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

claude-fable-5-2 high: cef-algebra

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.

A1 (Section 5.1 and Theorem 1.11) and A2 (Section 5.2) hold with no mathematical error found; A3 (Section 5.3) is false in its stated generality. A1: the only defects are an unprinted but immediate proof of pr:coinvariantfinite, typos, and loose wording about which sequence is representation stable. A2: the co-FI-ideal computation and the BS^1 identification check out; the result is conditional on the unsupplied [BS, Section 11.2] presentation and needs a one-line duality step the proof omits. A3: the setup wrongly calls the generators of J_n homogeneous, so the grading is undefined for non-monomial p. The proof's claim that rank varieties form an FI#-subscheme is false, with explicit matrices for p = T-1 and p = T^2. The theorem fails literally for the constant P=(1). Remark 5.3.3 fails for P=(T^2), r=(1), d=2 over C. What survives is the determinantal case p_l = c_l*T, where the printed proof works for all n >= 0, scheme and reduced. A new repair gives eventual polynomiality for monomial P. The scheme-theoretic 'all n >= 0' claim for P=(T^e), e >= 2, remains unresolved. All positive conclusions are conditional on core theorems (th:noetherian, th:FIequiv, thm:persinomial, thm:fisubalgebra, th:charFIsharp, thm:polyFIsharp) that were read but not audited in this packet. No tools, web access or external sources were used; line numbers inside the given ranges are hand-counted approximations; the journal text was not collated.

A1

Checked statement: (a) R(P)=k[M(1)]/((P)) is a graded FI-algebra of finite type, and R(P_d)^i_n is uniformly representation stable over Q. (b) W^inv is a co-FI-submodule and coinv(A)=A/((A^inv_{>0})) is a graded co-FI-algebra. (c) pr:coinvariantfinite: over a field of char 0, coinv(A) inherits finite type and slope <= m. (d) Theorem 1.11 (theorem:intro:polynomial:answer): the characters of R^{(r)}_J(n) are eventually polynomial of degree <= |J|, and dim R^{(r)}_J(n) is eventually a polynomial of degree <= |J|. (e) The three low-degree examples in eq:coinvariantexamples.

Returned assessment: no_issue_found

No mathematical error found; the conclusions are conditional on the cited core theorems (th:noetherian, th:FIequiv, thm:persinomial, thm:fisubalgebra), which I read but did not audit.

(a) k[M(1)] is generated by its grading-1 part M(1), which has V^0=0 and is finitely generated, so thm:fisubalgebra applies over any commutative k. The FI-ideal generated by homogeneous P_j is graded, and quotients of finite type are of finite type. The concrete description of R(P)_S is correct. Representation stability of R(P_d)^i over Q then follows from th:FIequiv.

(b) For a co-FI-module W, an injection f:T->T' and sigma in S_T, pick sigma' in S_{T'} with f sigma = sigma' f. Contravariance gives sigma_(f_ w) = f_(sigma'_ w) = f_* w for w invariant. Hence f_(W^inv_{T'}) lies in W^inv_T (the printed inclusion has a typo). Each f_ is a graded unital algebra map, so it sends the ideal generated by positive-degree invariants of A_{T'} into the corresponding ideal of A_T. Thus coinv(A) is a graded co-FI-algebra; no surjectivity of f_* is needed. The remark that the J_n do not form an FI-ideal for the covariant structure is correct.

(c) pr:coinvariantfinite has no printed proof. It follows at once from the co-FI paragraph before pr:cofisubalgebra: over a field, the surjection A^i ->> coinv(A)^i dualizes to an injection of FI-modules (coinv(A)^i)^* into (A^i)^*. The target is finitely generated by hypothesis, so th:noetherian (k contains Q) makes the submodule finitely generated. Weight passes to subquotients, which gives the slope statement.

(d) Checked step by step. The co-FI structure on M(1) (e_{f(s)} -> e_s, other e_t -> 0) has dual FI-module isomorphic to M(1), so W=M(1)^{+r} in grading 1 is of finite type with wt(W^)=1. pr:cofisubalgebra uses k{W}^ = k{W^}, valid because all pieces are finite-dimensional, followed by th:noetherian; it gives Q[M(1)^{+r}] finite type and slope <= 1. pr:coinvariantfinite passes this to R^{(r)}. The multigrading is preserved by the co-FI maps, and the ideal is multihomogeneous because multihomogeneous components of invariants are invariant. So (R^{(r)}_J)^ is a direct summand of ((R^{(r)})^{|J|})^*, hence finitely generated of weight <= |J|. thm:persinomial then gives a character polynomial of degree <= |J|. Characters of S_n-representations in char 0 are self-dual. Specializing to (n,0,0,...) gives dimension degree <= |J| since deg X_1 = 1. Passing from Q (proof) to a char-0 field k (statement in the introduction) is harmless: invariants and the ideal they generate commute with the flat base change.

(e) Verified by hand. Degree 1: quotient of the permutation representation by e_1 gives X_1 - 1 for n >= 1. Degree 2: Sym^2 has character C(X_1,2)+X_1+X_2; the ideal in degree 2 is e_1V_1 (character X_1) plus p_2, which is not in e_1V_1 for n >= 2; this gives C(X_1,2)+X_2-1, and the formula fails at n=1 as the stated range indicates. Bidegree (1,1): the ideal is e_xV_y + e_yV_x (dimension 2n-1) plus sum x_i y_i, independent for n >= 2; this gives X_1^2 - 2X_1 = 2C(X_1,2) - X_1.

Premises used

  • thm:fisubalgebra: an algebra generated by a finite-type graded sub-FI-module V with V^0=0 is of finite type (any k); slope <= m is inherited (char 0)
  • th:noetherian (k Noetherian containing Q): sub-FI-modules of finitely generated FI-modules are finitely generated
  • pr:cofisubalgebra and the preceding co-FI paragraph (field of char 0)
  • pr:Mmuweight: wt(M(1)) = 1
  • thm:persinomial: finitely generated FI-module in char 0 has a character polynomial of degree <= weight, valid eventually
  • th:FIequiv: finitely generated implies uniformly representation stable (char 0)
  • Self-duality of S_n-representations in char 0

Missing inputs

  • Chevalley's theorem, [CF, Thm 7.4] and [AAB, Cor 5.2] are cited only as comparisons; not supplied and not needed for the logic
  • Journal full text not collated

Scope changes

  • All finite-type conclusions for co-FI-algebras in Section 5.1 need a field of characteristic 0, as stated.
  • Theorem 1.11 gives only eventual polynomiality; no explicit range is claimed or derivable from this argument (it goes through th:noetherian).

Located concern

pr:coinvariantfinite, approx. lines 1730-1733

Missing proof (not an error): the proposition is stated with no proof or pointer.

No proof environment follows the statement. The needed argument is the dualization plus th:noetherian given in the argument field; it uses only the co-FI paragraph before pr:cofisubalgebra.

Add one sentence: coinv(A) is a quotient co-FI-module of A, so its graded dual is a graded sub-FI-module of A^*; apply th:noetherian and monotonicity of weight under subquotients.

None.

Located concern

para 'Coinvariant algebras' approx. line 1724; proof of theorem:intro:polynomial:answer approx. line 1746; pr:cofisubalgebra statement

Typographical slips: 'f_(W^inv_{T'}) subset f_(W^inv_T)' should have target W^inv_T; 'wt <= J' and 'degree <= J' should read |J|; pr:cofisubalgebra says 'If V has slope' for W.

Direct reading; the intended statements are the ones verified above.

Correct the typos.

None.

Located concern

Example after pr:coinvariantfinite, approx. line 1736

Imprecision: 'uniformly representation stable' is asserted for R^i_n, but the consistent sequence to which th:FIequiv applies is the dual FI-module (R^i)^*. The co-FI maps R_{n+1} -> R_n go the other way.

definition:repstab requires maps V_n -> V_{n+1}; these exist on the duals only.

State stability for the dual sequence. Multiplicity stability for R^i_n itself follows by self-duality.

None for multiplicities or characters.

A2

Checked statement: Proposition of Section 5.2: for k a field of char 0 there is a graded co-FI-algebra BS of finite type with BS_n = G(R_n/k) as graded S_n-representations (diagonal S_n in S_n x S_{n+1}); hence G(R_n/k)^i is uniformly representation stable. Also the displayed identification (BS^1)^* = M(1) tensor M(1) = M(2) + M(1,1) + M(1).

Returned assessment: no_issue_found

The construction and finite-type argument check out, conditional on the external presentation of G(R_n/k) from [BS, Section 11.2] and on th:noetherian and th:FIequiv.

Structure. R has R^0=M(0), R^1 = dual of the FI-module M(1), and zero above; it is a co-FI-algebra with maps x_{f(s)} -> x_s and x_t -> 0 for t outside f(S). The augmentation R ->> M(0) is a map of co-FI-algebras. So def:tensorbullet applies; functoriality holds because the augmentation commutes with the f_*.

Ideal. On the first factor, f_(x_t) is x_s if t=f(s) and 0 otherwise. On R^{tensor bullet}, a term with x_t in position t' is killed by the augmentation when t' is outside f(S); otherwise it goes to x_s in position s', or to 0. Hence f_(gamma_T(x_t)) = gamma_S(x_s) if t=f(s), and 0 otherwise. So J is a co-FI-ideal and BS_n = R_n^{tensor (n+1)}/J_n. The S_n action is the diagonal one, with S_n embedded in S_{n+1} fixing the extra tensor factor. There are no Koszul signs since R is treated as commutative.

Finite type. pr:Vtensorbullet is stated for FI-modules, while R is a co-FI-algebra; the proof leaves the bridge implicit. Since all pieces are finite-dimensional, (R^{tensor bullet})^* = (R^)^{tensor bullet}: the dual of 'augment the factors outside f(S)' is 'insert the unit M(0) -> R^ in the factors outside f(S)'. R^* is a graded FI-module with (R^)^0 = M(0) and (R^)^1 = M(1). I checked the printed proof of pr:Vtensorbullet; in this case it gives ((R^)^{tensor bullet})^i generated in degree <= 2i. pr:gradedtensor gives R^ tensor (R^)^{tensor bullet} of finite type. BS^ is a graded sub-FI-module of it, so th:noetherian (char 0) applies. th:FIequiv then gives representation stability of the dual sequence, and self-duality transfers multiplicities.

BS^1. In grading 1 the relations gamma(x_t) eliminate the position-0 generators, so BS^1 = (R^{tensor bullet})^1 as co-FI-modules. Its dual has basis the pairs (position t', variable t) in T x T with FI acting diagonally, i.e. M(1) tensor M(1). Splitting into equal and distinct pairs gives M(1) + M(2), and M(2) = M(k[S_2]) = M(trivial) + M(sign) in char 0. This matches the display.

Premises used

  • External: [BS, Section 11.2], that the defining ideal of G(R_n/k) in R_n^{tensor (n+1)} is generated by the elements gamma_n(x_i). Imported with available_from = arXiv:1006.2562 (June 2010 by its identifier, before the 2013-07-14 cutoff); date_basis = arXiv identifier as cited in the bibliography; version and section number not verifiable here.
  • def:tensorbullet (co-FI version), pr:Vtensorbullet (hypothesis V^0 = M(0), V of finite type), pr:gradedtensor
  • th:noetherian (char 0) for passage to the quotient BS
  • th:FIequiv for representation stability

Missing inputs

  • Text of [BS] Section 11.2 (not supplied): the exact statement, its characteristic hypotheses, and the grading on G(R_n/k). From my recollection of the Bhargava-Satriano definition (relations e_j(a^(1),...,a^(n+1)) = s_j(a) for all a), the claim is consistent in char 0: for nilpotent nu with nu^2=0 one has e_1^m = m! e_m, so the higher relations lie in the ideal generated by the e_1's. This rests on an unsupplied definition and is not a verification.

Scope changes

  • Conclusion is conditional on the [BS] presentation and on char 0 (used for th:noetherian, and plausibly for the presentation itself).
  • No stable range is claimed. The argument does give (BS^i)^* inside an FI-module generated in degree <= 2i, hence weight <= 2i.

Located concern

Proof, approx. line 1780 ('Since R is of finite type, Propositions pr:Vtensorbullet and pr:gradedtensor imply...')

Minor exposition gap: pr:Vtensorbullet is an FI-module statement applied to a co-FI-algebra without the identification (R^{tensor bullet})^* = (R^*)^{tensor bullet}.

Finite type for co-FI-modules is defined via the dual (para 'Co-FI-algebras'). The dual of the co-FI structure map of def:tensorbullet is the FI structure map of def:tensorbullet for R^* with unit M(0) -> R^*; this holds because R_T is finite-dimensional.

Insert the one-line duality identification.

None.

Located concern

Proposition statement, approx. lines 1761-1762

Imprecision: uniform representation stability holds for the dual consistent sequence (BS^i_n)^*; the co-FI maps on BS^i_n go from n+1 to n. 'Diagonal subgroup' tacitly uses the standard embedding of S_n in S_{n+1}.

Same as in A1; the proof itself writes (BS^i)^* = BS^i.

State stability for the dual FI-module and multiplicity stability for G(R_n/k)^i.

None for multiplicities or characters.

A3

Checked statement: Setup of Section 5.3 (J_n generated by 'homogeneous polynomials' det(B(p_l(A))), with O(X_{P,r}(n)) inheriting a grading). Theorem 5.3.2 (theorem:sections:polygrowth): for arbitrary P, r, d over any field, dim O(X_{P,r}(n))^d is a polynomial in n of degree <= 2d for all n >= 0, and likewise the characters in char 0. Remark 5.3.3: the same holds for the reduced variety by the same argument.

Returned assessment: claim_false

Three things must be kept apart: an ill-posed setup, a false step in the proof, and false statements.

(1) Setup. Entries of p(A) are homogeneous only when p is a monomial c*T^e. For other p the minors are inhomogeneous, J_n is not a homogeneous ideal, and k[x]/J_n inherits no grading. So O(X_{P,r}(n))^d is undefined for general P.

(2) Proof step. The FI# maps on Mat are 'restrict to the principal A x A block, then extend by zero'. Rank varieties are not preserved in general. Extension by zero fails when p(0) is nonzero, since p(M + 0) = p(M) + p(0)*I. Principal restriction fails already for p = T^2. So X_{P,r} is not an FI#-subscheme and O(X_{P,r}) is not a quotient FI#-algebra of O(Mat). Explicit matrices are in the issues.

(3) Theorem as stated. It is literally false for the constant polynomial P=(1), r=(1), d=1; here J_n is homogeneous, so the statement is meaningful. For non-monomial P it is ill-posed. For monomials T^e with e >= 2, the scheme-theoretic 'all n >= 0' claim is unproved and I could not decide it by hand.

(4) Remark 5.3.3. False for the nonconstant, homogeneous case P=(T^2), r=(1), d=2 over C: the reduced coordinate ring has degree-2 dimension 0, 1, then n^2(n^2+1)/2 - 1 for n >= 2, which is not a polynomial on all n >= 0. Full proof in the issues.

(5) What survives. Determinantal case (every p_l a nonzero scalar multiple of T, or zero): minors go to minors or 0 under both kinds of FI# map, J is a homogeneous FI#-ideal, and the printed proof works. O(X)^d is a quotient FI#-module of Sym^d(M(1) tensor M(1)), generated in degree <= 2d as an FI-module by pr:tensor (any k). thm:polyFIsharp (vi)/(vii) gives polynomial dimension over any field, and polynomial characters in char 0, for all n >= 0, of degree <= 2d. In positive characteristic the degree bound must come from generation degree, not from 'slope', which is defined only in char 0. The reduced version also holds here: if g^m lies in J_T then f^*(g)^m lies in J_S, so the radical of an FI#-ideal is an FI#-ideal, and it is homogeneous.

(6) New repair (my argument, not the printed proof): eventual polynomiality for monomial P. Suppose every p_l = c_l*T^{e_l} with e_l >= 1. Then J_n is homogeneous. The pullback along extension by zero (x_{f(s)f(s')} -> x_{ss'}, other variables -> 0) sends the generic matrix to A_S + 0, and p(A_S + 0) = p(A_S) + 0, so minors go to minors or 0. Thus J and its radical are co-FI-ideals, and O(X)^d and O(X_red)^d are quotient co-FI-modules of O(Mat)^d. The dual of O(Mat)^d is the FI-module of symmetric d-tensors on M(1) tensor M(1): a permutation module on size-d multisets of ordered pairs, generated in degree <= 2d, of weight <= 2d. In char 0, th:noetherian makes the dual of O(X)^d finitely generated, and thm:persinomial gives characters and dimensions eventually polynomial of degree <= 2d. In positive characteristic the same argument needs the Noetherian property and eventual polynomiality of dimension over arbitrary fields from [CEFN] (arXiv:1210.1854, October 2012 by identifier, before the cutoff; cited in the paper as Theorem A; version content not verified here). This gives 'for n >> 0', not 'for all n >= 0'.

Disclosure: I cannot remove later knowledge from model weights. Nothing above relies on post-cutoff results. I am not aware of, and do not cite, any published correction.

Premises used

  • remark:FIsharpremarks (FI# isomorphic to its opposite), thm:fisubalgebra, pr:tensor, thm:polyFIsharp, th:charFIsharp, thm:persinomial (for the determinantal case)
  • th:noetherian and thm:persinomial (for the new char-0 repair)
  • Elementary linear algebra: Jordan form; quadrics vanishing on rank <= 1 matrices are spanned by the 2x2 minors (proved in the issue below)
  • Hilbert Nullstellensatz over C (to identify the reduced ideal with the vanishing ideal)

Missing inputs

  • A proof or disproof of the scheme-theoretic 'all n >= 0' statement for P=(T^e), e >= 2. Unverified heuristic only: for P=(T^2), r=1, the relation [A tensor 1 + 1 tensor A, Lambda^2(A^2)] = 0 gives linear syzygies among the generators of J_n in degree 5 whose count suggests the same small-n defect as in the reduced case. I did not prove these are all the syzygies.
  • [CEFN] text (for the positive-characteristic version of the repair); exact theorem numbers and version not verified
  • Eisenbud-Saltman [ES] and Weyman [Wey] are not supplied; they are not used in the logic

Scope changes

  • Theorem 5.3.2 as printed is established only for determinantal varieties (each p_l a nonzero scalar multiple of T, or zero): any field, all n >= 0, degree <= 2d, both scheme and reduced versions.
  • For p_l = c_l*T^{e_l}, e_l >= 1: only eventual polynomiality (degree <= 2d) is available, via the new repair. Char 0 uses th:noetherian and thm:persinomial; positive characteristic is conditional on [CEFN]. Scheme version for all n >= 0 is unresolved; reduced version for all n >= 0 is false.
  • For non-monomial p_l the graded statement is ill-posed; a filtered reformulation would be a different theorem and is not audited here.
  • table:examples item 5 and cor:polydim:app: for general P the entry is not justified as a finitely generated FI-module. For monomial P with e_l >= 1 the dual of O(X_{P,r})^d is a finitely generated FI-module (char 0), which suffices for the eventual-polynomiality conclusion of cor:polydim:app.

Located concern

Section 5.3 setup, approx. lines 1801-1802 ('generated by the homogeneous polynomials ... inherits a grading')

Incorrect intermediate assertion: the generators of J_n are not homogeneous unless every p_l is a monomial, so the grading on O(X_{P,r}(n)) is undefined for general P.

P=(T-1), r=(1), n=2: J_2 is the principal ideal generated by (x_11 - 1)(x_22 - 1) - x_12 x_21. A principal ideal with an inhomogeneous generator is not homogeneous.

Restrict to monomial p_l (in particular p_l(0)=0), or replace the grading by the degree filtration (functions that are restrictions of polynomials of degree <= d) and restate the theorem.

The theorem is ill-posed for non-monomial P.

Located concern

Proof of theorem:sections:polygrowth, approx. line 1820 ('The rank varieties X_{P,r}(S) in Mat_{SxS} are preserved by these maps')

False step. Neither half of the FI# structure preserves rank varieties in general.

(i) Extension by zero, p(0) nonzero: P=(T-1), r=(1). A=(1) lies in X(1) since A-I=0. Its extension diag(1,0,0) has A'-I = diag(0,-1,-1) of rank 2 > 1, so it is not in X(3). (ii) Principal restriction, P=(T^2), r=(1), any characteristic: M = [[I_2, I_2],[-I_2, -I_2]] in Mat_4 has M^2=0, so M is in X(4). Its principal block on {1,2} is I_2 with rank(I_2^2)=2 > 1, not in X(2). On rings: det((A restricted to {1,2})^2) lies in J_2, but as an element of k[x_ij : i,j <= 4] it is nonzero at M in V(J_4), so it is not in J_4 or its radical. The covariant map O(X(2)) -> O(X(4)) therefore does not exist.

Restrict the FI# argument to the determinantal case p_l = c_lT with c_l nonzero. For monomial p_l = c_lT^{e_l}, e_l >= 1, only the co-FI half (extension by zero) survives; use it with th:noetherian to get eventual polynomiality (new repair, argument item 6).

The printed proof establishes the theorem only in the determinantal case.

Located concern

theorem:sections:polygrowth, approx. lines 1811-1818 ('Fix P, r, and d ... for all n >= 0')

False as stated in full generality: a degenerate literal counterexample satisfies every stated hypothesis.

P=(1) (constant polynomial in k[T]), r=(1), d=1, any field. p(A)=I_n. For n <= 1 there are no 2x2 minors, so J_n=0 and O = k[x_ij]. For n >= 2 the principal 2x2 minor of I_n is 1, so J_n=(1) and O=0. All generators are constants, so J_n is homogeneous and the statement is meaningful. dim O(X(n))^1 = 0, 1, 0, 0, ... for n = 0, 1, 2, 3, ...; a polynomial vanishing for all n >= 2 is zero, contradicting the value 1 at n=1.

Add hypotheses. With p_l = c_lT (c_l nonzero) the printed statement and proof hold. With p_l = c_lT^{e_l}, e_l >= 1, claim polynomiality only for n >> 0.

Theorem 5.3.2 must be restricted. Outside the determinantal case the 'all n >= 0' assertion is unproved (scheme version unresolved for T^e, e >= 2).

Located concern

Remark after the proof (reduced variety), approx. line 1828

False statement, with a nonconstant homogeneous P: the reduced-variety version fails for 'all n >= 0'.

k=C, P=(T^2), r=(1), d=2, X(n) = {A : rank(A^2) <= 1} with reduced structure. Let c_2 = sum over i<j of (x_ii x_jj - x_ij x_ji).

Claim: for n >= 2 the quadrics vanishing on X(n) are exactly the multiples of c_2.

(a) c_2 vanishes on X(n). A^2 is invertible on the sum of generalized eigenspaces for nonzero eigenvalues, so rank(A^2) <= 1 forces at most one nonzero eigenvalue with multiplicity, and c_2(A) = e_2(eigenvalues) = 0.

(b) X(n) contains all rank <= 1 matrices. A quadric q vanishing on all u v^T lies in the kernel of Sym^2(U tensor V)^* -> Sym^2 U^* tensor Sym^2 V^*. That map is surjective, so the kernel has dimension C(n,2)^2 and is spanned by the linearly independent 2x2 minors m_{I,J}. Write q = sum a_{I,J} m_{I,J}.

(c) I and J disjoint: A = E_{ij} + E_{i'j'} has A^2=0 and its only nonzero 2x2 minor is m_{I,J}(A)=1, so a_{I,J}=0.

(d) I={a,b}, J={a,c}, with a,b,c distinct: A = E_{aa} + E_{bc} has A^2 = E_{aa} of rank 1 and its only nonzero minor is +-m_{I,J}, so a_{I,J}=0.

(e) Principal minors: B = [[1,1,0],[-1,0,1],[1,0,-1]] has B^2 = [[0,1,1],[0,-1,-1],[0,1,1]] of rank 1 and B^3=0. Its principal 2x2 minors on {1,2}, {1,3}, {2,3} are 1, -1, 0. Placing B on rows and columns (a,b,c) and zero elsewhere gives a point of X(n) with q = a_{ab} - a_{ac}. Hence all a_{I,I} are equal and q is a multiple of c_2.

For n <= 1, X(n) = Mat_n and no quadric vanishes. So dim O(X_red(n))^2 = 0, 1 for n = 0, 1 and n^2(n^2+1)/2 - 1 for n >= 2 (values 9, 44, 135, 324, 665, ...). The only polynomial agreeing with these for n >= 2 takes the values -1 and 0 at n = 0 and 1. The same argument works for P=(T^e), any e >= 2.

Restrict the remark to the determinantal case (radicals of FI#-ideals are FI#-ideals), or weaken to 'for n >> 0' for monomial P via the co-FI structure and th:noetherian.

Remark 5.3.3 is false in the stated generality; only the eventual statement survives for monomial P.

Located concern

Proof, approx. lines 1824-1826 ('finite type and slope <= 2 ... follows from thm:persinomial and thm:polyFIsharp')

Minor gap in the surviving determinantal case: for a field of positive characteristic, the degree bound 2d cannot come from 'slope', since weight is defined only in char 0.

de:weight is restricted to char 0. The needed input is that O(X)^d is generated in FI-degree <= 2d, which follows from pr:tensor applied to (M(1) tensor M(1))^{tensor d} and passage to quotients. Then H_0 vanishes above 2d and thm:polyFIsharp (iii)=>(vi) gives degree <= 2d.

Cite generation in degree <= 2d instead of slope when char k > 0.

None in the determinantal case.

Outside the audit

  • Proofs of th:noetherian, th:FIequiv, pr:stabdegstabrange, thm:persinomial, th:charFIsharp, thm:polyFIsharp, thm:fisubalgebra, pr:cofisubalgebra (used as stated; only interfaces checked)
  • External sources: [BS] Section 11.2 and Theorem 7; Chevalley; [CF] Thm 7.4; [AAB]; [ES]; [Wey] Cor 6.1.5(d); [DK]; [CEFN]. None supplied.
  • Example 5.1.A (Arnol'd algebra, arXiv-only) beyond noting it lies in the line range; its dependence on eq:dlmn and cor:subalgebraslope was not checked
  • Scheme-theoretic Hilbert functions of rank(A^e) <= r varieties for e >= 2 beyond degrees <= 2e (unresolved)
  • Sections 6-7 and all other parts of the paper; journal version
  • Preserved source-access qualification: CEF-v4-TeX (context read for interfaces, not fully audited: th:noetherian, th:FIequiv, thm:persinomial, pr:Mmuweight, pr:tensor, remark:FIsharpremarks, th:charFIsharp, thm:polyFIsharp, pr:gradedtensor, thm:fisubalgebra, pr:cofisubalgebra, def:tensorbullet, pr:Vtensorbullet)
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}
}