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-rank-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 unrestricted rank application has both ill-defined cases and false, well-defined instances. The false preservation assertion explains the failure of the proof but is not the numerical counterexample. For the same homogeneous test p=T², r=1 over C, the radical quotient fails all-n polynomiality in degree two and the minor quotient fails it in degree five. The latter conclusion follows here from a linear-relation proof, not trust in the reported table. Ordinary determinantal conditions retain the all-n result; homogeneous positive monomials retain an eventual dimension result.

REX1

Checked statement: For the stated polynomial tuples and positive rank bounds, the minor quotient inherits the standard grading, the indicated partial injections preserve the rank schemes, and the degree-d dimensions are polynomial for all n.

Returned assessment: claim_false

There are three distinct failures. Some allowed data do not define the graded piece in the question. Even for positive homogeneous data, restriction to a principal block need not take a point of the locus to a point of the locus. Finally, there are allowed data for which the graded pieces are defined but their dimensions contradict the theorem. The journal images retain all three relevant features of the TeX: arbitrary polynomials, the preservation assertion, and the all-n conclusion.

Premises used

  • The definition uses the quotient by the generated ideal J_n, not a choice of grading after a change of coordinates.
  • A partial injection on matrices takes the principal block on its domain, relabels it, and pads with zeros in the target. The arrows on coordinate rings are pullbacks.
  • The bounds r_l must be positive; the polynomials need not have positive degree or zero constant term. Degree d=1 of the coordinate ring is expressly in scope.

Missing inputs

Scope changes

  • The verified all-n determinantal case and the eventual result for positive monomials are distinguished in REX3.
  • The failure concerns the rank-variety application, not the other applications listed in the abstract.

Located concern

CEF-rank-source lines 1798-1806; journal-rank-p55

The generating equations and even the ideal need not be homogeneous.

For p=T−1, r=1, n=2 the sole generator is det(A−I)=det(A)−tr(A)+1. The ideal is proper (evaluate at I); if homogeneous, it would contain the constant component 1. Thus the images of the standard homogeneous pieces do not give the asserted quotient grading.

Require homogeneous ideals, for instance by taking scalar monomials for the nonzero polynomials. A filtration or a translated grading is a different question. A sufficient condition on the polynomials should not be read as a necessary condition on every individual ideal: there may, for example, be no minors of the specified size.

Question 5.3.1 and the graded theorem are not well-defined in the asserted sense for general inputs. Fixing the grading alone does not fix the proof.

Located concern

CEF-rank-source lines 1820-1823; prior-report-A-rank-p7 §5.2; prior-report-B-rank-p8 §2

Principal restriction does not preserve the polynomial-rank condition.

With p=T², r=1, A's block matrix [[I₂,I₂],[−I₂,−I₂]] squares to zero: each block product sum cancels. Its first principal 2-block is I₂, whose square has rank 2. In B's indexing, N has rows (1,1),(−1,−1), so N²=0 and N⊕N restricts on {1,3} to I₂. These examples are permutation-conjugate and work even in characteristic 2. The determinant of the square of the restricted block evaluates to 1, so its pullback does not even vanish on the source point.

Use the FI-sharp argument only with verified ideal preservation, as for ordinary rank minors. For positive powers retain zero extensions but not general partial injections.

The indicated subscheme and its claimed quotient FI-sharp algebra do not exist via these maps in general. Passing to the radical or identifying FI-sharp with its opposite does not fix a failure on points. This disproves an intermediate assertion; by itself it would not disprove a dimension formula.

Located concern

CEF-rank-source lines 1820-1821; prior-report-A-rank-p7 §5.2; prior-report-B-rank-p8 §2

Even the forward zero extensions have a constant-term obstruction.

p(A⊕0_s)=p(A)⊕p(0)I_s. For p(0)≠0 this adds s to the rank, so arbitrary padding from a nonempty source can violate any fixed bound. For example p=1, r=1 already fails from size 1 to size 2. This is an obstruction to the family of maps, not a claim that every single padding with p(0)≠0 violates the bound.

For the zero-extension argument, p_l(0)=0 suffices: a target minor pulls back to a source minor or to zero. Separately ensure homogeneity to use graded pieces.

One cannot even infer the proposed one-directional functor for unrestricted data.

Located concern

CEF-rank-source lines 1794-1798, 1814-1815; journal-rank-p56 Theorem 5.3.2

The all-n numerical theorem is false, independently of any undefined grading.

Take m=1, p=1, r=1, d=1. At n=1 there are no 2-minors and the dimension is 1. At every n≥2 a principal 2-minor of I_n is 1, so the ring is zero and the dimension is 0. Both ideals (zero and unit) are homogeneous and radical. With the empty matrix the dimensions start 0,1,0,0,…; even omitting n=0 makes no difference to the contradiction. A polynomial zero on the infinite tail is zero at 1. In characteristic zero a character polynomial evaluated at the identity would give a dimension polynomial.

Narrow the hypotheses or weaken the conclusion and prove it anew. Excluding constants removes this example, not the positive-power problems in REX2 and REX3.

Refutes Theorem 5.3.2 as written, including its characteristic-zero character assertion, and already contradicts the unrestricted reduced claim. It is not a counterexample to any abstract FI-sharp classification or polynomiality theorem.

REX2

Checked statement: Remark 5.3.3's all-n conclusion for the radical quotient, tested at p=T², r=1, d=2 over C.

Returned assessment: claim_false

Write Δ_{I,J} for a minor with two ordered-increasing row and column indices, and e₂=Σ_I Δ_{I,I}. Here is a check in both directions of the alleged ideal calculation.

(1) Every rank-one matrix is in X_n since rank(A²)≤rank(A). Substitute x_ij=u_i v_j into a homogeneous quadratic q vanishing on X_n. The result is the zero polynomial over C. For a given pair of row and column multisets, the image monomial in u,v has at most two monomial preimages. There are two exactly when both multisets have distinct indices, and their difference is a minor. Comparing coefficients therefore gives q=Σ c_{I,J}Δ_{I,J}.

(2) For disjoint I,J take A sending the two basis vectors in J bijectively to those in I and killing the others. Its only nonzero 2-minor is Δ_{I,J}=±1 and A²=0. For a one-index overlap, write J={b,c}, I={c,a}; send b to c to a to zero on basis vectors, with no other arrows. The same unique minor survives and A² has rank 1. Hence every off-principal coefficient is zero.

(3) Put J=E₁₂+E₂₃ on a coordinate triple and zero elsewhere. Its square is E₁₃. Conjugating by I+tE₃₁, with inverse I−tE₃₁, gives J+tE₃₂−tE₂₁. Its principal minors on 12 and 23 are t and −t; all other principal 2-minors vanish. Evaluation at t=1 equates the two coefficients. Relabeling equates coefficients on any two pairs sharing one index, and such steps connect all pairs. For n=2 no comparison is needed. Thus the vanishing quadratics lie in C e₂.

(4) Conversely, on the direct sum of generalized eigenspaces with nonzero eigenvalues, A² is invertible. The dimension of this sum is at most 1 on X_n. The coefficient of t² in det(I+tA) is both e₂ (determinant expansion) and the second elementary function of the eigenvalues with multiplicity, hence is zero. This also gives radical membership without assuming that the minor ideal defines a reduced scheme: modulo any prime containing J_n, pass to the algebraic closure of the fraction field. The same rank and eigenvalue argument kills e₂ there and hence modulo the prime. The intersection of these primes is the radical (an element not nilpotent in the quotient can be inverted to obtain a nonzero ring and then avoided by pulling back a maximal ideal). Conversely any radical element vanishes at every point. Since e₂ is a nonzero polynomial for n≥2, the degree-two radical is exactly the claimed line.

Premises used

  • The locus is a cone and J_n is generated by homogeneous quartics, the 2-minors of A².
  • There are n² independent degree-one variables before quotienting.
  • The rank-one substitutions and the conjugations above are tests on points; radical membership of the surviving equation is checked separately.

Missing inputs

Scope changes

  • The calculation is over C and for the specified bound and grading; one such field suffices against the universal assertion.

Located concern

journal-rank-p56 Remark 5.3.3; prior-report-A-rank-p8

The reduced degree-two Hilbert function has a small-n exception incompatible with any single polynomial.

For n≥2 the dimension is binom(n²+1,2)−1. At n=1 the reduced scheme is the whole affine line and the dimension is 1 rather than 0. Agreement on all n≥2 determines the polynomial uniquely.

In this example the displayed polynomial is an eventual formula only. Restrict an all-n radical argument to ideals for which grading and the partial-injection pullbacks have actually been checked.

Report A's homogeneous counterexample to Remark 5.3.3 holds. It also precludes an all-n character polynomial by specialization to the identity.

Located concern

CEF-rank-source lines 1801-1802 versus line 1829

The surviving quadratic equation of the reduced variety is not a degree-two equation of the minor scheme.

The quartic generators give (J_n)_2=0, so the nonreduced-presentation quotient has degree-two dimension binom(n²+1,2) even when e₂ is nonzero in the polynomial ring.

Keep J_n and √J_n separate, and test the scheme claim in a degree where the generators can contribute.

This particular degree-two refutation cannot be used against Theorem 5.3.2 for the scheme at p=T².

REX3

Checked statement: For the minor scheme at p=T², r=1, the proposed degree-five obstruction to all-n polynomiality, the commutator relation count, and the scope of the proposed replacements.

Returned assessment: claim_false

The all-n scheme conclusion fails here too. The following is a characteristic-zero derivation of the ranks, not a reproduction of the report's computation. Work over C. Set V=C^n, W=∧²V, m=binom(n,2), H=End(V). For L:H→End(W) linear, form μ(L)(A)=tr(L(A)R(A)), R(A)=∧²(A²). The entries of R are exactly the quartic generators, so im μ=(J_n)_5: this accounts for every variable times every generator, without assuming independence. Write D(A)(v∧w)=Av∧w+v∧Aw. It commutes with R(A) by AA²=A²A. Thus L(A)=[D(A),C] for fixed C∈End(W) gives a relation. Here are details proving these exhaust ker μ.

(i) Take L in the kernel. In any basis, evaluate on h=diag(h₁,…,h_n). The monomials h_k h_i² h_j² (i<j) distinguish both k and the pair, including when k is in the pair. Their coefficients are (L(E_kk)){ij,ij}; all vanish. Thus L(h) has zero diagonal in the exterior basis for every such h, and this test works in every choice of eigenbasis. Write χ_I(h)=Σ{i∈I}h_i. In fact, for I≠J the linear entry L(h){I,J} is a multiple of χ_I−χ_J. To see this, match I\J bijectively to J\I by i↦j(i) and impose h_i=h{j(i)}. The shear g=1+tΣ E_{j(i),i} commutes with h. In the sheared exterior basis the diagonal still vanishes. Equivalently tr(L(h)G P_I G⁻¹)=0, where G=∧²g and P_I is the coordinate projection onto e_I. In G P_I G⁻¹ the covector e_I* stays fixed and the highest power of t in G e_I has coefficient ±e_J. This gives L(h)_{I,J}=0 on the imposed equalities. For one replacement these cut out χ_I=χ_J. For disjoint pairs the two matchings give two codimension-two subspaces whose intersection equates all four coordinates (codimension three); their sum is the hyperplane χ_I=χ_J. Linearity proves the assertion. In particular on each diagonal algebra the map has the form L(h)=[D(h),C] with one C independent of h, by choosing its entries to be the respective multiples.

(ii) Subtract such a commutator so that L(h)=0 on the fixed standard diagonal algebra. Fix a≠b and put B=L(E_ab). Apply (i) in the basis changed by g=1+tE_ab: the map h↦G⁻¹L(ghg⁻¹)G=t(h_b−h_a)G⁻¹BG is again a commutator on diagonals. For t≠0, G⁻¹BG therefore has zero diagonal and entries only between pairs with weight difference proportional to h_b−h_a. These are just {b,i} and {a,i}, i distinct from a,b, in either direction. This support holds at t=0 too, by polynomiality. Since G=1+tD(E_ab), the first-order commutator of B with D(E_ab) has zero diagonal. In the oriented basis vectors e_b∧e_i, e_a∧e_i, the latter operator sends the first to the second with coefficient 1 and kills the second. Hence the reverse entry of B is zero, separately for each i. We have proved that L(E_ab) only sends e_b∧e_i to l_{ab,i} e_a∧e_i.

(iii) The coefficients are differences of potentials on unordered pairs. To verify the consistency equations, use a matrix A supported on a coordinate pair or triple, acting there by a cycle and zero outside. It is diagonalizable. In an eigenbasis on those coordinates, completed by the unchanged others, L(A) has zero exterior diagonal by (i). Thus on the exterior block consisting of one vector from those coordinates wedged with a fixed outside spectator i, the trace of L(A)D(A)^p is zero; the same holds on the exterior square of the supported coordinates. These blocks are preserved by the replacement formula of (ii). For A=E_ab+E_ba, take p=1 on the spectator block: l_{ab,i}+l_{ba,i}=0. For A=E_ab+E_bc+E_ca, take p=2 there: l_{ab,i}+l_{bc,i}+l_{ca,i}=0, since D(A) cycles the three vectors. On the internal exterior square of that triple, D(A) sends e_a∧e_b to −e_b∧e_c, then cyclically with the same minus sign. The zero trace for p=2 now gives l_{ca,b}+l_{ab,c}+l_{bc,a}=0. For each spectator i the first two identities give numbers b_{ai} with l_{ab,i}=b_{bi}−b_{ai} (fix a base vertex and use the triangle identity; for two vertices just use antisymmetry). The internal identity says d_{ab}+d_{bc}+d_{ca}=0 for d_{ab}=b_{ab}−b_{ba}. Thus d_{ab}=t_a−t_b for some t_a by the same base-vertex argument. Replacing b_{ai} by b_{ai}+t_i makes them symmetric without changing any l. The diagonal operator C with entries b_{ai}+t_i on {a,i} now satisfies L(A)=[D(A),C] on all matrix units, including diagonals. This finishes the exhaustion. For n=2 there is no replacement to make: after the diagonal subtraction (ii) forces L=0.

(iv) For n≥2 the kernel of C↦[D(−),C] consists of scalars: commuting with all diagonals in this representation forces C diagonal by distinct pair weights, and commuting with each D(E_ab) equates entries on pairs differing by one replacement. The pair graph is connected (a singleton for n=2). Hence dim ker μ=m²−1 and g(n)=dim im μ=(n²−1)m²+1. For n=0,1 there is no minor, so the relation domain and g are zero. In particular this proves the numbers 0,0,4,73,541,2401,7876,21169,49393,103681 at sizes 0 through 9 over C, by formula rather than by an asserted rank computation.

Premises used

  • The degree-five space in question is for J_n, not its radical. The dimension of the ambient degree-five polynomials is binom(n²+4,5).
  • Linear relations here mean relations among formal products of a linear coordinate with an indexed quartic minor; the trace pairing parametrizes their coefficient arrays.
  • The proof uses polynomial identities over an infinite field and diagonalization of short cycles over C. The counterexample requires only one field.

Missing inputs

  • For an eventual character-polynomial extension via general FI theory, the general characteristic-zero theorem for finitely generated FI-modules (and its weight bound if used) is an input whose proof is outside the supplied ranges. It is not needed for the counterexample or for the eventual dimension repair below.
  • No computational implementation or certificates were supplied for the prior report's rank calculations.

Scope changes

  • The exact relation-space calculation is asserted here over C, not as a modular rank computation.
  • The direct eventual dimension argument concerns standard-graded quotients compatible with zero extension, not arbitrary inhomogeneous rank data.
  • No general formula or failure threshold for bounds r>1 is inferred from degree five.

Located concern

prior-report-B-rank-p8 table through prior-report-B-rank-p9; journal Theorem 5.3.2

A table read from a report is not a verified rank calculation; a commutator construction alone gives relations but not their exhaustion.

The derivation above supplies exhaustion in characteristic zero. Independently the product count gives g(n)≤n²m²=O(n⁶); thus if the quotient dimension were polynomial, its difference g from the ambient dimension would be polynomial of degree ≤6. Put G(n)=(n²−1)binom(n,2)²+1. The actual g equals G except that it is smaller by 1 at 0 and 1. Since Δ⁷G=0, Δ⁷g(0)=−[(-1)⁷+7(-1)⁶]=−6. Certified values through 7 would already suffice for this bounded-degree contradiction; finite differences alone would not prove the proposed tail formula.

Use the rank proof or verified rank certificates for the finite contradiction. For this example the quotient dimension equals binom(n²+4,5)−G(n) for n≥2, not for all n.

Confirms a nonconstant homogeneous counterexample to the scheme statement and hence to its all-n character version. The exception at n=1 persists if size zero is excluded. No extrapolation to larger r or arbitrary characteristic is required.

Located concern

prior-report-B-rank-p9 'What survives'; CEF-rank-source lines 1820-1827

For ordinary determinantal conditions the all-n conclusion does survive, and arbitrary-characteristic dimensions need not be argued using characteristic-zero weight.

For scalar multiples of T, target minors under a partial-injection pullback are signed source minors times the prescribed scalar power, or zero. This gives compatible homogeneous ideals; pullback compatibility passes to radicals by taking powers. The radical of a homogeneous ideal is homogeneous: in the graded quotient the top component of a nilpotent is nilpotent, and one can subtract it and descend in degree. Thus both quotient systems are graded FI-sharp modules via partial inverse. Ambient degree-d monomials use at most 2d index labels. Relabeling from these supports generates the ambient piece in sizes ≤2d and hence its FI quotient too. Values are finite dimensional. In the supplied FI-sharp decomposition the H₀ coefficients therefore vanish above 2d; the subset sum gives a binomial dimension polynomial for every n. In characteristic zero the fixed-subset trace sum at lines 1392-1397 gives a character polynomial of weighted degree ≤2d for every n.

Retain Theorem 5.3.2 and Remark 5.3.3 with verified homogeneous FI-sharp ideal compatibility, in particular in the determinantal case.

This is an all-n repair with narrower hypotheses. It does not require the general rank-preservation assertion or an external determinantal Hilbert-function calculation.

Located concern

prior-report-B-rank-p9 'What survives'; CEF-rank-source lines 1490-1493, 1819-1829

Positive monomials give only the zero-extension direction in general. Noetherianity alone is not the assertion of dimension polynomiality.

For p_l=c_l T^{s_l}, s_l≥1, padding makes p_l(A) a zero-padded block. This proves minor-ideal pullback compatibility, also for radicals, and homogeneity holds. The degree-d coordinate quotient is co-FI. Its pointwise dual U embeds as an FI-submodule of the ambient degree-d dual, whose dual monomial basis is carried by injections to the relabeled dual monomials (pullback kills monomials using new labels). Ambient finite generation by support does not by itself give an FI-sharp structure on U.

Here is a direct eventual dimension argument over any field. Restrict to ordered sets [n] and increasing injections. Order the ambient dual basis at each n lexicographically by exponent arrays with variables in lexicographic pair order. An increasing injection preserves comparisons since it inserts only zeros elsewhere. Leading basis indices of vectors of U are thus closed upwards under these injections. An index has a pattern on its q used labels in increasing order (q≤2d), and a tuple of q+1 nonnegative gaps of unused labels before, between and after them. There are finitely many patterns for fixed d. Within one pattern, componentwise enlargement of gaps is realized by an increasing injection. An upward set of gap tuples has finitely many minimal tuples: any infinite sequence admits an infinite componentwise nondecreasing subsequence by successively extracting one in each coordinate (a repeated value or a strictly increasing subsequence). Every tuple dominates a minimal one by descent. The number of leading indices at size n equals dim U_n by elimination. For one minimal tuple b, its upper orthant contributes binom(n−Σb_i,q) for n≥q+Σb_i, by distributing the remaining unused labels. Intersections use coordinatewise maxima, so finite inclusion-exclusion gives an eventual polynomial of degree ≤q per pattern. Summing proves eventual dimension polynomiality of degree ≤2d, also for the radical quotient. Moreover, choosing one leading vector for each minimal tuple gives finite FI generators by leading-term cancellation at fixed size. In characteristic zero an eventual character conclusion can then be drawn conditional on the general finite-generation character theorem; the dual has the same character since a permutation and its inverse are conjugate.

Justifies the eventual dimension repair without importing a later proof as validation of the original one. The size threshold need not be zero, as both positive-power counterexamples illustrate.

Outside the audit

  • Unrelated portions of the reports on integral generator counts and configuration-space arguments.
  • Unsupplied external papers, omitted general FI theorem proofs, and journal pages outside the supplied comparison.
  • Any claimed computer execution, modular analogues of the exact degree-five syzygy count, and conjectural higher-rank-bound patterns.
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}
}