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-foundations

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 principal foundational results in this packet are supported by the manuscript and the indicated local repairs, conditional on the classical symmetric-group inputs. There are literal false assertions about unrestricted dimension polynomiality, a rank count, unrestricted self-duality, and finite-type subquotients without Noetherianity. The substantive gap in the printed representation-stability proof is cancellation of infinite multiplicities; it does not affect the finite-dimensional applications, and a new transition-block proof recovers the full stated scope and range. The arbitrary-ring FI-sharp growth argument needs a smaller repair: prove finite generation before choosing a residue field that sees a summand. The review used the recorded received-date cutoff 2013-07-14; bibliography dates are version-specific evidence at the precision shown, not inspection of external texts. The model's learned knowledge cannot be time-filtered directly; no later theorem is used to discharge a printed obligation, and new arguments are distinguished from citation checks. This is an assessment of CEF-v4-TeX in the assigned scope, not of the whole paper or its journal version.

F1

Checked statement: The FI and FB constructions, the adjunction defining M, representables, finite generation, the tensor-product bound, H_0, and the accompanying numerical and projectivity assertions in Section 2.

Returned assessment: claim_false

The foundational functors and generation criteria check out. For M, the component of the indexing category with image T has the inclusion T→S as a terminal representative; equivalently, a natural transformation M(W)→V is specified by its restrictions to the W_T at T, and on a summand W_T at S it must then be followed by V(T→S). This verifies the adjunction, its counit, and pointwise exactness directly. Taking the orbit of a size-a subset gives the stated induced representation, including over arbitrary k. An element in degree m specifies a map from M(m), with image the span of its injection images. For the tensor product, any pair of basis injections factors through the union of its images; this proves the degree bound, with the reduction for possibly infinitely generated modules made using arbitrary sums of representables rather than only finite sums. For H_0 the quotient precisely kills the maps from strictly smaller sets, which proves its adjunction. Lifts of generators of the quotient in each size generate by induction on size. Right exactness follows from the adjunction, and non-exactness can be seen by including the positive-size truncation of M(0) into M(0) and applying H_0 in size 1. The printed numerical errors do not affect these arguments. The projectivity assertion in the rendered arXiv remark also checks: M sends projectives to projectives by the adjunction with exact restriction. Conversely, P projective is a retract of a sum of representables, so H_0(P) is projective over FB, being a retract of a sum of the corresponding regular FB-modules. Split the FB surjection πP→H_0(P) and use adjunction to obtain M(H_0(P))→P inducing an isomorphism on H_0. This map is onto by the size induction, splits by projectivity of P, and has zero kernel because its split kernel has H_0=0. This is a direct verification, not an audit of the cited Lück or Weibel sources.

Premises used

  • An FI kernel or cokernel can be formed on each object; naturality gives its structure maps.
  • The basis in \label{def:Mm} is the set of ordered injections, not the set of images of injections.
  • The finite-dimensional split irreducible representation convention over characteristic-zero fields is the representation-theoretic input used here and below; it is not a finite-dimensionality hypothesis on every FI-module.

Missing inputs

Scope changes

  • The projective FI-module remark is included because \remove passes its argument through in CEF-v4-TeX; no comparison with the journal remark is made.

Located concern

599–804, \label{eq:MWinduced}+5: the sentence beginning ‘This shows that for any W’.

The conclusion that the dimension of M(W)_n is a single polynomial for arbitrary W is false, even if every value is finite-dimensional.

Take k=Q and W_a=Q with trivial action for every a≥0. The displayed dimension formula is correct and gives ∑_{a=0}^n binom(n,a)=2^n, which is not a polynomial on the nonnegative integers.

Keep the displayed dimension formula, but impose finite support and finite-dimensional W_a for the polynomial assertion. These are the relevant conditions in \label{thm:polyFIsharp}.

The unconditional dimension assertion is false; the adjunction, induced decomposition, and later finite-generation-based polynomiality are not contradicted.

Located concern

599–804, \label{pr:tensor}+5: the parenthetical rank in the first proof paragraph.

The rank of the tensor product of two representables is miscounted.

For m_1=2, m_2=0, n=2, there are two injections of a two-element set and one empty injection, so the rank is 2, not 1. In general the rank is (m_1! binom(n,m_1))(m_2! binom(n,m_2)).

Replace the rank formula by the product of the two injection counts. In the degree-only assertion explicitly allow possibly infinite sums in the reduction to representables.

Incorrect intermediate numerical claim; the subsequent support argument proves the proposition.

Located concern

599–804, \label{sec:free}, paragraph ‘Irreducible representations of S_n in characteristic 0’ preceding \label{def:Vlambdan}.

‘Every representation of S_n is self-dual’ overstates the finite-dimensional fact, given the definition of representation in this section and the allowance of infinite multiplicities in Section 3.3.

The trivial Q-representation ⊕{j≥1}Q has a countable underlying set, whereas its full algebraic dual ∏{j≥1}Q is uncountable. They are not isomorphic even after forgetting the action.

Say that every finite-dimensional representation is self-dual, and distinguish the finite-dimensional irreducibles from arbitrary direct sums.

The self-duality used for finite-dimensional characters is unaffected. The sentence cannot supply any finiteness assumption for Proposition 3.3.3.

F2

Checked statement: The coinvariant functors and stability degrees, the branching and weight calculations, the width bound, and the Noetherian proof over a Noetherian ring containing Q.

Returned assessment: repairable_gap

The transition in Φ_a descends because permutations of the unlabeled source extend over the new element, and choices of injection are conjugate by postcomposition on the target. Averaging identifies invariants and coinvariants when the orders are units; in particular Φ_a is exact over a Q-algebra. In the direct-sum computation for M(W), representative summands are indexed by a frozen subset U and a number b of unlabeled elements; passing to coinvariants within an orbit imposes exactly the relations from its stabilizer S_b×S_{n-b}. The second factor acts trivially. With standard inclusions old summands map identically and new summands first occur at n=b. Thus the injectivity, surjectivity, and quotient/subobject bounds work over their stated rings. For M(m), size a=0 detects the first nonzero regular coinvariant at n=m. Conditional on the printed horizontal-strip branching rule, removing boxes from the rightmost b of the nonempty columns always gives a partition for b≤μ_1. For b=μ_1 it gives the tail λ. In (iii), λ[n] adds one box in the first n−|λ| columns, and after removal the c=a−|λ| surviving added boxes can use at most λ_1+c columns. Hence n−a≥λ_1 is sufficient, including equality. Part (iv) uses n≥a. Induction by a horizontal strip gives weight ≤m for M(μ), with equality at μ[n] as soon as n≥m+μ_1. The weight tensor inequality can also be verified without a separate Kronecker-coefficient premise: each V(λ)_n, when defined, occurs in M(λ)_n; hence the tensor constituents are bounded by the weight of the pointwise tensor of the two M-modules, which is generated in degree at most |λ|+|μ| by \label{pr:tensor}. For the Noetherian proof, use the hypothesis Q⊂k from the theorem, in addition to Noetherianity. Φ_a(V) is generated over k[T] by the finitely generated k-modules in sizes through the stabilization bound. The relevant Hilbert basis step follows by stabilizing the ascending ideals of leading coefficients of polynomials of bounded degree in an ideal of k[T], then reducing leading terms with finitely many representatives; submodules of finite modules are then finitely generated. Lifting homogeneous generators of Φ_a(W) and adding generators below a gives the printed candidate W-tilde. Restriction to Q preserves the counit presentation and the direct sums defining M, and commutes with coinvariants because both are the quotient by the same difference relations. Thus one may finish entirely on the underlying Q[S_n]-modules: generation in degree ≤a bounds the weight of V and of W/W-tilde, so Φ_a(W/W-tilde)=0 detects every value in size n≥a. The smaller values are zero by construction. No finite Q-dimension of a finitely generated k-module is needed, nor is semisimplicity of k[S_n] asserted for a general k.

Premises used

  • Historical record for the classical symmetric-group input: available_from=1991 at year precision; date_evidence=\bibitem[FH]{FH} in CEF-v4-TeX; date_basis=publication year shown in supplied bibliography; version=Fulton–Harris, Representation Theory: A First Course, GTM 129 (1991). The exact input used is the characteristic-zero split partition parametrization and the horizontal-strip rule as stated in \label{lemma:branching}: induction with the trivial S_b-representation adds a horizontal b-strip, with multiplicity one; coinvariants of restriction remove such a strip. The latter follows from the former by reciprocity and averaging. The FH full text was not inspected.
  • Semisimplicity for characteristic-zero finite group actions on arbitrary-dimensional vector spaces follows by averaging a linear projection onto any subrepresentation. Consequently subquotients cannot introduce new constituents, including after restriction from k to Q.
  • \label{eq:MWdecomp}, \label{lemma:Mspan}, and \label{prop:charfg} supply the presentations and the pointwise finite generation needed in the proof of \label{th:noetherian}.

Missing inputs

  • The external source text for the classical representation-theoretic input is not supplied. The combinatorial consequences have been verified from the rule as reproduced in \label{lemma:branching}, not by checking the cited book.

Scope changes

  • The Noetherian verdict uses the rendered hypothesis in \label{th:noetherian}, not the commented-out restatement near the proof. It does not invoke the later arbitrary-Noetherian-ring result advertised in the introduction.

Located concern

805–1038, \label{pr:surjdeg}+7: rendered paragraph beginning ‘The converse of Proposition’.

The stated vanishing conditions imply the surjectivity-degree bound, but do not imply failure of generation below d+1 merely from V_{d+1}≠0.

Over Q take d=1. In size 1 take Q, and in size 2 take Q^2/Q(e_1+e_2) with the permutation action. Use the maps of M(1) followed by this quotient, and set all higher values and V_0 equal to zero. This is an FI quotient of M(1), is generated in size 1, and its size-2 value is nonzero sign with zero S_2-coinvariants. Conversely a module supported only at size 2 with that sign representation has the same vanishings and is not generated in size 1.

Specify the module supported only at size 2 over Q as the counterexample, or require the size-d+1 quotient by lower-size images to be nonzero.

The universal assertion about generation in the example is false. The proposition and the advertised failure of its converse are both correct.

Located concern

805–1038, \label{lem:coinvariants} proof, first paragraph following the statement; \label{pr:stabdegMlambda}, last paragraph of its deferred proof; \label{pr:narrow}, paragraph defining the transition with subscript m−s−1.

The proof of (i) writes only one direction explicitly; the sharp lower bound for M(λ) needs an existential choice of a; and the width proof invokes a negative subscript if m≤s.

For the converse in (i), subtract one bottom box from each of the rightmost b columns when b≤μ_1; column heights stay weakly decreasing. For M(λ), the factor with b=λ_1 need not occur for a given a: if λ=(1,1) and a=0, the transition at n=0 is between zero modules. For the width proof, μ_1≤m proves the claim immediately when m≤s; m−s−1 is a valid parameter only otherwise.

Insert the column argument; choose a=|λ|−λ_1 for the sharp lower bound when λ is nonempty; for the empty partition use λ_1=0 and the constant module directly. In \label{pr:narrow} handle infinite s vacuously and m≤s before using the transition. In (iv) make n≥a explicit.

These are proof and indexing repairs, not counterexamples to the stated bounds.

Located concern

805–1038, \label{pr:stabdegprojective}, last proof paragraph beginning ‘Finally, for M(m)’.

The claimed sharpness needs the usual nonzero coefficient-ring convention.

If the zero ring is admitted, M(m)=0 and its surjectivity degree is 0 for m>0, rather than m. The proof explicitly uses k≠0.

Specify nonzero k for the equality if the zero ring is included in the paper's ring convention; the upper bound needs no change.

Only a convention-sensitive endpoint; no effect on the Noetherian theorem for rings containing Q.

F3

Checked statement: Proposition 3.3.3 in its stated scope, the equivalence of finite generation with uniform representation stability plus pointwise finite dimension, and the existence, degree, uniqueness, and stable range of character polynomials.

Returned assessment: repairable_gap

Write s=stab-deg(V) and d=wt(V), both finite for a nonvacuous range. The factorizations of T prove Conditions I and II: injectivity of T detects the kernel after S_{n−d}-coinvariants, and surjectivity of T detects the cokernel of the adjoint induction map after S_{n+1−d}-coinvariants. This uses exactness and weight detection, and gives n≥s+d including equality. The sufficient consistency condition also checks: the complement group is generated by adjacent transpositions, each of which fixes the image by the two-step condition and equivariance of subsequent transitions. Consequently σ composed with a standard inclusion defines an injection map independently of the extension σ. For Condition III, the printed proof is valid for finite-dimensional values, but the printed proposition is not so restricted. New argument for the full scope: fix n≥s+d. For any constituent of tail λ at either endpoint, |λ|≤d and λ_1≤s, so the corresponding padded partitions are defined at both endpoints. Choose multiplicity spaces for these isotypic components, allowing dimension zero or infinite. Ordinary one-box restriction implies that a source of weight j cannot map to a target of smaller weight; at equal weight the only possible tail is the same λ, with branching multiplicity one. Use the transition of the FI-module V(λ) of \label{pr:Vlambda} as a nonzero reference map for this branching line. Its construction and stability bounds use only Sections 2, 3.1 and 3.2. The λ-to-λ block of the transition of V is therefore that reference map tensored with a linear map A_λ on the multiplicity spaces. Prove A_λ is an isomorphism by induction on j=|λ|. Apply the transition on Φ_j: weights >j vanish, and the direct sum of the weight-j summands is preserved. On the quotient consisting of smaller weights the map is triangular for the finite weight filtration. Its diagonal blocks are isomorphisms by induction and by stability of Φ_j(V(λ)) in unlabeled degree n−j≥s≥λ_1. The quotient transition is thus an isomorphism, as is the full transition by definition of s. Hence the map on the weight-j piece is an isomorphism. There its λ-block is A_λ tensored with the isomorphism on the nonzero S_j-module V_λ, which forces A_λ to be an isomorphism. This also starts the induction at j=0 with zero lower quotient. It proves the asserted range without cancellation of infinite multiplicities. Moreover, for finite-dimensional values this same block argument proves the monotonicity conclusion independently: the projection of the span of k copies of a λ-type at size n onto the λ-type at size n+1 contains k copies, and the target isotypic projection preserves any S_{n+1}-subrepresentation. For the finite-generation equivalence the printed proof only needs the finite-dimensional case: a finite free presentation with generator degrees ≤g and kernel generated in degrees ≤r gives surj-deg≤g and inj-deg≤r by lifting in the Φ_a diagram. The latter diagram uses injectivity on the free module and surjectivity on the kernel, uniformly in a. Thus the range n≥max(g,r)+g follows using wt(V)≤g. In reverse, Condition II alone generates from sizes ≤N, and pointwise finite dimension turns this into a finite set of generators. For the character theorem the stable decomposition is finite, and nonzero stable labels satisfy both inequalities |λ|≤d, λ_1≤s. This explains exactly why every constituent formula applies at n=s+d, not just beyond it. Uniqueness, including for agreement along an infinite subsequence of sizes, follows by prescribing any nonnegative integer tuple of short-cycle counts and padding with one cycle longer than the largest variable at a sufficiently large permitted size. The difference polynomial then vanishes on the whole nonnegative lattice. The same padding shows that P_V takes integer values there; traces are rational by split rational forms and integral because eigenvalues are roots of unity. Iterated finite differences give integer coefficients in the product binomial basis, so P_V is also integer-valued on the entire integer lattice. Specialization to the identity gives the claimed dimension bound.

Premises used

  • Historical record for the printed padded-character formula: available_from=1995 at year precision; date_evidence=\bibitem[Mac]{Mac} and \cite[Example I.7.14]{Mac}; date_basis=publication year in supplied bibliography; version=Macdonald, Symmetric Functions and Hall Polynomials, second edition. The exact input cited is that the character of V_{λ[n]} is one rational polynomial of weighted degree |λ| for all n≥|λ|+λ_1. No different edition was used and the external example was not inspected.
  • New independent verification of the padded-character input, conditional only on the printed Pieri rule and rational split characters: the trace of M(λ)_n is given for all n by the cycle-subset sum of \label{thm:polyFIsharp}, with weighted degree |λ|. For n≥|λ|+λ_1, the tails of the constituents of M(λ)_n are exactly the partitions ν obtained from λ by removing at most one box from each column. Indeed, deleting the first row from a partition obtained by adding a horizontal strip to λ gives these column heights; conversely ν[n] has first row of length at least λ_1 in this range, so it adds the required horizontal strip to λ. Each occurs once. Apart from ν=λ, these tails have smaller size and satisfy n≥|ν|+ν_1. Inductively subtract their character polynomials from the trace polynomial of M(λ). The term of degree |λ| cannot cancel, as evaluation at the identity has leading term dim(V_λ)·binom(n,|λ|). This is a new argument, not the printed appeal to Macdonald.
  • Historical record for the printed monotonicity interface: available_from=2012 at year precision; date_evidence=\bibitem[Ch]{Ch}, Invent. Math. 188 (2012), 465–504; date_basis=journal publication year recorded in TeX; version=cited 2012 journal article. The exact uses attributed to Theorem 2.8 and Proposition 2.3 are eventual monotonicity of the induced sequences M(i) and eventual monotonicity of their quotient when the kernel sequence is uniformly representation stable. These citation interfaces were not externally inspected; the new block argument does not use them.
  • The representation-stability definition used is the one reproduced in \label{definition:repstab}, not an imported theorem about a different convention for consistent sequences.

Missing inputs

  • The full texts of Macdonald Example I.7.14 and Church Theorem 2.8, Proposition 2.3, and Definition 1.2 were not supplied. The cited versions have pre-cutoff bibliography records, but the citation contents are not independently collated.

Scope changes

  • The new proof preserves the range n≥wt(V)+stab-deg(V) for finite bounds and arbitrary multiplicity spaces. An infinite bound yields no finite stable range.
  • Undefined padded labels have zero occurrence; for a label that occurs in the claimed range the width and weight inequalities ensure it is defined throughout that range.
  • The stated ranges are sufficient bounds, not assertions of an optimal character range for each module. For example, finitely generated FI-sharp modules have character polynomials at all sizes.

Located concern

1039–1150, \label{definition:repstab} (Condition III), \label{pr:stabdegstabrange}, and \label{eq:repstabcoinvariants}+3, the paragraph canceling the first summand.

Constancy of an entire representation and one direct summand does not imply constancy of a complementary summand with infinite multiplicities. The proposition expressly allows this scope.

Already for trivial representations, a countably infinite direct sum with a zero complement and the same sum with an extra one-dimensional complement are isomorphic; the two complements are not. This is a counterexample to the inference, not to the FI proposition. No finite-dimensionality hypothesis for the proposition can be recovered from \label{definition:repstab}.

Use the new transition-block proof above for the original scope. A shorter alternative is to restrict the proposition to pointwise finite dimension, where the printed cancellation is valid.

Missing proof in the unrestricted printed argument, not a false theorem. The finite-dimensional cases needed for both introductory theorems and the printed character theorem are unaffected. Restricting Proposition 3.3.3 alone would also require qualifying any unrestricted consumer, notably \label{co:fisharprange}.

F4

Checked statement: The FI-modules V(λ), Murnaghan and Schur stability, the FI-sharp equivalence and growth results, and finite-type graded FI/co-FI constructions including the variable tensor product.

Returned assessment: claim_false

The main stated theorems in this block survive the identified issues. For V(λ), the only constituent of M(λ)n of weight |λ| is λ[n] when n≥|λ|+λ_1. Restricting any other constituent to S_m cannot increase its number of boxes below the first row: iterate the one-box restriction rule. This proves preservation for a standard injection and hence for any injection by relabeling. The subset direct sum makes the transition injective and the image from the first nonzero size spans the target irreducible under S_n. This gives finite generation in exactly that first size. For λ nonempty, the Φ|λ| transition out of unlabeled size λ_1−1 detects the sharp lower bound, while the quotient from M(λ[|λ|+λ_1]) gives the matching upper bound. For λ empty use the constant module directly. Murnaghan follows from pointwise tensor and finite generation, with a finite set of stable labels by the weight bound and finite multiplicities. In the Schur proof a normalized Young symmetrizer (or just its image construction) gives a natural quotient of the tensor power; hence the finite-generation and degree/weight assertions follow. The empty partition gives the constant tensor unit. The two stability applications use finite-dimensional values, and the monotonicity application has the interface discussed under F3. For FI-sharp, I_C I_D=I_{C∩D}. The minimum-size induction in the printed proof makes the lower-size terms zero, so the summands of the projector are orthogonal. Functoriality follows by matching C with its image under the partial injection on precisely the surviving terms. The maps from the size-n summands supply the image of the projector; its kernel vanishes through size n. Iteration is finite at each object S, and applying H_0 identifies the residual generating FB-module at each size. This proves the natural equivalence over arbitrary k without semisimplicity. In \label{thm:polyFIsharp}, the equivalence preserves directed unions by the subset decomposition, and finite generation in FB means finite support with finitely generated k[S_i]-modules, equivalently finitely generated k-modules, in each size. Conversely polynomial generator growth first gives finiteness of each W_i by taking it as a quotient of sufficiently large values. A nonzero finitely generated W_i has a simple quotient k/𝔪: take a maximal proper submodule, which exists since a chain of proper submodules cannot exhaust a finitely generated module. Hence dim_{k/𝔪}(W_i⊗k/𝔪)≥1, and taking the fiber of the direct summand W_i^{⊕binom(n,i)} forces i≤d. This supplies the missing order of justification for the fiber argument. In the field case the binomial dimension sum is exact for every n, including n=0. In characteristic zero the trace formula counts invariant subsets by choosing cycles and is likewise exact for every n and every σ. A support bound i≤d gives weighted character degree ≤d; a nonzero highest W_i contributes the highest dimension degree at the identity. For \label{co:fisharprange}, Φ_a commutes with direct sums, so the inequality stab-deg≤wt holds even for infinite multiplicity spaces at finite weight; obtaining all of uniform stability then uses the repaired F3 result. The arXiv projective-resolution remark can use finitely generated terms even if the given finite resolution does not: choose successive finite free surjections using \label{th:noetherian}. Comparing two projective surjections onto an object by their pullback identifies the two kernels after adding the other projective summand; iterating this comparison shows the last kernel of the finite-free construction is projective when a finite projective resolution of that length exists. Then character additivity gives all-size polynomiality. The arXiv tensor examples follow by sorting pairs by union of supports: for two 2-subsets the exact-support S_2, S_3, S_4 representations are respectively trivial, regular, and the permutation representation on 2-subsets of a four-element set. Pieri gives the stated S_4 decomposition. The same support argument gives the inequalities in \label{eq:Kronecker}, ordinary Kronecker multiplicities for equal support sizes, and induction from S_|λ|×S_|μ| on disjoint supports. The latter is the usual induction description of the Littlewood–Richardson coefficient. For the graded claims, each total degree of a tensor product uses finitely many bidegrees. In the free algebra each positive-degree word has length at most its total grading, so only finitely many shapes and tensor powers contribute. Dualization in the co-FI argument is degreewise; finite type of W* forces W_S^i to be finite-dimensional, so the duals of the finite sums and tensor products at a fixed grading identify as needed. Finite type passes to the dual submodule using Noetherianity over the stated characteristic-zero field, and slope passes by the weight bound on subquotients. For the variable tensor product, naturality of the unit and augmentation, and augmentation composed with unit equal to the identity, verify composition in \label{eq:tensorbullet}, including insertion followed by deletion. In \label{pr:Vtensorbullet} choose the free cover to be an isomorphism in degree zero. A basis tensor in grading i is determined on the union of its positive-degree factor positions and all injection images attached to those factors. The bound in the proof is precisely a bound on that union. There are finitely many positive-degree multisets at fixed i, since each factor has positive grading and each grading of the cover has finitely many free generators; this proves finite type over arbitrary k.

Premises used

  • The one-box restriction rule needed in \label{pr:Vlambda} follows by iterating \label{lemma:branching}(ii) with b=1; it does not require an additional restriction theorem from an unsupplied exercise.
  • Historical record for the Schur construction and the classical identification of induction coefficients: available_from=1991 at year precision; date_evidence=\bibitem[FH]{FH} and the Lecture 6 citation in Section 3.4; date_basis=publication year recorded in CEF-v4-TeX; version=Fulton–Harris, GTM 129 (1991). The input is the characteristic-zero construction of the Schur functor as the image of a Young symmetrizer on the |λ|-fold tensor power, and the classical identification of the outer-product induction multiplicities with the Schur product coefficients. The book is not supplied. The exact-support induction calculation itself does not need that identification.
  • The ring hypothesis needed for the finite-type submodule step is \label{th:noetherian}'s Noetherian Q-algebra hypothesis, automatically satisfied by a characteristic-zero field. No arbitrary-ring Noetherian theorem is used.
  • The growth proof uses right exactness of tensoring only to preserve a surjection and uses the subset decomposition to preserve a split summand. It does not assume tensoring with a residue field preserves arbitrary injections.

Missing inputs

  • The external representation-theory sources are not supplied; the review of the core FI arguments is conditional on the stated classical inputs. The historical priority claim about the first complete proof of Murnaghan's theorem is not determined by the supplied citation alone.

Scope changes

  • The all-size FI-sharp dimension and character conclusions are kept distinct from eventual polynomiality for a general FI-module.
  • The rendered arXiv projective and tensor additions are checked as such. The displayed tensor products are understood for the finite-dimensional irreducibles over characteristic-zero fields; the earlier FI-sharp classification is ring-general.

Located concern

1151–1648, \label{ss:graded}, paragraph immediately following the definition ‘Graded FI-modules of finite type’, sentence invoking \ref{th:noetherian}.

The prose drops Noetherianity in asserting that finite type passes to subquotients when k contains Q.

Let k=Q[x_1,x_2,…], and put the constant module M(0)=k in grading 0. The constant submodule with value I=(x_1,x_2,…) is not of finite type. Finitely many proposed ideal generators involve only finitely many variables; setting those variables to zero kills the generators but not a variable outside that finite set. Constant FI maps cannot add k-module generators.

Require that k be a Noetherian ring containing Q in this sentence. The analogous introductory shorthand before \label{th:noetherian} also needs that qualification if read as a standalone assertion.

The unqualified subquotient assertion is false. Neither the stated Noetherian theorem nor the co-FI proposition over a characteristic-zero field is contradicted.

Located concern

1151–1648, \label{thm:polyFIsharp}+20–21, the paragraphs beginning ‘Let c_i’ and ‘From’.

The quantifier describing (iii) is reversed for the finiteness of the lower values, and the choice of a nonzero residue fiber in (iv)⇒(iii) is made before establishing finite generation of W_i.

The displayed ‘i≥d’ condition allows W_0 to be infinitely generated when d=1. A nonzero arbitrary module can also have zero residue fibers, for example Q as a Z-module. The latter is not a counterexample to (iv)⇒(iii): (iv) itself makes each W_i finitely generated, since it is a quotient of V_n for arbitrarily large n≥i by \label{eq:MWdecomp}. This also handles asymptotic big-O with no bound initially imposed at n=0.

Require c_i<∞ for i≤d and c_i=0 for i>d. In the reverse direction first use the quotient to prove finite generation in every size, then use the simple-quotient argument to get a nonzero fiber and apply the binomial lower bound.

The arbitrary-ring equivalence and the degree bound remain true; no Noetherian hypothesis must be added.

Located concern

1151–1648, \label{pr:Vlambda}, last proof sentence involving Φ_|λ| in degree λ_1−1; \label{pr:schurfunctor}, first proof paragraph.

The sharpness argument has an undefined negative degree for the empty partition. The idempotency assertion for a Young symmetrizer needs normalization if the term denotes the ordinary row-sum times column-alternating-sum element.

For the empty partition the natural first-row convention is λ_1=0 and V(λ)=M(0), with stability degree 0. For the shape (2) the unnormalized symmetrizer 1+(12) squares to twice itself, not to itself.

Handle the empty partition directly and say ‘normalized Young symmetrizer’ when requiring an idempotent. The proof only needs the image to be a natural quotient.

Convention and endpoint repairs only; the stated construction and stability corollaries hold.

Located concern

1151–1648, \label{remark:FIsharpremarks}, paragraph on endomorphisms; remark immediately after \label{th:charFIsharp}, paragraph beginning ‘Since FB is a full subcategory’.

The endomorphisms form a monoid, not an algebra before linearization. The comparisons of categories incorrectly use ‘full’.

End_FI-sharp([n]) consists of partial bijections under composition. Its linearization acts on V_n, with no requirement that the empty partial map act by zero: it acts by the identity on M(0). In FS there is a surjection [2]→[1] absent from FB. In the category of based sets and all based maps there is a based map taking two distinct nonbasepoint elements to the same nonbasepoint element, absent from FI-sharp. The stated subcategories have all objects, so these omissions contradict fullness.

Say rook monoid, or its uncontracted monoid algebra when linearizing, and replace ‘full’ by ‘wide’ for the indicated subcategories. For partial injections specify injectivity on the inverse image of the nonbasepoint part of the target.

Errors in explanatory statements, not in the direct proof of the FI-sharp classification or an occasion to import Pirashvili's theorem.

Located concern

1151–1648, \label{thm:fisubalgebra}, first proof paragraph and displayed list after the proof; definition of unital graded algebra at \label{remark:FIalgebras} (599–804); \label{pr:cofisubalgebra}, final hypothesis.

The parenthesization count is misindexed in the usual Catalan convention; the stated unital definition does not literally include the free Lie algebra under its bracket; the co-FI slope clause refers to V rather than the generator W.

A product of three factors has two parenthesizations, C_2 rather than C_3=5. The tensor unit is a separate zero-factor case. In a nonzero Lie algebra a two-sided bracket unit e would satisfy [e,e]=e but [e,e]=0, impossible. The statement of the co-FI proposition only introduces A and W.

Use C_{j−1} for j≥1 and one zero-factor term in the unital case. For Lie algebras use the same proof for the nonunital free algebra, omitting the zero-factor term, and take its Lie quotient. Replace V by W in the co-FI slope clause.

The finite-type and slope estimates are unchanged. The claimed applicability to Lie algebras needs the stated nonunital variant rather than a literal application of a unital definition.

Outside the audit

  • Journal full text and differences from the final-author arXiv version; the recorded DOI evidence is received-date metadata only.
  • External reference texts and claims of historical priority. In particular, the bibliography is not an audit of the cited book, journal, or preprint versions.
  • The application sections beyond the packet ranges and the external alternative Noetherian and categorical-equivalence routes mentioned in the introduction.
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}
}