their stated inputs.

Outside the audit

  • Unprovided parts of the original TeX, the journal pages and underlying reports referenced in the candidate, and the full texts of Church §3.3, Forster's Satz 1, and the omitted rational character-range result. Bibliography entries do not verify their contents.
  • The second, PL-thickening counterexample and any reported Smith computations. Neither is needed for the connected-sum counterexample; no computations were run.
  • Topics outside V1 and V2, including rank varieties and other non-configuration applications.
Source record

Additional supporting archive items are identified in the original report. View the manuscript hashes and repository snapshot.

Cite this review

If this review or its supporting reports inform your work, please cite it.

Igor Rivin. Review of FI-modules and stability for representations of symmetric groups. Math Referee, .

Download .bib
View BibTeX
@misc{rivin2026review-human-church-ellenberg-farb-fi-modules,
  author = {Rivin, Igor},
  title = {{Review of FI-modules and stability for representations of symmetric groups}},
  year = {2026},
  month = {November},
  date = {2026-11-20},
  howpublished = {Math Referee},
  url = {https://mathreferee.dimensionreducers.ai/papers/human-church-ellenberg-farb-fi-modules/panel/},
  note = {Review report}
}