Representation theory and topology

Representation theory and homological stability

Thomas Church and Benson Farb

91 pages66 claims indexed80 referencesManuscript 2013-06-10

gpt-6.2-astra max: cf-foundation-challenges

Reviewer record, not the publication verdict. The full response includes unaccepted proposals and source limitations. The curated report states what was adopted. Initial packets used a conservative 2010 cutoff; focused packets used the verified 2012 journal received date. The focused Fable answer needed a metadata-only derivative; its original failure and partial source access remain recorded.

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 negative findings survive adversarial checks in the intended algebraic category: the pointwise closure and cancellation statements and the two unrestricted mixed/dual implications have actual counterexamples. They do not propagate into a general failure of uniform closure. The printed projection proof fails, but finite-dimensional uniform cancellation has a valid bounded-support, common-rank replacement. Infinite cardinal absorption is a separate obstruction. This was a non-blind adjudication of the supplied proposals against the TeX and the journal page images, not independent discovery. The historical cutoff used is the recorded journal received date 2012-01-19 (supplied first-page date record), not the superseded 2010 fallback or the journal publication date. Sources for mathematical premises are identified by dated versions; the journal images are evidence of the claims under review. No searches or computations were executed. Later knowledge cannot be removed from model weights, but no later theorem is used as a premise for the original argument.

FC1

Checked statement: The pointwise multiplicity conclusions of Theorem 3.1(1)–(5) for SL, of Theorem 3.5 for SL, GL and Sp, and of the tensor-square assertion in Remark 3.6; separately, the scope of cancellation with infinite multiplicities.

Returned assessment: claim_false

The counterexamples are to the printed hypotheses, not just to a conjectured extension. The journal images confirm both the finite-dimensional, nonuniform SL clause of 3.1 and the distinction between the two clauses of 3.5. Write E_n for the standard module, 1 for the trivial module, and a_n=1 for even n, 2 for odd n. Multiplicity stability asks only for an eventual value at each fixed label. A moving simple, even with varying multiplicity, can contribute zero to every one of these eventual values. Neither the first clause of 3.1, the cancellation in 3.5, nor the assertion about squares requires injective or generating transition maps. If consistency alone is requested, zero maps suffice. Full representation stability of the inputs is a different proposed hypothesis; FC3 supplies actual maps for a counterexample to tensor closure even under that strengthening.

Premises used

  • [HW] available_from=1991 (year precision); date_evidence=CF-foundations 4475–4477; date_basis=book publication; version=Fulton–Harris, Representation Theory, GTM 129 (1991), classical highest-weight and Schur-module results, including §§6 and 15.3 as cited. Statements used as in CF-foundations 682–738, 771–791, 809–836: in the finite-dimensional rational algebraic, characteristic-zero category for these split groups there is complete reducibility, simples have a one-dimensional highest line and are determined by their dominant weight; S_p(E) is the polynomial GL simple for a partition p of length at most dim E. SL identifies full tuples differing by a constant; Sp simples are the traceless Schur modules of length at most the symplectic rank and are self-dual. In particular Sym^r(E) for Sp is simple: every pair contraction of a symmetric tensor vanishes. These are the split forms used over Q in the manuscript, not a classification of abstract group actions.
  • [LR] available_from=1991; date_evidence=CF-foundations 4475–4477; date_basis=book publication; version=Fulton–Harris (1991), Eq. 6.7 as reproduced at 1488–1501. For polynomial Schur modules in characteristic zero the tensor coefficient is the Littlewood–Richardson coefficient, output size is the sum of input sizes, and a shape too long gives zero. Its Pieri cases with a row or column add a horizontal or vertical strip of the specified size with multiplicity one. These size assertions precede, not follow, SL normalization.
  • Elementary checks used below: wedge pairing identifies Λ^(n−1)E with E* on SL and with det⊗E* on GL; End(E)=1⊕sl(E) by trace in characteristic zero. The traceless summand is simple, also seen by applying the one-box rule to E⊗Λ^(n−1)E and removing the determinant twist. The trace pairing is nondegenerate on this summand.

Missing inputs

Scope changes

  • Polynomial GL repair, checked separately (prior-foundations 238–256): with ordinary Schur names S_p(Q^n) padded by zeros, polynomial degree is a nonnegative grading. For an output of degree b in a tensor product or a fixed Schur construction only input degrees ≤b can contribute (apart from factors used zero times). These finitely many degree slices have finitely many labels; one can truncate to them and use the fixed-label formulas. This is not what the constant-tail naming rule says about all polynomial modules. In the paper's GL names, E_n and (Λ^(n−1)E_n)^{⊕a_n} are pointwise stable, yet their product has det_n with multiplicity a_n by Pieri, in addition to a_n copies of S_(2,1^(n−2))E_n. The determinant has a fixed tail-1 name. Thus mere polynomiality is not the fix, and 3.1 did not assert pointwise GL closure.
  • New restricted cancellation argument, not the printed proof: for finite-dimensional polynomial GL modules in ordinary Schur names, suppose W and C=V⊗W are pointwise stable and some stable multiplicity of W is positive. Let d be the least degree with nonzero stable class w_d. In each fixed degree the whole class stabilizes, since there are finitely many partitions. Below d the classes eventually vanish simultaneously. For fixed m, compare large ranks in degree d+m: w_d v_(m,n)=c_(d+m)−Σ_{j=d+1}^{d+m} w_j v_(d+m−j,n), where lower-case letters denote homogeneous classes. By induction on m the right side is fixed. All products here have degree at most d+m and can be compared with identical LR tables at a common rank greater than d+m. Its character ring is a domain (FC2); w_d≠0 forces equality of the degree-m classes. This proves pointwise cancellation under the stated extra hypotheses. Without the positive stable slice, Sym^n(E_n) still defeats cancellation in ordinary Schur names.

Located concern

CF-foundations 1369–1385, 1554–1564, 1642–1667; prior-foundations 98–152

The fixed-output-size bound used in the nonuniform closure proof is false for normalized SL labels.

Use SL_n, n≥3. The inputs E_n and a_n copies of E_n* are pointwise multiplicity stable: the latter's only label is (1^(n−1)), different at every rank. Their product is a_n copies of 1⊕sl(E_n), with trivial multiplicity a_n. For X_n=E_n⊕(E_n*)^{⊕a_n}, Sym²(X_n) has exactly a_n invariants: the cross term gives those invariants, and neither E_n⊗E_n nor E_n*⊗E_n* has an invariant since E_n is not self-dual. This gives (2) and (3); the one-box outer Schur functor gives (4) from the tensor counterexample, and the one-box inner functor gives (5) from the Sym² counterexample. The lost degree in the tensor example is a whole column of height n.

Delete the pointwise-only SL clause, or impose a condition that prevents the moving full-column contributions, for example the finite-dimensional uniform hypothesis already present in the theorem.

All five advertised pointwise closure operations fail, by one mechanism. The uniform conclusions are not refuted. A second valid numerical witness is X_n=Λ*E_n: there are two trivial copies and n−1 other distinct exterior simples paired with their duals. Hence dim (X_n⊗X_n)^SL=4+(n−1); Sym²(X_n) has at least 3+floor((n−1)/2) invariants from the trivial part and distinct complementary pairs. This uses only multiplicities; with the usual degree-preserving inclusion the image actually misses the new top exterior degree even after taking its orbit span.

Located concern

CF-foundations 1855–1860, 1885; prior-foundations 156–192

Nonvanishing in each rank is not a nonzero cancellable limit. Even a persistent constituent does not in general suffice.

In each of SL, GL and Sp take W_n=Sym^n(E_n), V_n=1^{⊕a_n}. The simple W_n has label (n) for SL and Sp and polynomial part (n) with tail 0, i.e. (n,0), for GL. Both W_n and a_n W_n are everywhere nonzero with every fixed multiplicity eventually zero. V_n is not multiplicity stable. For a persistent version in SL, one can even keep W_n=E_n uniform and alternate V_n between 1 and sl(E_n). In E_n⊗sl(E_n), adding a box to (2,1^(n−2)) gives (3,1^(n−2)), (2,2,1^(n−3)), and (2,1^(n−1)); only the last has n rows and it normalizes to (1). Thus the product always has one standard constituent and all extra labels escape. The other persistent example in prior-foundations also checks out: put A=(3,1^(n−2)). By the vertical-strip rule Sym²(E)⊗E*=E⊕V(A); by the one-box rule E⊗V(A)=Sym²(E)⊕V(4,1^(n−2))⊕V(3,2,1^(n−3)). In the latter formula the new bottom box gives (3,1^(n−1)), hence (2) on SL. Taking (V,W)=(Sym²(E),1⊕E*) at odd ranks and (E,1⊕V(A)) at even ranks therefore gives persistent trivial multiplicity one in W and persistent multiplicities one of E and Sym²(E) in the product. Every other displayed product label has n−1 rows.

For these families use the finite-dimensional uniform hypothesis on both known decompositions, with an eventually nonzero multiplier (FC2). Requiring a nonzero pointwise multiplicity in each known decomposition, or uniformity of the known factor alone, is not that hypothesis.

The nonuniform theorem is false for every family listed; the strengthened persistent examples already work in SL. The compactness argument does not fix it: the span of high-size simples need not be an ideal (tensor a high simple with its dual and obtain the unit as a constituent). Even in a positively graded polynomial ring a degree truncation can kill the multiplier.

Located concern

CF-foundations 1888–1889; prior-foundations 194–210; sol-foundations F3

Recovery from a square at a single rank does not imply recovery of pointwise stability from square stability.

For SL take V_n=1 at odd ranks and V_(2m)=Λ^m E_(2m). Adding a vertical m-strip to (1^m) gives exactly the shapes (2^j,1^(2m−2j)), 0≤j≤m, once each. The j=0 term of the square is the determinant, trivial on SL. All others are already normalized and have 2m−j≥m rows. Thus the squares have stable trivial multiplicity one and all other fixed multiplicities eventually zero, while V_n has oscillating trivial multiplicity. The Sol alternative works for both SL and GL: alternate 1 and the traceless adjoint. The adjoint is self-dual simple, so its square has one trivial copy. In SL all weights of that square have total size 0 modulo n (sums of roots, modulo full columns), excluding every fixed nonempty partition for n larger than its size. In GL scalar weight is zero, whereas a fixed ordinary label of polynomial part p and constant tail c has scalar weight |p|+nc, eventually nonzero unless the label is trivial. This test does not assume the center of SL_n(Q) contains a primitive nth root of unity.

Retain a finite-dimensional uniform square statement with the proof in FC2. At fixed rank one may recover actual classes from their squares, but an algorithm reading successive largest weights would have to subtract the full products already accounted for, and has no demonstrated pointwise-limit property.

The pointwise assertion in Remark 3.6 is false. These finite-dimensional square witnesses do not by themselves decide a separate Sp-only square assertion.

Located concern

CF-foundations 893–896, 920–923, 1855–1873, 1888–1889; prior-foundations 234–236; sol-foundations F3

Independently of moving labels, unrestricted infinite multiplicity invalidates even uniform cancellation.

Let W_n be a countable sum of trivial lines and V_n=1^{⊕a_n}. W_n and the product have the identical constant countably infinite trivial multiplicity. V_n does not stabilize. There is also a square obstruction: put H_n equal to countably many copies of 1⊕E_n^{⊗2}, and Y_n=H_n⊕E_n^{⊕a_n}. At large rank E_n is absent from H_n (scalar weight in GL, size modulo n in SL, parity under −I in Sp). In the square, H_n^{⊗2} already absorbs the finite a_n² copies of E_n^{⊗2}; the cross terms have countably infinite multiplicities on their support for either a_n. The result is a countable sum of copies of the fixed tensor powers E_n^{⊗j}, 0≤j≤4, with each power repeated countably often. Their finite list of decompositions stabilizes uniformly by the fixed-input laws in FC2.

Require finite-dimensional algebraic modules for the uniform ring argument. Finite multiplicities together with known finite support suffice for finiteness; local finiteness alone does not. The coefficients of the ordinary representation ring are integers, not infinite cardinals.

This is a missing finiteness qualification in 3.5 and the analogous square statement under Section 2's literal convention. It is not the explanation of any of the finite-dimensional failures above, nor an objection to the explicitly finite-dimensional uniform multiplicity clause of 3.1.

FC2

Checked statement: Finite-dimensional uniform closure in Theorem 3.1, and finite-dimensional uniform cancellation in Theorem 3.5 and Remark 3.6: validity of the conclusions, the printed proof, and the proposed replacement.

Returned assessment: repairable_gap

In this verdict the subject is the FINITE-dimensional uniform inference. It is true. First, III′ for a finite-dimensional sequence gives a fixed finite support: at its common threshold N every later positive label must already occur in the finite decomposition at N. In GL use the name (p;c) for S_p(E_n)⊗D_n^c, D_n=det_n, with p padded by zeros and ℓ(p)<n. For a full tuple λ take c=λ_n and p_i=λ_i−c. Thus the rank extension repeats the last entry of λ, not necessarily zero. In SL use p alone with last entry zero; in Sp use the partition of length ≤n.

Uniform closure: for fixed SL polynomial parts take the rank larger than the possible sizes in the polynomial LR products, so no output has a full-height column. In GL the same calculation has the additional tensor exponent c+d. For a fixed outer Schur functor S_α on a single GL type, ordinary plethysm applies to p and the twist is D_n^{c|α|}, by homogeneity. For Sp one needs the tensor law only in stable rank. For example, it follows from the quoted Littlewood restriction rule without any assertion about tensor coefficients on the rank boundary: on partitions of size ≤b, for n≥b, the restricted GL Schur classes equal the corresponding Sp simples plus integer combinations of strictly smaller sizes (the empty paired-row partition gives diagonal coefficient one). This finite triangular matrix has a rank-independent inverse. Express two Sp simples this way, multiply the restricted polynomial classes by LR, and convert back by restriction, with b the sum of the two input sizes. This gives the whole finite product table for the fixed pair. For Sp Schur functors on a single simple the fixed-input plethysm statement quoted from Kabanov suffices. Tensor products distribute over the fixed finite lists of inputs, and the direct-sum formula S_α(A⊕B)=⊕{β,γ} c^α{βγ}(S_β A⊗S_γ B) reduces a Schur functor on any such list to the single-simple cases. This proves uniform multiplicities in (1)–(3); composing the tensor and Schur operations gives (4)–(5). No stabilization map is used.

New repair of uniform cancellation (verifying the proposal, not the projection proof): let B,F be the finite stable supports of W_n and C_n=V_n⊗W_n and assume W_n≠0 eventually. Fix μ in B of positive multiplicity. For every input type λ of V_n the Cartan type λ+μ belongs to F: the tensor of highest vectors is nonzero, of that weight and killed by the positive root algebra, hence gives that simple constituent by complete reducibility. In SL the normalized sum still has last entry zero; in Sp add padded nonnegative partitions; in GL the sum of (p;c) and (q;d) is (p+q;c+d). Thus subtracting μ from each member of F leaves only a fixed finite candidate set A for λ, discarding non-dominant or inadmissible differences. This bound uses positivity of actual representations, not virtual differences. Take n,m past the finitely many definition and product-table thresholds for A×B and past stabilization of W,C. Interpret the two candidate decompositions of V_n,V_m as classes x,y at one rank t≥max(m,n) beyond the same thresholds. With the same nonzero class w from W they satisfy xw=[C]_t=yw because these particular product tables agree at all three ranks. In the finite-dimensional representation ring at t the character map into the integral group ring of the weight lattice is injective: in a nonzero finite difference of simple classes a maximal highest weight has a nonzero coefficient in the character. The lattice is free abelian, so that group ring is a Laurent polynomial domain. Cancel w and compare the distinct simple basis elements at t. This proves III′ for V, not just III. If C is eventually zero and W eventually nonzero, V is eventually zero as well. If W is merely nonzero somewhere before the stable range, cancellation need not say anything.

New repair for uniform squares: the types 2λ in the square bound the input support by the halves of its fixed finite output support, including halving both the polynomial part and the exponent when possible in GL. Use the fixed product tables on this candidate set squared. At a common rank x²=y² implies (x−y)(x+y)=0; for effective classes x+y is nonzero unless both are zero. This proves the finite-dimensional uniform square assertion.

Premises used

  • [HW] and [LR] with the 1991 version, dates, evidence and hypotheses recorded in FC1. Only highest-weight triangularity and the elementary lattice-domain argument are needed for cancellation at a fixed rank, not a theorem identifying the entire Weyl-invariant subring.
  • available_from=1991; date_evidence=CF-foundations 4475–4477; date_basis=book publication; version=Fulton–Harris (1991), Exs. 6.11, 6.12, 6.17(a) as cited at 1566–1575, 1605–1611, 1699–1708. In characteristic zero, Schur of a direct sum has the LR expansion above; restriction of S_p(E_n) to GL_(n−k) has coefficients Σ_τ c^p_{ντ} dim S_τ(Q^k); S_α(S_p E) has a finite Schur expansion with coefficients independent of E and degree |α||p|, omitting shapes too long for E.
  • available_from=1991; date_evidence=CF-foundations 4475–4477; date_basis=book publication; version=Fulton–Harris (1991), Eqs. 25.36 and 25.39 as stated at 1733–1757 and 1774–1786. In characteristic zero, for one symplectic pair removed the coefficient for ν counts nonnegative interlacings λ_i≥p_i≥λ_(i+1), p_i≥ν_i≥p_(i+1) with the indicated zero ends. For a polynomial GL_(2n) shape ρ with at most n rows, the Sp_(2n) restriction coefficient of τ is Σ_η c^ρ_{ητ} over η with rows in equal adjacent pairs. The triangular tensor calculation uses only this last rule with n≥b. In particular it does not require the displayed Sp tensor law at 1531–1541 to hold in arbitrary small rank.
  • available_from=1997; date_evidence=CF-foundations 4547–4548; date_basis=journal publication; version=Kabanov, J. Algebra 195 (1997), Theorem 7 as quoted at 1592–1598. For fixed partitions α,ρ and the characteristic-zero Sp simple of type ρ, the coefficients of S_α(V(ρ)_n) stabilize for n≥ℓ(ρ)|α| with output lengths ≤ℓ(ρ)|α| (use ranks where the input exists). Only the eventual finite decomposition is needed here.
  • For the distinction from 3.2: available_from=1938; date_evidence=CF-foundations 4583–4585; date_basis=journal publication; version=Murnaghan, Amer. J. Math. 60 (1938), §1 as cited at 1827–1831, stabilization of the full padded S_n tensor decomposition for two fixed partitions in characteristic zero. The quoted quantitative form n≥|λ|+|μ|+λ_1+μ_1 has available_from=2011; date_evidence=CF-foundations 4416; date_basis=journal publication; version=Briand–Orellana–Rosas, J. Algebra 331 (2011), Theorem 1.2 as cited. The 2011 journal version predates the recorded cutoff; Sol's concern about needing a pre-2010 version is not applicable. For the shift, available_from=1991; date_evidence=CF-foundations 4475–4477; date_basis=book publication; version=Fulton–Harris (1991), Eq. 4.42 as cited at 1134–1138, the characteristic-zero one-box restriction rule for S_n.

Missing inputs

  • The external books and articles containing the classical formulas and the cited stable ranges are not reproduced as external sources here. The closure checks take the identified, pre-cutoff formulas as premises, not their full proofs as audited.
  • The later graded theorems and proofs are not in CF-foundations. To certify an application one needs its exact degreewise hypotheses, the eventually nonzero factor, which known classes are uniformly stable, and any requested range or map conclusion.

Scope changes

  • For a later argument on actual finite-dimensional graded pieces that needs eventual uniform numerical cancellation in one fixed degree at a time, the replacement gives exactly that step if W and the product satisfy its hypotheses there. Finite grading pieces alone do not guarantee those hypotheses. It gives no common range over all degrees, no comparison of chosen transition maps, and no simple-stability conclusion controlling multiplicities from the first rank at which a label exists. No unseen Lie-homology proof is thereby certified.
  • Uniform finite support is automatic here because of finite dimension, not because there are only finitely many algebraic-group labels at rank N. In particular the finite-type shortcut in the proposed Sol extension for arbitrary infinite-dimensional strong inputs would itself need justification or replacement. Such an extension is not needed for the finite repair.
  • Theorem 3.2 has a different scope: S_N really does have finitely many simple types. Under III′ there is a common finite label list, so the fixed-pair symmetric-group tensor law and iterated one-box branching apply simultaneously, with cardinal sums if multiplicities are infinite. This supports its uniform multiplicity conclusion on those premises; it does not assert orbit generation.

Located concern

CF-foundations 1873–1885; journal-foundations-p27 (printed p. 276)

The printed additive projection does not preserve multiplication. Fixed-rank uniqueness therefore does not justify projecting a solution.

Already for SL_3→SL_2 the label projection kills E_3*=(1,1) and sl(E_3)=(2,1), but π(E_3⊗E_3*)=1. In GL_(n+1) it kills Λ^n E (last two entries 1,0); E⊗Λ^n E contains D_(n+1), which projects to D_n. In Sp_(2n+2) it kills the simple (1^(n+1)), whose square contains the unit by self-duality. No surviving positive summand can cancel the surviving term in these tests.

Bound the unknown labels first by the Cartan argument, and transfer only the finite controlled product tables to a common rank, as in the new repair above.

The ring-map inference is invalid, including in the proof purporting to handle uniform inputs. The separate assertion that a high-length input produces at least one high-length output can follow from the Cartan sum; it does not make the projection a ring map. The finite uniform theorem survives by replacement.

Located concern

CF-foundations 1392–1397, 1699–1796; journal-foundations-p26 (printed p. 275, end of restriction proof)

Restriction conclusions need the intended source groups and a sufficiently large remaining rank, not a universal boundary-rank formula or a map claim.

For a fixed shift k in SL or GL, write E_n=E_(n−k)⊕Q^k and apply the restriction formula; D_n restricts to D_(n−k). Take n−k past the polynomial row bounds to read normalized labels. For Sp, once n−1≥ℓ(λ), trailing zeros force the tail of the one-step interlacing pattern to be zero and the entire finite list of coefficients is fixed. Iteration gives any fixed k; at n=ℓ(λ) a same-length output need not even be available in rank n−1. In (7), GL_n→SL_n forgets the exponent c. GL_(2n) or SL_(2n)→Sp_(2n) forgets any determinant and uses the paired-row restriction formula for n at least the polynomial row bound, with input multiplicities taken at 2n.

State (6)–(7) as eventual uniform multiplicity statements in those restriction directions, taking a maximum of the input and branching thresholds. Do not read (7) as a theorem for arbitrary embeddings into a symplectic source: for example the Levi embedding of SL_n in Sp_(2n) gives E_n⊕E_n* from the symplectic standard, not a uniform SL decomposition.

The intended finite-dimensional uniform restriction assertions follow on the quoted formulas. They assert neither injective generating structure maps nor stability starting at the first admissible rank.

Located concern

CF-foundations 1024–1034, 1080–1151, 1162–1228, 1398–1400, 1445–1452, 1798–1822; journal-foundations-p26

The strong conclusion has extra map hypotheses. Its proof needs control of highest lines, not uniqueness of arbitrary weights above an old weight.

Uniformly stable maps alone can miss an entire output under tensoring: in 1445–1452, (a,v,w)↦(a,az,w+vz), z the new basis vector, is injective by the disjoint old and mixed symmetric-square subspaces and its orbit generates all target simples. But the standard projections of its image are all multiples of z, so the exterior-square projection of the tensor image is zero. This works also for GL with polynomial standard spaces. Conversely, a claim that a restricted weight has a unique lift among all occurring weights is too strong: in 1⊕E_(n+1) both the trivial weight and the last coordinate weight restrict to zero.

Finite-dimensional strong clause, clarified argument: in a same-label new simple the old simple has multiplicity one by the Schur restriction formula in large rank (also after a fixed determinant twist). Its canonical old-space inclusion is fixed by the last-column unipotent P. Thus IV on the inputs gives P-fixed images. Sums, tensors and Schur functors preserve this property and preserve injections (linear injections split as vector-space maps). For an old highest vector in an output, its image is now killed by the old positives and Lie(P), hence by all new positives. Its projections into new simples are on highest lines, not arbitrary weight spaces. For the fixed finite output support at large rank, restriction of these highest weights identifies their labels uniquely, since the normalized partitions have finished and the GL tails are constant. A nonzero combination of highest lines in copies of the same simple generates one diagonal copy. This gives IV for every old simple submodule without rechoosing its vector. For k old copies of one type, injectivity and same-type branching multiplicity one force their span to contain k new copies; IV gives at most k. By equality of the finite multiplicities this fills the target, giving II.

The finite-dimensional strong assertion for SL and GL in (1)–(5) survives. It is not a consequence of the pointwise theorem, not a strong restriction assertion, and not stated for Sp. The same proof cannot simply omit finiteness: e.g. an injective proper shift on a countable trivial multiplicity space preserves type and multiplicity but not generation.

FC3

Checked statement: The assertions near Definition 2.16 that mixed stability implies ordinary stability and that ordinary stability gives mixed stability of the duals, with the actual GL names and the forward map conditions; scope of the underlying representation category.

Returned assessment: claim_false

The journal image retains both implications. Consistency is equivariance along the inclusion, I is injectivity, and II is generation by the next group's orbit, not surjectivity of the linear map. Ordinary representation stability adds pointwise III; uniform stability uses simultaneous III′, and strong stability also requires the correct next simple span of every simple copy. Multiplicity stability alone drops I–II. Definition 2.16 uses I–II and a pointwise test for each fixed mixed pair. These definitions are distinguishable and usable; the problems here are with implications between them. In particular the formula (786–788), the example V(3,1)=V(3,1,1,1) (802–804), and the explicit inclusion of full GL tuples (1883) corroborate the constant-tail reading, despite the abbreviated weight at 793–795 omitting the repeated entries. A stable polynomial Schur functor with fixed partition p instead has name (p;0) in the notation of FC2 at ranks greater than ℓ(p).

Premises used

  • [HW] with available_from=1991, date_evidence=CF-foundations 4475–4477, date_basis=book publication, version=Fulton–Harris (1991), as specified in FC1. Used only for the algebraic simple modules, their distinct highest weights, and complete reducibility in characteristic zero. In particular E, E*, Sym²(E) and the traceless adjoint used below are simple at the large ranks in question.
  • CF-foundations 1310–1333 supplies the label test directly: the full GL weight of a fixed mixed pair has fixed positive and negative ends separated in large rank by zeros. Dualizing negates and reverses the full tuple, so interchanges the pair. This is an identity of individual finite-dimensional representations.

Missing inputs

Scope changes

  • The GL argument depends on the manuscript's repeated-tail convention. With only fixed ordinary polynomial Schur labels to test, the empty-negative-part mixed pairs would already test those labels. This does not establish a general implication for the printed GL convention.
  • No map property for a dual, restriction, or tensor sequence has been inferred merely from stable multiplicities. Reversed-map stability is separately defined at 1002–1017.

Located concern

CF-foundations 1335, 780–807; journal-foundations-p20; prior-foundations 645–670

In GL, MTIII does not test ordinary types of fixed nonzero determinant tail.

Use B_n=1⊕E_n^{⊕a_n}⊕Sym² E_n and T_n=D_n⊗B_n. Identify E_(n+1)=E_n⊕Qz. At a 1→2 increase send (q,v,w) to (q,v,qz,w), with old spaces included; at a 2→1 decrease send (q,v_1,v_2,w) to (q,v_1,w+v_2 z). These are GL_n-equivariant since z is fixed. They are injective, in particular Sym²(E_n) and E_n z are disjoint. At an increase, pure v and w generate the first standard and symmetric-square targets. The orbit of (1,z) in the trivial and second standard targets separates them: subtract a translate to get a nonzero pure standard vector, generate that copy, then recover the trivial vector. It does not have to separate two isomorphic copies from a diagonal. At a decrease the pure q,v_1,w already generate all targets. Twist the maps on determinant lines by d↦d∧z; this is equivariant for the old group and nonzero between lines, and twisting the new group action by a character does not change orbit spans. The three full weights of T_n are (1^n), (2,1^(n−1)), (3,1^(n−1)), with multiplicities 1,a_n,1. No fixed mixed pair matches any of them once it has a zero middle. But the fixed ordinary name V(2,1)_n=D_n⊗E_n has multiplicity a_n.

For SL the implication does hold: each fixed normalized ordinary label is tested by the mixed pair with empty negative part. For GL one can add all fixed determinant twists to the mixed tests, or hypothesize eventual absence of ordinary nonzero-tail types, for example under a fixed polynomial degree bound. A bound merely on the number of ordinary types is insufficient: this example uses just three, all polynomial as modules.

Mixed representation stability need not imply ordinary GL representation stability, even with injective generating transitions and finite-dimensional algebraic inputs. The individual mixed labels and Definition 2.16 need not be discarded.

Located concern

CF-foundations 1339–1344, 1017; journal-foundations-p20; prior-foundations 620–644

Swapping labels under duality does not turn the ordinary pointwise tests into all mixed tests, and transposition reverses the arrows.

For SL let Z_n=1⊕(E_n*)^{⊕a_n}⊕sl(E_n). Write f̄ for extension by zero on z and Ā=diag(A,0). At a 1→2 increase use (q,f,A)↦(q,f̄,qz*,Ā), where z* is the last coordinate covector. At a 2→1 decrease use (q,f_1,f_2,A)↦(q,f̄_1,Ā+z⊗f̄_2). The added matrix is traceless and confined to the bottom-left row, disjoint from the old block. Thus both maps are equivariant and injective. At an increase the pure old covector and adjoint generate the first dual copy and the entire new adjoint. The orbit of (1,z*) in the trivial and second dual targets gives both, by the difference-of-translates argument. At a decrease the pure q,f_1,A generate every target simple. In Z_n only the trivial name persists: the dual and adjoint have normalized labels (1^(n−1)) and (2,1^(n−2)). Thus I, II and III all hold. In Z_n* the fixed standard mixed pair ((1);∅) has multiplicity a_n. No choice of maps for the duals can fix that numerical failure. The same block maps even work for GL: the two moving full weights end in −1 with normalized polynomial parts of length n−1. The Sol variants using an optional standard copy and a Sym² buffer, or their covector versions, use the same disjoint old-square and last-coordinate-times-old subspaces.

Retain V(λ;μ)_n* ≅ V(μ;λ)_n. For finite-dimensional modules, mixed multiplicity stability of the original modules gives mixed multiplicity stability of their duals by this identity. Map conditions require a separate specification and proof; the natural transposes are not the forward maps in Definition 2.16.

The asserted sequence-level dual implication is false already for SL. As a cross-check for FC1, Z_n and E_n are fully representation stable with the stated maps, but (Z_n⊗E_n)^SL has dimension a_n: only the dual copies pair with E_n. Thus ordinary map conditions do not repair nonuniform tensor closure.

Located concern

CF-foundations 602–606, 676–695, 893–897

The literal definition of an action on a Q-vector space is broader than the algebraic representation theory invoked.

For the abstract group GL_n(Q), let g act on Q² by a unipotent shear with upper-right entry v_2(det g). Valuation additivity makes this an action. It is finite-dimensional but not semisimple: diag(2,1,…,1) acts by a nonidentity Jordan block and the fixed line has no invariant complement.

For the classical algebraic groups require rational algebraic finite-dimensional modules, or modules locally a sum of such modules in the locally finite formulation. A sum of submodules that are sums of simples is semisimple. The algebraic-group heading, Schur constructions and highest-weight discussion strongly indicate this intended context.

This is a category qualification, not a refutation of the stability framework or of finite-group semisimplicity. Conversely, adding the intended qualification does not defeat any of the tensor, cancellation or mixed-label counterexamples: their modules are algebraic.

Outside the audit

  • Full proofs in the externally cited representation-theory sources, including sharp stable-range results; bibliographic dating and use of the cited statements are not external proof audits.
  • Omitted sections, in particular the detailed hypotheses, maps and ranges of the later Lie-homology applications. No integral or field-coefficient homology conclusion is established by this packet.
  • A standalone finite-dimensional Sp-only classification of pointwise tensor-square recovery, and maximal infinite-dimensional extensions of the positive uniform results.
  • Other Section 2 examples, transfer assertions, and other Sol or prior-report targets outside FC1–FC3.
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 Representation theory and homological stability. Math Referee, .

Download .bib
View BibTeX
@misc{rivin2026review-human-church-farb-homological-stability,
  author = {Rivin, Igor},
  title = {{Review of Representation theory and homological stability}},
  year = {2026},
  month = {November},
  date = {2026-11-21},
  howpublished = {Math Referee},
  url = {https://mathreferee.dimensionreducers.ai/papers/human-church-farb-homological-stability/panel/},
  note = {Review report}
}