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-astra high: cef-configuration-challenges

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.

The published all-n assertion about the total integral generator minimum is false under its own hypotheses. Both proposed manifolds witness the failure, whereas the FI# classification correctly predicts an eventual polynomial. The published residue-field inference is invalid but the bounded-generation theorem is repairable, even without finite Krull dimension. In the rational argument the filtration converse and the negative numerical degree assertions fail literally; removing the converse and clipping the bounds leaves the stated stability range intact. None of these distinctions depends on agreement between the reports.

CEX1

Checked statement: The all-n polynomial assertion for the minimum number of abelian-group generators in journal Theorem 1.9(c), arXiv item (3), tested on both proposed 7-manifolds.

Returned assessment: claim_false

Both counterexamples work, for ordinary ordered configurations and ordinary integral cohomology. Here are the topology and the group calculations.

Common diagonal step: in X=M^n each pair diagonal is a closed submanifold of codimension 7. For a finite relative chain in (X,Conf_n(M)) of dimension t<7 one can put its simplices in general position with respect to every pair diagonal, compatibly on faces and keeping the relative boundary in the open complement. Indeed, approximate and perturb on skeleta, extending relative to the already perturbed faces; transverse inverse images in each simplex have dimension at most t−7. There are only finitely many conditions, and it is not necessary for the union itself to be a manifold. The resulting relative chain is zero. Thus relative homology vanishes below 7, and inclusion is an isomorphism on H_j for j≤5 (one needs relative vanishing in j+1 as well as j). By naturality of integral cohomological UCT it gives the needed cohomology isomorphisms in degrees 3 and 4. There are no diagonals to check for n=0,1.

A: form the oriented connected sum N=L^7(3)#UTS^4#UTS^4. All summands are closed connected oriented 7-manifolds: the lens quotient uses a free orientation-preserving scalar action on S^7, and the oriented sphere bundle has oriented base and fiber. The complement of a point has a collared end and is diffeomorphic to the interior of the compact complement of an open coordinate ball; its boundary in this compact model is S^6. To compute the lens homology one can view the quotient of the Hopf bundle as a circle bundle over CP^3. The transition functions are cubed, so the Euler class is ±3h for the degree-two generator h. Multiplication by this class between successive even cohomology groups of CP^3 through degree six is multiplication by ±3. Gysin gives intermediate cohomology Z/3 in degrees 2,4,6 and zero in odd degrees; duality gives Z/3 in intermediate odd homology. For S^3→UTS^4→S^4 the Euler number is 2 (the height gradient on S^4 has two zeros of index +1). The nonzero middle Gysin map is Z→Z, multiplication by 2, so the only intermediate cohomology is H^4=Z/2, hence the only intermediate homology is H_3=Z/2. Excision of a ball does not change intermediate homology: the relative group is concentrated in degree 7 and the orientation class maps to its generator. In gluing punctured summands along S^6, Mayer–Vietoris adds their homology in degrees 1 through 5; degree 6 is also zero here and the oriented sum has a top class. Puncturing kills that class. Thus for M=N minus a point, H_1=Z/3, H_2=0, H_3=Z/3 ⊕ (Z/2)^2, and all positive-degree rational homology vanishes.

Write X_m=M^m. Split integral homological Künneth gives H_1(X_m)=(Z/3)^m and H_2(X_m)=(Z/3)^binom(m,2): adjoining a factor adds H_1(X_m)⊗H_1(M) in degree two. In degree three the new tensor terms, besides H_3(X_m), are H_3(M) and H_2(X_m)⊗H_1(M); the new Tor term is Tor_1(H_1(X_m),H_1(M)). Thus each step adds two order-two factors and 1+binom(m,2)+m order-three factors. The splitting as groups matters; this is not merely an order calculation. Summing from m=0 to n−1 gives H_3(X_n)=(Z/2)^(2n) ⊕ (Z/3)^[n+binom(n,2)+binom(n,3)]. Rational Künneth gives H_4(X_n;Q)=0, so Hom(H_4(X_n),Z)=0. UCT identifies H^4 with Ext^1_Z(H_3,Z), isomorphic as a group to this H_3. By the diagonal step this computes H^4 of the configuration space. For a finite abelian group the generator count is the largest of its primary generator counts: reduction at a prime gives the lower bound, and cyclic generators at different primes can be paired in the same slots by the Chinese remainder theorem. The count is therefore max(2n,(n^3+5n)/6). The cubic exceeds 2n for n≥3 since the difference is n(n^2−7)/6; at n=1,2 the counts are 2,4 instead of 1,3. Agreement on the infinite tail would force a polynomial to equal the cubic identically.

B: take finite polyhedral mapping cones of degree 3 on S^1 and degree 2 on S^2, wedged at a vertex. Such maps can be taken PL (for example suspend a degree-two circle map), and the mapping cylinders and cones admit finite subdivisions. The reduced homology is read off from the two attaching maps: H_1=Z/3 and H_2=Z/2, no other reduced homology. The resulting complex K has dimension 3. Place the vertices of a triangulation in general affine position in R^7 and extend linearly. Two simplices involve at most eight vertices, affinely independent in their union, so their convex hulls intersect exactly in their common face. This is an embedding, not just an immersion. Take a closed PL regular neighborhood in R^7. More explicitly, in an ambient subdivision with K a full subcomplex, use a derived neighborhood, cutting off the star of K by a level strictly between 0 and 1 of the barycentric weight on vertices outside K. This is a compact codimension-zero neighborhood with collared boundary (the PL regular-neighborhood construction); normalizing the weights on K and shrinking the outside weights retracts it, and also its interior, onto K. The positive intermediate levels have a product collar by rescaling the inside and outside weights. Its interior M is an open subset of R^7, hence orientable, and is connected. The boundary is nonempty, since a nonempty compact codimension-zero submanifold of R^7 cannot also be open and closed. One can equally use a regular thickening by small neighborhoods of the simplices with corners rounded. No prescribed exotic embedding or embedding of a 7-manifold in lower dimension is required.

Now H_2(M^n)=(Z/2)^n ⊕ (Z/3)^binom(n,2). The single factors give the first summand and pairs of H_1 factors give the second; all possible degree-two Künneth Tor terms involve a free H_0. Positive rational product homology vanishes, in particular Hom(H_3(M^n),Z)=0. Hence Ext^1_Z(H_2(M^n),Z) gives exactly the asserted H^3, unchanged by deleting the diagonals. Its generator count is max(n,binom(n,2)), equal to binom(n,2) for n≥3 (tie at 3), but not at 1 or 2. This independently disproves the all-n claim. Neither proof uses the alleged small-n Smith computations.

Premises used

  • Standard topology used in the constructions: PL regular neighborhoods of finite subcomplexes in Euclidean space, general position for finite chains, excision, Mayer–Vietoris, oriented-bundle Gysin sequence and integral Poincaré duality.
  • Integral homological Künneth over Z has an abstract group splitting, not necessarily natural. Integral cohomological UCT has kernel Ext^1_Z(H_{i−1},Z) and quotient Hom_Z(H_i,Z); Ext^1_Z(Z/m,Z)=Z/m by the multiplication-by-m resolution.
  • The all-n quantifier and the word 'minimum' in the journal image agree with the native claim; a polynomial upper bound or an eventual formula is not the assertion tested.

Missing inputs

Scope changes

  • The counterexamples need neither a non-Noetherian coefficient ring nor a manifold of infinite type.
  • They do not disprove all-n polynomiality for a fixed prime, rational rank, or field dimension.

Located concern

CEF-configuration-source 436, 2064; journal-config-p7 (c), journal-config-p66

The FI# decomposition gives an additive formula for rank and for each primary generator count, but not for the total minimum.

In A the controlling prime changes from 2 to 3; in B the two primary polynomials are n and binom(n,2). Both changes occur within the stated manifold hypotheses.

Replace item (c) by an eventual polynomial, with the exact all-n maximum formula in CEX2.

A false theorem item, not just a failed deduction. It does not contradict bounded generation of the integral FI#-module.

CEX2

Checked statement: The replacement for the total generator count, the residue-field step in Theorem 6.4.3, and the two universal-coefficient repairs with their support and finiteness hypotheses.

Returned assessment: repairable_gap

Put D_i=i in manifold dimension d≥3 and D_i=2i in dimension 2. The general-ring proof has a gap, but the generation theorem, unlike item (c), admits a repair. Here are details that avoid assuming a residue-field comparison for a Leray subquotient.

FI# input. The collar construction discards undefined labels and inserts new labels in a connected end, after compressing the old configuration. For composition one can use embeddings h_u with collar coordinate t∈(0,1) sent to u+(1−u)t, identity off the collar, for 0≤u<1. Use u>0 when inserting points with t<u and h_0 otherwise. In a composite the surviving original labels are moved by a single h_v, and all other output labels are constants in the vacated collar t<v. Configurations of a fixed finite list of constants there are path connected: in dimension ≥2 one can move points one at a time avoiding a finite set, using temporary positions if needed. Rescale their collar depths as v changes to the parameter for the single composite map. If there are no constants just isotope the compression. This proves the required equality of homotopy classes, without requiring strict functorial choices. Homology is thus covariant on FI#; contravariant cohomology is made covariant by taking the inverse partial bijection. On total injections this last convention is pullback along forgetting labels.

The classification writes V_S=⊕_{T⊆S} W_T. At each stage of the supplied proof, partial identities on distinct subsets of the same size multiply through smaller sets; after the smaller-support part has been removed they are orthogonal projections. Their images are copies of the values on those subsets via inclusion and partial retraction. Iterating gives the displayed decomposition, locally finite at each S. Extracting W (equivalently H_0 on FI#-modules) is exact, since this is an additive equivalence with FB-modules, where exactness is valuewise. In particular a bound on the sizes with W_T nonzero passes through extensions in FI#, not through arbitrary subobjects in FI. Finite generation in degrees ≤D means W_r=0 above D and the remaining W_r are finitely generated over the coefficient ring (the automorphism groups are finite). The growth test at lines 1385–1388 uses genuine summands: if V_n has a polynomial generator bound, each W_r is a quotient of a finitely generated V_n for n large enough, and a nonzero finitely generated W_r has a simple quotient k/m. There are binom(n,r) copies for every n≥r, forcing r≤D for a bound O(n^D). This does not require tensoring an arbitrary inclusion injectively. The finiteness condition in line 1385 should say finite also below d.

Field growth and integral finiteness. The stated field E_2 description would suffice: a base class in degree p uses at most p positive-degree factors by field Künneth and connectedness, and q pair classes have O(n^(2q)) choices. For p+(d−1)q=i this costs O(n^D_i), meaning O(n raised to D_i). But the citation in A is not a supplied proof in all characteristics. Here is an independent derivation. Write X_n=Conf_n(M). Forgetting the last point is a fiber bundle over X_m, m=n−1, with fiber M−P for the labeled m-point configuration P. Local trivializations come from moving the points in disjoint coordinate balls by ambient isotopies. Excision gives H^j(M,M−P;F) zero except for F^m in degree d, for any field F. Around a loop this is a constant system: the labels are ordered and the local orientations are preserved. The old H^*(M;F) is also constant since the transport extends by an isotopy of M. In the pair sequence H^b(M−P;F) is consequently built from constant pieces: the image from H^b(M) is a constant quotient of dimension at most b_b(M;F), and the remaining piece is a constant subspace of F^m only for b=d−1. This asserts a filtration, not trivial monodromy on the extension. Connectedness gives H^0=F. By the long exact sequence for coefficients on the base and the cohomological Serre spectral sequence, writing c_{m,j}=dim_F H^j(X_m;F) and setting negative subscripts to zero, one gets 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)}. Here c_{m,0}=1, since the fibers and bases are path connected. Induct on i: for d≥3 the increment bound is O((m+1)^(i−1)), since d−1≥2. For d=2 it is O((m+1)^(2i−1)). Summing in m proves the desired O((n+1)^D_i) over each field, including characteristic two. No constant uniform across fields is needed below.

The base manifold has finitely generated integral homology: the compact model is homotopy equivalent to its interior by the collar, and a compact smooth or PL manifold has finite CW type. For topological manifolds the usual compact ENR property suffices: take a finite union of small closed cubes in a retracting Euclidean neighborhood covering the embedded compact manifold, to obtain finite domination. To propagate finite generation to X_n, use the homological pair sequence for the same punctured fibers. The only relative groups are the local copies of Z in degree d. Fiber homology thus has a finite filtration by constant finitely generated abelian systems (subgroups or quotients of those local groups and of H_(M)); the maps between the constant systems are monodromy invariant. If H_(X_m;Z) is finitely generated degreewise, its homology with any constant finitely generated abelian coefficients is so by UCT. The local-coefficient long exact sequence and the homological Serre sequence now give degreewise finite generation at m+1, starting from X_0. This supplies an important prerequisite: field growth by itself does not exclude divisible integral groups.

Repair A. Decompose integral H_j(X_•)=M(W^(j)); the W_r^(j) are finitely generated abelian summands. If W_r^(j) is nonzero, choose F=Q for positive rank or F=F_p detecting some primary torsion. The homological UCT injection H_j(X_n;Z)⊗F → H_j(X_n;F) then bounds the latter dimension below by binom(n,r) for this fixed field as n varies. Field cohomology is dual to field homology, so the upper bound above gives r≤D_j. Now apply the natural cohomological UCT over Z, with abelian coefficient group the underlying group of k: 0 → Ext^1_Z(H_{i−1},k) → H^i(X_•;k) → Hom_Z(H_i,k) → 0. The maps are k-linear by naturality in the coefficients, and natural for configuration maps without choosing a splitting. Hom and Ext in the first variable reverse arrows; after partial inversion they send a subset summand to the corresponding relabeled dual-coefficient summand if the subset survives, and to zero otherwise. This follows directly from the projection, permutation and inclusion description of M(W) and additivity on finite sums. Thus the right end has support ≤D_i, the left ≤D_{i−1}≤D_i (left end zero for i=0). On a free cyclic summand Hom gives k and Ext gives 0. On Z/h they give k[h]=Ann_k(h) and k/hk, respectively, by the two-term free resolution. These are finitely generated k-modules if k is Noetherian. Exact extraction of W and lifting finite generating sets across a short exact sequence prove the result with no Krull dimension restriction.

Repair B. The alternate sequence is also usable, but one must justify change of coefficients for cochains. Degreewise finite generation of integral homology suffices. Indeed the free singular chain complex admits a degreewise finite free chain model: take finite free Z-presentations of each H_j with free relation group, lift generators to cycles and relations to bounding chains, and sum these two-term complexes in degrees j,j+1. The map induces a homology isomorphism and its free acyclic mapping cone is contractible by splitting onto free boundary groups over Z. This argument establishes the coefficient comparisons, not a functorial choice of finite model. In particular the natural maps H^j(X;Z)⊗P → H^j(X;P) for free abelian P are isomorphisms, even for infinite rank, by computing on the finite-in-each-degree model. A free resolution of the abelian group k and the coefficient long exact sequence give naturally 0 → H^i(X;Z)⊗k → H^i(X;k) → Tor_1^Z(H^{i+1}(X;Z),k) → 0. This is functorial in X and k, hence k-linear and FI#-compatible; a natural splitting is not claimed. The integral support bound used by B follows already from repair A at k=Z. Alternatively integral cohomology is degreewise finitely generated by UCT, and the left injection with k=F_p detects any nonzero finitely generated coefficient W_r of H^i(Z) for a suitable p, giving r≤D_i by field growth. For the right term write H^{i+1}(Z)=M(W′). At every r>D_i exactness of coefficient extraction and the field case for H^i(F_p) give Tor_1^Z(W′_r,F_p)=0 for every p. This is W′_r[p]; hence these finitely generated groups are free and have zero Tor with any k. Tensor and Tor respect the finite subset sums and relabelings, so both ends have support ≤D_i. B's single excess degree i+1 applies in d≥3; in d=2 one must cover both 2i+1 and 2i+2. For the remaining finite coefficients, tensor gives free and k/hk pieces and Tor gives k[h] pieces, finitely generated in the Noetherian case. This completes B's proof with the surface modification. Over arbitrary commutative rings one obtains the support bound, not automatic finite generation of annihilator ideals.

Arithmetic conclusion. For the integral H^i write r_j=rank W_j and t_{j,p}=dim_Fp(W_j[p]) for its coefficient groups. Put R(n)=Σ_j r_j binom(n,j), T_p(n)=Σ_j t_{j,p} binom(n,j). There are only finitely many relevant j and torsion primes. The exact minimum is R(n)+max_p T_p(n) (take zero for absent torsion). Indeed reduction modulo p requires r+μ(T_p) generators of a group of free rank r, and juxtaposing the free generators with aligned primary cyclic generators attains the maximum. Primary generator counts, not primary group orders, are additive. The finite list of differences T_p−T_l has constant signs for all sufficiently large n, ignoring identically zero differences. Thus one polynomial controls the entire tail, of degree at most D_i. Each fixed T_p and the rational rank are polynomials for every n. Each field Betti number has its own all-n binomial sum from its field FI#-module; mod-p cohomology need not equal reduction of integral cohomology. Hence Theorem 1.9(a),(b),(d) survive with degree upper bounds, while (c) has only the maximum formula globally. Read (a) as rational cohomology: the journal image actually prints F_p inside the purported rational dimension, and the arXiv gloss omits the cohomology symbol.

Premises used

  • Manifolds and configurations as in 2023–2026 and 2044–2048; the connected collar component and d≥2 are important to the FI# construction.
  • Classification and support/generation test in the supplied FI# section; ordinary exactness in the abelian functor categories.
  • Standard bundle Serre spectral sequences with local coefficients, collar and compact-manifold finiteness (compact ENR in the topological category), excision and the ordinary integral and field universal coefficient theorems.
  • k is a commutative ring with unit. Noetherianity is sufficient for the finiteness of k[h]. No uniform field estimate, Forster theorem or later configuration-space generation theorem is used in the repairs.

Missing inputs

  • The earlier proof/external all-characteristic Leray presentation cited for E_2 is not supplied for independent verification. The forgetful-bundle estimate above replaces that input; it does not certify that citation.

Scope changes

  • Remove finite Krull dimension in the repaired finite-generation theorem, retaining Noetherianity as a sufficient finiteness assumption.
  • For (c) change polynomiality for every n to eventual polynomiality; the all-n description is a finite maximum plus the free-rank polynomial.
  • The rational all-n character-polynomial consequence for the boundary case also survives: in the finite subset sum only invariant subsets contribute trace, counted by choosing cycles as at lines 1392–1397.

Located concern

CEF-configuration-source 2057–2061; journal-config-p65–p66

A dimension bound on a fiber of a module does not pass to fibers of its submodules. A spectral sequence over the residue field is also not automatically the tensor of the original one.

In R=C[x,y], m=(x,y), F=R/m, the fiber of R has dimension 1 but the fiber m/m² of its ideal has basis x,y. Even m^s/m^(s+1) has dimension s+1 inside this example of submodules of a one-generator ambient module.

Use either natural UCT argument above after establishing field growth and integral finiteness. Forster's estimate needs a valid fiber bound and a finitely generated module to start with (without finiteness, e.g. Q/Z over Z has all residue fibers zero).

The general-k inference in the printed proof fails. Field dimension really is monotone on vector-space subquotients, and subgroups of an N-generated abelian group need at most N generators (lift to a subgroup of Z^N); those arguments, given the respective input bounds, do not have this particular defect.

Located concern

prior-report-B-config-p10 Repair

The indices i and i+1 and the phrase 'top piece' do not prove the surface version as written.

For surfaces integral degree i+1 has support bound 2i+2, whereas the desired bound in degree i is 2i.

Test the Tor quotient for every r>D_i and every F_p in degree i, using exact extraction on FI#.

The modified repair covers both dimension ranges and gives finite generation over every commutative Noetherian k.

Located concern

CEF-configuration-source 425, 432–437; journal-config-p6–p7

The growth and support arguments give no equality of the polynomial degree with i or 2i.

For example H^1 with integral coefficients is torsion free by UCT since H_0 is free, so its fixed-prime torsion generator polynomial is zero, also on surfaces.

Use 'degree at most' throughout, correct the Betti-number glosses, and change the all-n assertion for the total minimum separately.

Literal exact degrees are false independently of the prime switch in (c).

CEX3

Checked statement: The equivalence of filtered and graded-piece bounds in Lemma 6.3.2(b), the numerical injectivity bound of Theorem 6.3.1 and its page induction, and the consequences for rational stability.

Returned assessment: claim_false

Write Φ_a(V)n=(V{n+a})_{S_n} with the a distinguished letters fixed. In ranges of injectivity or surjectivity one tests the transition n→n+1 for every a. These functors are exact over a ring containing Q: averaging identifies coinvariants with invariants and permits invariant lifts in a surjection. For a short exact sequence the transition diagram then has exact rows. If both outer vertical maps are injective so is the middle one (project a kernel element to the right and then test it on the left); if both are surjective so is the middle one (lift on the right and correct on the left). Iterating gives graded bounds ⇒ filtered bounds, separately for the two thresholds. Conversely take the constant FI-module V=Q, with all transition maps identity, and its sub-FI-module F zero in sizes below N≥1 and equal to V thereafter. Injections never decrease size, so this is a valid filtration 0⊂F⊂V. V has both degrees zero. For a=0 at n=N−1 the F transition is 0→Q and the V/F transition is Q→0. Thus F fails surjectivity and V/F fails injectivity in the claimed range. Indeed F has surjectivity degree exactly N since for n≥N the transition is the identity for every a. Pointwise splitting does not give an FI splitting. This defeats the 'only if' even with finite generation and rational coefficients.

Part (a) passes its separate test. In the transition diagram of complexes exactness identifies the tested homology with the homology of the tested complex. To prove injectivity on homology, if a cycle maps to a boundary, lift a witnessing preboundary using surjectivity on X and subtract its boundary, then use injectivity on Y. For surjectivity lift a target cycle using surjectivity on Y; its differential vanishes by injectivity on Z. This gives exactly max(B,C) and max(D,E).

Numerically, the degree is a nonnegative threshold, so a bound i+2−d<0 is not a possible degree inequality. In degree zero, e.g. d=3, configurations are path connected (move labels along paths avoiding finitely many points), including the singleton empty configuration space at n=0. H^0 is the constant Q module of injectivity degree 0, not ≤−1. Interpreted as 'the maps are injective for all allowed n past that threshold', the intended map property makes sense and is unchanged by clipping at zero.

Here is the induction with actual degree bounds. Use the supplied rational E_2 bounds and put D=d−1≥2, h=p+2q+1−D, h_+=max(0,h). Rows are indexed by qD. On the first possible differential (page D+1) the incoming term, if in the quadrant, has surjectivity bound p−D−1+2(q+1)=h, and the middle and outgoing terms have injectivity bound zero. Absent terms impose no condition. Part (a) gives injectivity ≤h_+ and surjectivity ≤p+2q on the next distinct page. For k≥2 retain surjectivity bound s_k=p+2q+min(k−2,q−1)(D−2) for q>0. At the step using page kD+1, the incoming term at (p−kD−1,(q+k)D), when present, has bound p−kD−1+2(q+k)+(k−2)(D−2)=p+2q+3−2D≤h. Together with injectivity ≤h_+ in the middle this preserves h_+. For 0<q<k there is no outgoing term, and s_k≤s_{k+1} suffices. For q≥k the outgoing term at (p+kD+1,(q−k)D) has clipped injectivity bound p+2q+(k−1)(D−2), already nonnegative, even at q=k. The middle surjectivity bound is p+2q+(k−2)(D−2); their maximum is s_{k+1}. No negative degree was used for a zero module outside the quadrant.

Each fixed bidegree stabilizes after finitely many pages by the quadrant bounds. At p+Dq=i the final injectivity bound is ≤max(0,i+1−D)=max(0,i+2−d). For q>0 the final surjectivity bound is p+2q+(q−1)(D−2)=i+2−D≤i. For q=0 there was never an outgoing differential, and surjectivity ≤p=i passes from E_2 to its quotient. Weight is bounded on these subquotients by p+2q≤i, and the same bound passes through the finite filtration (in characteristic zero, irreducible constituents of an extension come from its pieces). Line 1977 uses just graded ⇒ filtered. Consequently stability degree is ≤i, weight is ≤i, and the cited weight-plus-stability character range n≥2i is unchanged within the original finite-dimensional character setting. The colored-configuration application likewise needs only stability degree ≤i: with |μ| distinguished letters the remaining coinvariants stabilize for n−|μ|≥i, equivariantly for S_|μ|, hence so do the dimensions after taking S_μ-invariants. It is this subgroup, not necessarily the whole S_|μ|, whose fixed space is needed in the last sentence at line 2011.

Premises used

  • The ordinary nonnegative-degree convention for injectivity/surjectivity; a threshold condition on n≥0 can equivalently be stated with a negative lower limit, but its least nonnegative degree cannot be negative.
  • Rational first-quadrant FI spectral sequence, row spacing and initial bounds E_2^(p,qD) ≼ (0,p+2q), weight ≤p+2q, as stated at lines 1932–1938 and in journal-config-p61.
  • The cited conversion of stability degree plus weight to the rational character range used at line 1983; the induction does not reprove the general character-polynomial theorem.

Missing inputs

  • The prior rational E_2 decomposition and the general character-range theorem are inputs from outside the supplied proof segment; the range check is relative to these inputs.

Scope changes

  • Replace an equivalence by a sufficient condition; do not infer graded bounds from filtered bounds for a general FI-filtration.
  • Distinguish an injectivity-degree inequality from an assertion about which nonnegative transition indices work.

Located concern

CEF-configuration-source 1928; journal-config-p61 Lemma 6.3.2(b)

The converse filtration implication is false, not a consequence of exact coinvariants.

The threshold submodule of a constant module and its quotient have bad opposite transitions while the ambient transition is the identity.

Delete 'only if'; retain propagation of bounds from all pieces to the extension.

No loss in the spectral-sequence application. Report B's statement that the diagram chase was checked does not verify the converse.

Located concern

CEF-configuration-source 1920, 1940, 1950–1952, 1961–1967, 1977–1980; journal-config-p61

The unclipped injectivity bound is literally false in low total degrees.

The base step takes a maximum with the middle injectivity bound zero; omitting that maximum can give −1 already for p=q=0, d=3.

Clip each injectivity estimate at zero, omit nonexistent incoming/outgoing terms, and state final injectivity ≤max(0,i+2−d).

Changes the literal injectivity assertion, not the uniform bound on stability degree, the rational character range, or the colored range.

Located concern

CEF-configuration-source 423–425; journal-config-p6 Theorem 1.8

Specializing a character polynomial of degree at most i gives only degree at most i for the Betti polynomial.

For M=S^3 and i=1, the codimension-three diagonal argument gives an isomorphism on H_1 with (S^3)^n, hence b_1=0, not a degree-one polynomial on the tail.

Say 'of degree at most i' in the Betti conclusion.

No change to polynomial agreement for n≥2i.

Outside the audit

  • Unprovided arXiv sections, other journal pages and unsupplied external papers, including the full proofs of the cited Leray description, Forster estimate and general rational character-range theorem.
  • Report B's claimed Smith normal form runs: no code supplied or execution performed; the group formulas were derived instead.
  • Rank-variety, graded-algebra, multiplicity and other non-configuration targets incidentally mentioned on the supplied pages; no scheme-theoretic or reduced rank-variety conclusion is assessed here.
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}
}