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-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's diagonal-coinvariant theorem survives the audit of its functorial and finite-type interfaces. The proof needs minor corrections, and the rendered Arnol'd example contains an invalid inference about individual summands that can be replaced by the support calculation. A2's co-FI construction, diagonal action, and stability argument check out with the dual-sequence interpretation and the stated conditional identification of the closure. A3 is false in the stated generality, separately for the minors scheme and for the reduced conclusion. The polynomial-rank maps also fail independently of the conclusion, including for a homogeneous polynomial with zero constant term. The historical cutoff is 2013-07-14 on the supplied journal-received metadata. The dated records above come only from the supplied TeX or metadata; the underlying external versions were not inspected. Later knowledge cannot be excluded from model training, but no later result is used to close a proof obligation here. This is not a whole-paper verdict or a comparison with the journal full text.

A1

Checked statement: Section 5.1: the FI-algebra from a symmetric presentation; the rendered Arnol'd example; and the co-FI construction, finite type, and character assertions for diagonal coinvariants, including Theorem 1.11.

Returned assessment: repairable_gap

For a symmetric presentation, an injection relabels a generator indexed by an injection g to the generator indexed by the composite. Thus the ideals are covariant, and the homogeneous quotients are FI-algebras. In grading i the ambient polynomial FI-module is generated by finitely many monomials on at most i labels, so this finite-type assertion needs no restriction on the cardinality of the family of relations. The nilpotent example is a special case; its ideal is even preserved by partial deletions. The subsequent coinvariant construction correctly uses the other direction: for f:S→T, restriction on polynomial rings sets variables outside f(S) to zero and relabels the rest by f⁻¹. The extension of a permutation of S by the identity on the complement makes restriction equivariant for that permutation. Consequently invariants restrict to invariants, and grading preservation takes positive-degree invariants and their generated ideal to the corresponding ideal. For the ordinary coinvariant ideal, failure in the covariant direction really does follow in degree one: the degree-one ideal at n+1 is the line along the sum of all n+1 variables, not the sum of only the first n, for n≥1.

For each multidegree J, the dual ambient co-FI polynomial piece has the dual monomial basis. Its covariant maps relabel that basis, and each monomial uses at most |J| distinct labels. It is therefore finitely generated in degree at most |J|. The invariant ideal is multihomogeneous: the multihomogeneous components of an invariant are invariant, and the ideal is generated by such components of positive total degree. Dualizing the quotient in multidegree J thus gives an FI submodule of the ambient dual. Over a characteristic-zero field the printed Noetherian theorem gives finite generation and the ambient weight bound passes to this submodule. This argument works over the field k of the introductory setup, not only over Q as written in the proof. Equivalently, the Q proof extends by scalar extension since averaging commutes with that extension. At each n the dual character has value χ(σ⁻¹)=χ(σ), since a permutation and its inverse are conjugate. The range delivered by thm:persinomial is n≥stab-deg((R_J^(r))* )+wt((R_J^(r))* ); Theorem 1.11 itself does not promise a numerical bound on the first term. Substitution of (n,0,…) yields the asserted degree bound on dimension.

The low-degree checks agree with eq:coinvariantexamples. Write E=k^n and e=Σx_j. In degree one the quotient is E/ke for n≥1. In single-color degree two, the subspace eE has character X₁, and the invariant Σ_{a<b}x_ax_b gives one further independent relation modulo eE for n≥2: comparing coefficients of squares rules out its membership in eE. Subtract these from the character of Sym²E. In bidegree (1,1) the linear relations first give (E/ke)⊗(E/ke); the remaining invariant is the image of the diagonal tensor. It is nonzero for n≥2, as evaluation on (x₁*−x₂*) in both factors gives 2. The resulting character is (X₁−1)²−1. These checks do not require the formulas to hold below their stated ranges.

For the rendered Arnol'd example, conditional on the cited presentation, the degree-two calculation checks directly by support. On three labels any pair of distinct edges meets. Its wedge space is the exterior square of the three-dimensional edge permutation representation, namely a sign line plus the two-dimensional standard representation. The printed relation spans the sign line. On four labels, a basis up to signs is given by ordered complementary edges with reversal acting by −1. The traces for cycle types 1⁴, 2·1², 2², 3·1, 4 are respectively 3, 1, −1, 0, −1, the traces of the standard S₄ representation. There are no other supports in grading two. Partial deletions kill a term unless its full support survives. This gives eq:h2conf1 and, by counting invariant supports, the introductory character polynomial for every n, including n<3. The degree bound 2i for FI generation follows by using at most 2i labels in a product of i edges; with the printed weight and FI# range results this gives n≥4i. In grading two the sharper n≥7 in the introduction follows from the horizontal-strip decompositions of M(2,1) and M(3,1), or from their separate stability degrees 2 and 3.

Premises used

  • pr:coinvariantfinite follows by dualizing each graded quotient over a field and applying th:noetherian to the finite-type dual when k has characteristic zero. Weight decreases under subquotients; pr:fgweight applies to the ambient dual generated in degree ≤|J|.
  • thm:persinomial is used as a manuscript prerequisite: finitely generated FI-module over a characteristic-zero field, weighted degree at most weight, valid from stability degree plus weight. Its proof cites the padded irreducible character-polynomial statement: for every partition λ and n≥|λ|+λ₁, χ of V(λ)_n is one rational polynomial of weighted degree |λ|. Historical record for that dependency: available_from=1995 at year precision; date_evidence=Mac bibliography entry and proof citation to Example I.7.14; date_basis=edition publication year in supplied TeX; version=Macdonald, second edition. The external example was not inspected.
  • For the Arnold calculation the imported input is the presentation of rational cohomology of the ordered configuration space in the plane by degree-one pair classes with the three-label quadratic relations in eq:arnoldideal. Historical record: available_from=1969 at year precision; date_evidence=Ar entry, Mathematical Notes 5(2), 138–140; date_basis=journal publication year in supplied TeX; version=the cited Mathematical Notes publication. Its content was not externally inspected.
  • The representation-theoretic range checks use the horizontal-strip induction rule stated in the manuscript in lemma:branching for S_m×S_j induced from an irreducible over Q and a trivial factor, with extension to characteristic-zero fields as stated there. Historical record for the cited source: available_from=1991 at year precision; date_evidence=FH entry; date_basis=book publication year in supplied TeX; version=Fulton–Harris, GTM 129. The external text was not inspected.

Missing inputs

  • The cited external results underlying the manuscript prerequisites were not independently collated. The main diagonal-coinvariant inference is conditional on those manuscript prerequisites, and the Arnol'd presentation is a separate conditional input.

Scope changes

  • For ordinary and diagonal coinvariants the asserted natural stability is for the dual FI-module; character and dimension conclusions concern the original representations as well.
  • example:arnoldalgebra was checked because the supplied preamble renders it. This is not a finding about the journal version.

Located concern

CEF-v4-TeX lines 1723–1733, paragraph before def:generalcoinvariantalgebra.

The displayed inclusion for invariants has the wrong right-hand side: f_* is applied there to a subspace of its target rather than its source.

For f:T→T′ the co-FI map has source W_T′ and target W_T. The permutation-extension identity in the immediately preceding sentence gives f_*(W_T′^inv)⊆W_T^inv.

Correct the formula to that inclusion. Also distinguish failure to be invariant from failure to lie in the generated ideal in the preceding ordinary-coinvariant example; its degree-one calculation proves the latter.

Typographical error in the invariants formula and a short missing justification in the example; neither invalidates the co-FI quotient.

Located concern

CEF-v4-TeX lines 1735–1748, classical coinvariant example and proof of theorem:intro:polynomial:answer.

The classical example attributes uniform representation stability to the co-FI pieces without specifying forward maps. The diagonal proof also writes two bounds as ≤J instead of ≤|J|.

th:FIequiv applies to a covariant FI-module with forward maps and finite-dimensional terms. Here the quotient maps go backward; the dual goes forward and has objectwise the same character. J is a tuple, while the total grading used in the proof is |J|.

State stability for the dual sequence, or specify equivariant identifications that transport the dual's forward maps to the original spaces. Write |J| in both bounds and describe the character as polynomial in cycle counts, with the specialization at the identity polynomial in n.

No change to the character or dimension theorem; the repair specifies the stable sequence to which the definition applies.

Located concern

CEF-v4-TeX lines 1668–1715, example:arnoldalgebra, especially the calculation using eq:dlmn.

An FI# automorphism preserves the isotypic component containing two copies of M(2,1), not necessarily each copy. The printed assertion about individual summands does not justify the next assertion that the copies are exchanged.

By the manuscript's own classification th:charFIsharp, an invertible 2×2 scalar matrix can act on the multiplicity space for two copies of the same irreducible FB-module, so neither displayed copy is canonical. The support calculation in the argument computes the antisymmetric eigenspace without choosing these two copies.

Use the support calculation above as a new elementary replacement for the summand argument. For the geometric maps on the pair classes, in the planar collar construction one can compress all retained points into a disk and put new points outside the closed disk. A pair with one new point has a difference vector varying in a disk avoiding zero, so its angular class pulls back to zero; two new points give a constant. Two retained points give the relabelled class by isotopy. This verifies the partial-deletion action on generators used in the calculation.

The asserted all-n grading-two decomposition and character survive; the invalid inference about individual copies is unnecessary.

A2

Checked statement: Section 5.2: the co-FI-algebra modelling the Bhargava–Satriano closure, the diagonal action, finite type of the dual of each graded piece, and the claimed stability and grading-one description.

Returned assessment: no_issue_found

Use labels {}∪T for the tensor factors and T for the square-zero variable labels. For f:S→T the restriction sends x_{t,u} to x_{s,v} when t=f(s) and either u=f(v) or u=v=; otherwise it sends it to zero. Here x_{t,u} means the variable x_t in factor u. This formula preserves the same-factor square-zero relations, is unital and graded, and composes exactly by testing survival of the two labels. For a bijection both ordinary labels are relabelled, with * fixed, which gives the asserted diagonal action under the usual inverse convention for co-FI representations. The sum γ_T(x_t)=Σ_{u∈{*}∪T}x_{t,u} maps to γ_S(x_s) if t=f(s), and to zero otherwise. Thus the generated ideals are a co-FI-ideal, including at the empty set where the ambient algebra is k and the ideal is zero.

There is also a direct check of the required square-zero closure presentation from the defining characteristic-polynomial relations of the closure. This is a new calculation, conditional only on that definition. If a∈k and u is in the square-zero ideal, multiplication by a+u is triangular with all diagonal entries a in the basis 1,x_t. Put N=|T|+1. Its characteristic polynomial is (Z−a)^N. The difference between e_j of the N copies of a+u and the corresponding characteristic coefficient (using unsigned coefficients) is a linear combination of the e_l of the copies of u for l≥1, by expanding e_j. Each such e_l equals (Σ_u′ u^{(u′)})^l/l!, because each individual copy squares to zero. These differences lie in the ideal generated by the γ(x_t), and the degree-one characteristic relations on x_t give precisely those generators. This verifies the asserted ideal in characteristic zero without needing a separate presentation theorem.

Each ambient grading has a finite monomial basis formed by choosing distinct factors, with at most one variable in each chosen factor. The dual basis is covariantly relabelled. A grading-i basis vector uses at most i ordinary factor labels and i variable labels; the distinguished factor uses no ordinary label. Thus the ambient dual is finitely generated in degree ≤2i, an independent check of the printed use of pr:Vtensorbullet and pr:gradedtensor. Dualizing the quotient embeds its graded dual into this finite-type dual. The characteristic-zero hypothesis then permits the stated Noetherian and representation-stability results. In grading one, the γ relations eliminate exactly the distinguished-factor variables. The surviving ordinary pairs form a co-FI basis with deletion maps. Its dual is M(1)⊗M(1): equal pairs supply M(1), and unequal pairs supply M(k[S₂]), which splits into its trivial and sign summands. This checks both the maps and the claimed small grading.

Premises used

  • The closure definition used in the independent presentation calculation is the quotient by e_j of the copies of each element minus the corresponding unsigned coefficient of its multiplication characteristic polynomial, for all elements and 1≤j≤dim_k R. Historical record: available_from=unresolved for the exact version containing the defining formula or §11.2; date_evidence=BS bibliography entry gives arXiv:1006.2562, identifying a June 2010 preprint series, without a version suffix; date_basis=arXiv identifier and supplied bibliography, not an inspected version; version=unspecified preprint version. The calculation does not assume any later presentation theorem.
  • def:tensorbullet supplies the co-FI maps by deletion with the augmentation in deleted factors. The ambient grading is finite-dimensional objectwise. th:noetherian applies to its graded dual over any characteristic-zero field, and th:FIequiv applies to the resulting finitely generated FI-module.
  • The last isomorphisms in the printed proof are valid at the level of S_n-representations. Uniform representation stability is a statement about the dual's forward maps or the forward sequence transported along equivariant self-duality isomorphisms. It is not a stability assertion for the backward restriction maps regarded as forward maps.

Missing inputs

  • A dated Bhargava–Satriano version containing the defining relations and the cited §11.2 was not supplied. The identification with the named closure is conditional on the defining formula specified above; the quotient and its functoriality were checked independently.

Scope changes

  • No numerical uniform stable range is asserted by this proposition. The ambient degree bound does not, by itself, bound the generation degree of its submodule obtained by dualizing the quotient.
  • The final question whether the algebra itself extends to FI# is not a claimed result and is not decided here.

A3

Checked statement: Section 5.3: the grading and FI# maps for arbitrary polynomial rank conditions, all-n polynomial dimension and character in Theorem 5.3.2, and the separate reduced-scheme assertion of Remark 5.3.3.

Returned assessment: claim_false

There are separate failures of the grading, of the proposed partial-injection maps, and of the conclusion. The constant-polynomial counterexample below is already a homogeneous scheme counterexample. There is also a nonconstant homogeneous counterexample for the reduced conclusion, so requiring p(0)=0 and an honest grading would not suffice to save that conclusion.

For the reduced counterexample work over Q, take p(T)=T², r=1, and d=2. Write J_n for the ideal of 2×2 minors of A². I compute the degree-two part of √J_n for n≥2. First, any such quadric vanishes on every rank-one matrix, since rank A²≤rank A. Substitute x_ij=u_iv_j. The resulting monomials are indexed by their unordered row pair and column pair, with repetitions allowed. A repeated row or column occurs in only one input monomial for these pairs; for distinct rows and columns the two possible monomials differ by the corresponding 2×2 minor. Hence every quadric in √J_n is a linear combination of the minors of A, or equivalently L(∧²A) for a linear functional L on End(∧²Qⁿ).

Write L(w⊗η)=η(Dw) for a linear map D on ∧²Qⁿ. For independent u₁,u₂, extend them to a basis and let w=u₁∧u₂. Take η=v₁∧v₂ to be any wedge of two distinct dual basis vectors other than the pair dual to u₁,u₂. The matrix A=u₁v₁+u₂v₂ has ∧²A=w⊗η, and A² factors through the 2×2 matrix (v_i(u_j)), which has determinant zero. Thus L(w⊗η)=0. All these η together imply Dw is proportional to w. Every decomposable w is therefore an eigenvector of D. On the wedges of a fixed basis, the eigenvalues agree for any pairs sharing one index, by applying the same property to the sum of their wedges with consistent signs. The graph on pairs with this adjacency is connected for n≥3, and n=2 is one-dimensional. So D is scalar, and the only possible quadric is a multiple of tr(∧²A).

Conversely tr(∧²A) lies in √J_n. Over the fraction field of the quotient by any prime containing J_n, triangularize A after passing to an algebraic closure. Since A² has rank at most one, at most one diagonal eigenvalue of A is nonzero: two nonzero entries on the diagonal of a triangular A² give a nonzero principal 2×2 minor in that basis. Therefore tr(∧²A)=0 in that field. It belongs to every prime over J_n and hence to the radical. For n≥2 it is a nonzero polynomial, so the dimension of the reduced degree-two quotient is binom(n²+1,2)−1 for every n≥2. At n=1 it is 1, because the rank condition is vacuous. The unique polynomial agreeing for n≥2 gives 0 at n=1. This refutes the all-n reduced statement even with a positive-degree homogeneous p. It does not refute the scheme statement for this particular p and d: J_n is generated in grading four so the scheme has the full ambient degree-two space.

A restricted theorem does follow for p_ℓ(T)=a_ℓT, including zero polynomials. Principal restriction followed by zero padding sends each defining minor to a minor on retained labels, up to the relevant scalar and sign, or to zero. The associated coordinate ring maps therefore descend, compose, and preserve grading; they descend to radicals as well, since a ring map carrying I into J carries √I into √J. The radical of a homogeneous ideal is homogeneous: in the quotient, the highest-degree component of a nilpotent element is nilpotent, and subtracting it and iterating proves the assertion. The exact all-n polynomial bound for this restricted family can be seen directly. In degree d the ambient monomials have support of size ≤2d. The partial identity on a subset B induces a projection E_B retaining exactly the monomials supported in B. For each U the operator Σ_{B⊆U}(−1)^{|U|−|B|}E_B projects onto exact support U and descends to both quotients. These projections split the degree-d quotient into pieces on subsets U, and inclusion and restriction identify a piece naturally with the full-support piece on U. Dimensions therefore sum as Σ_{j≤2d} c_j binom(n,j). In characteristic zero only invariant supports contribute to the trace; the number of supports of a given cycle type α is the product of binomial polynomials choosing its cycles. Each term has weighted degree |α|≤2d. This is a new proof for the restricted class, not a proof of the printed statement for arbitrary P.

Premises used

  • The scheme is the one defined by the specified minors ideal; the reduced scheme is obtained by taking that ideal's radical. These are the conventions stated in subsection:determinantal and its closing remark.
  • The counterexamples and the restricted repair use elementary matrix and polynomial calculations, not Eisenbud–Saltman or later results. The reduced counterexample works with geometric points and also gives the stated obstruction if one uses a rational-point vanishing ideal over Q.

Missing inputs

  • None needed for the refutations. No assertion is made that the restricted linear-polynomial repair is the maximal valid subfamily.

Scope changes

  • The all-n repair is restricted to scalar multiples of T. It is not a repair of the theorem for arbitrary P.
  • The homogeneous square-power example separates scheme and reduced behavior in the same degree. No classification of which other families have eventual polynomiality is claimed.

Located concern

CEF-v4-TeX lines 1794–1807, subsection:determinantal, grading assertion before question:rankvar.

The minors need not be homogeneous, and their quotient need not inherit the claimed grading.

Take k=Q, p(T)=T−1, r=1, n=2. The ideal is generated by det(A−I₂)=det A−tr A+1. It is proper by evaluation at A=I₂. A homogeneous ideal containing this polynomial must contain its constant component 1, so this proper ideal is not homogeneous.

Require conditions that actually imply a homogeneous ideal if retaining the stated graded question. A filtered formulation would be a different assertion and would need a new proof.

For allowed inputs the degree-d component in the statement can be undefined.

Located concern

CEF-v4-TeX lines 1819–1823, theorem:sections:polygrowth proof, ‘The rank varieties ... are preserved by these maps’.

The proposed FI# maps do not preserve polynomial rank conditions in general, even for homogeneous p with p(0)=0.

Take k=Q, p(T)=T², r=1, and the 4×4 matrix with 2×2 blocks [I I; −I −I]. Its square is zero. The partial injection from four labels to two keeping only the first two labels sends it to I₂, whose square has rank two. This is a pointwise failure, hence also a failure for the stated scheme maps and their reductions. The zero-padding part can separately fail when p(0)≠0: with p(T)=T−1 and r=1, padding I₂ by two zero labels gives p of the padded matrix of rank two.

Replace the preservation step by an actual check for a restricted class, such as p_ℓ(T)=a_ℓT. Checking total injections with zero padding when p_ℓ(0)=0 is not enough to obtain FI#: principal restriction must also be checked.

Neither the ambient FI# grading nor the cited FI# classification gives all-n polynomiality for the printed rank conditions.

Located concern

CEF-v4-TeX lines 1811–1818, theorem:sections:polygrowth, and the reduced remark at lines 1829–1833.

The all-n conclusion for the specified scheme is false even when the quotient has the stipulated grading.

Take k=Q, one polynomial p(T)=1, r=1, and d=1, all permitted by the statement. For n≤1 there are no 2×2 minors and the ring is the ambient polynomial ring; at n=1 its degree-one dimension is 1. For every n≥2 the matrix p(A) is I_n and a principal 2×2 minor is 1, so the ring is zero. No polynomial is zero at every n≥2 and 1 at n=1. Both ideals, zero and unit, are already radical. At the identity the character is the dimension, so the all-n character assertion fails too.

Narrow the theorem's hypotheses. Correcting the proof without changing the stated generality cannot remove this counterexample.

Refutes the actual scheme theorem and its character consequence, not just the claimed functoriality. It also refutes the reduced remark as stated.

Located concern

CEF-v4-TeX lines 1829–1833, unlabelled reduced-variety remark after theorem:sections:polygrowth.

Taking radicals does not restore all-n polynomiality even on the nonconstant homogeneous square-power family.

For k=Q, p(T)=T², r=1, d=2 the independent computation above gives reduced dimension binom(n²+1,2)−1 for n≥2 and dimension 1 at n=1. The scheme has no degree-two relations for the same input.

State a separate reduced conclusion under hypotheses that yield the coordinate-ring maps and the support decomposition, such as the linear-monomial restriction proved above.

An independent failure of the reduced claim that does not rely on empty rank varieties or a nongraded quotient.

Outside the audit

  • The journal full text, other arXiv versions, and the content of unsupplied external publications.
  • Independent verification of the foundational manuscript theorems used conditionally, beyond the stated interfaces.
  • Historical priority claims, unrelated applications, and the open FI# question at the end of §5.2.
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}
}