When stable inputs do not stay stable
Explicit counterexamples in Church–Farb, and the hypotheses that preserve important uniform results. A scoped review with repairs, source checks and a decision graph.
If two sequences of representations stabilize, must their tensor products stabilize too? The answer depends on what “stabilize” means.
That distinction lies at the heart of our review of Thomas Church and Benson Farb's Representation theory and homological stability. The paper develops representation stability as a way to organize algebraic and topological objects whose dimensions grow. Our audit finds explicit counterexamples to several statements, alongside replacement arguments that preserve important uniform results. The useful story is both the failure and the mathematics that survives.
The disputed passages were checked against the published 2013 Advances in Mathematics edition. They remain there. This is a targeted comparison with the journal version, not an exhaustive audit of every page or every cited result.
Stable inputs, an unstable product
Think of a representation as an object built from irreducible pieces. For $SL_n$, these pieces have labels that can be followed as $n$ increases. Pointwise multiplicity stability means that each fixed label eventually appears a constant number of times. The threshold can depend on the label.
Let $E_n$ be the standard $n$-dimensional representation. Its dual $E_n^*$ has the label $(1^{n-1})$, which changes with $n$. If we take one copy of the dual at even ranks and two at odd ranks, every fixed label still eventually has multiplicity zero. This sequence passes the pointwise test.
But
The tensor product contains one trivial piece for each dual copy. Its trivial multiplicity therefore alternates between one and two. Both inputs are pointwise multiplicity stable; the product is not.
This contradicts the nonuniform tensor assertion in Theorem 3.1. Related examples defeat the nonuniform cancellation assertion in Theorem 3.5. They use finite-dimensional algebraic representations, not an exotic interpretation of the word “representation.”
Why the uniform version survives
Uniform stability asks for one threshold that works for all labels. For finite-dimensional representations this supplies a fixed finite list of constituents. The moving-label examples above cannot satisfy that condition.
The printed cancellation proof has a separate problem: its projection between representation rings does not preserve multiplication. Fixing the theorem's hypotheses does not make that proof step valid.
A replacement starts by bounding the unknown constituents. Every highest-weight constituent of an unknown factor produces a highest-weight constituent of its tensor product with a fixed nonzero factor. A finite list of possible outputs therefore gives a finite list of possible inputs. Once these lists are fixed, we can compare two ranks inside a single larger representation ring and cancel there legitimately.
The resulting finite-dimensional uniform theorem survives. The report states the classical representation-theory inputs it uses and the limits of their audit.
A new representation cannot be generated by zero
The paper's “simple stability” condition requires the image at rank $n$ to generate the whole representation at rank $n+1$, under the larger group action, for every rank.
The exterior square exposes the boundary problem:
No group action can turn the image of zero into a nonzero space. This example occurs in the paper's own canonical applications: degree-two free Lie words and the second homology of an abelian Lie algebra. The all-rank conclusions of Corollaries 5.7–5.9 therefore fail as stated.
The polynomial multiplicities can still be constant wherever a representation exists. Allowing a new type to first appear gives a useful changed statement, and eventual strong stability can follow after all relevant types have appeared. The report gives a forward homology reconstruction and distinguishes it from a fully re-audited replacement of every all-rank implication.
At prime two, the coefficients change the question
Theorem 8.4 calls an integral homology group an $\mathbb F_2$ representation. An $\mathbb F_2$ vector space has $2v=0$ for every vector. The integral group in question has elements of order eight, according to Sato's earlier computation. Our report also gives a separate explicit quotient of order four, which already contradicts the vector-space assertion.
Passing to homology with $\mathbb F_2$ coefficients is a meaningful alternative, but it changes the object. It is not a harmless change of notation. A second calculation shows that two congruence cases in the exterior-cube factor table are interchanged.
There is later work on the corrected field-coefficient problem: Tudur Lewis's preprint on first homology of congruence subgroups states composition-factor and uniform-periodicity results. We inspected those statements, not their full proofs. We make no claim that the problem went unnoticed or that the modern result remains entirely unresolved.
Read the findings and their limits
The campaign produced eight completed mathematical assessments across Sol, Fable and Astra, including focused challenges after the initial independent round. Three other initial attempts exhausted their allowance and supplied no completed verdict. The public record retains that distinction.
The checked pure-braid argument has no demonstrated defect in our reviewed scope. Several uniform results have useful repairs. Full modular stabilization-map checks, some imported proofs and broader claims outside the focused targets remain open in this audit. A broken prerequisite is not by itself a counterexample to every later theorem that cites it.
The report includes explicit counterexamples, replacement arguments, a decision graph, source-version records and an executable exact-check bundle. The neutral reading packet lets readers examine the paper without first seeing our verdicts.