The claim
These are standard results, checked here by reasoning. The orthogonality relations they use are proved in Peter and Weyl (LIT-329, read in NOTE-305, eqs. 6–8), for compact Lie groups by Schur’s method with integration over the group; a finite group is the zero-dimensional case, where the integral is the average over G. Serre’s Linear Representations of Finite Groups (LIT-893) is their standard textbook source and is unread here; it is cited, not stood on.
- The character χ_ρ(g) = Tr ρ(g) is a class function, χ(hgh⁻¹) = χ(g), and it is unchanged when ρ is replaced by SρS⁻¹. So a change of coordinates cannot change it.
- For a finite group G and finite-dimensional representations over ℂ, or over any field of characteristic zero, χ_ρ₁ = χ_ρ₂ exactly when ρ₁ ≅ ρ₂. Over ℂ the reason is short. Maschke’s theorem makes every representation a direct sum of irreducibles, and the irreducible characters are orthonormal for ⟨χ, ψ⟩ = |G|⁻¹ Σ_g χ(g)ψ(g)̄. So the multiplicities m_i = ⟨χ, χ_i⟩ can be read off the character, and they fix ρ up to isomorphism.
- The irreducible characters form a basis of the class functions, and there are as many of them as there are conjugacy classes.
- The same holds for continuous finite-dimensional complex representations of a compact group.
Two steps the list compresses. Over a field K of characteristic zero, KG is semisimple (Maschke), and each simple module S_i has a central idempotent e_i ∈ KG that acts as the identity on S_i and as zero on every other simple (Wedderburn). Extend χ_V linearly to KG. If V ≅ ⊕ S_i^{m_i}, then e_i projects V onto its S_i-isotypic part, so χ_V(e_i) = m_i · dim_K S_i. Because the characteristic is zero, dim_K S_i ≠ 0 in K, so χ_V fixes each m_i, and so fixes V up to isomorphism. In characteristic p the same equation fixes m_i only mod p, which is the first failure below. For a compact group G that is not a Lie group, the image of ρ₁ ⊕ ρ₂ is a closed subgroup of a general linear group, so it is a compact Lie group H (Cartan’s closed-subgroup theorem). Both representations factor through H, and equal characters on G are equal characters on H. So item 4 reduces to the Lie case.
The exchange states the core correctly. A194 §1: “For finite-dimensional complex representations of finite groups”, ρ₁ ≅ ρ₂ ⇔ χ_ρ₁ = χ_ρ₂, which “follows from complete reducibility and the orthogonality of irreducible characters”. A198 §2 gives the multiplicity formula. A218 §3.2: “Under the standard finite-group, finite-dimensional complex assumptions, characters classify linear representations up to isomorphism.”
Where it fails
- Characteristic p, even when everything is semisimple. Over F_p, for the trivial group, the one-dimensional and the (p+1)-dimensional trivial representations both have character 1. Multiplicities are known only mod p. So semisimplicity is not enough.
- Non-semisimple settings. For the cyclic group C_p over F_p, the regular representation, which is indecomposable, and p copies of the trivial representation both have character 0.
- Infinite groups. ℤ acting on ℂ² by n ↦ [[1, n], [0, 1]], and ℤ acting trivially on ℂ², both have the constant character 2. The two representations are not isomorphic. Equal characters give at most isomorphic semisimplifications.
A194 §4 says completeness “depends on conditions such as semisimplicity”, which is hedged and incomplete. A218 Ch3.5, “Completeness of characters under finite-group semisimplicity assumptions”, is wrong as worded. It should read “characteristic zero”. A218 §3.2 states the condition correctly.
What the character forgets
- The isomorphisms. A character fixes the isomorphism class, not the representation, and the isomorphisms are not unique. By Schur’s lemma the automorphisms of ⊕ S_i^{m_i} form ∏ GL(m_i). A character lives in the representation ring, where the intertwiners are gone. It is a decategorification.
- The group. The dihedral group D₄ and the quaternion group Q₈ are not isomorphic and have the same character table.
- The permutation action. The permutation character of a G-set counts
fixed points, χ_X(g) = |Fix_X(g)|, and non-isomorphic G-sets can share it.
A198 §4: “Two G-sets can even have the same permutation character without
being isomorphic as G-sets.” Two checked instances:
- In the Klein four-group, take the regular G-set plus two fixed points, and the union of the three G-sets G/H₁, G/H₂, G/H₃ for its three subgroups of order two. Both have permutation character (6, 2, 2, 2). Their orbit sizes are {4, 1, 1} and {2, 2, 2}.
- GL(3, 2) acts on the seven points and on the seven lines of the Fano plane. The two actions have the same permutation character and are not isomorphic as G-sets.
The exact link to Yoneda. Over ℂ, ⟨χ_V, χ_i⟩ = dim Hom_G(S_i, V) for each irreducible S_i. So a character records the dimensions of the hom-sets from the irreducibles. That is a counts-only relational profile, taken on a restricted family of probes. THEORY-032’s first condition says that bare counts of morphisms do not in general fix an object, and that only special classes are exceptions. Finite-dimensional representations of a finite group in characteristic zero are one such class, because the category is semisimple. The character theorem is a case of the Yoneda moral, and it holds only up to isomorphism, as Yoneda does.
The loss is of a familiar kind. Equal entropies do not give equal relational structure (CLAIM-043), and a factorization fixes its latent coordinates only up to a gauge (CLAIM-148). In each case an invariant is kept and structure is discarded.
Kinds, not tokens
Take V irreducible over ℂ. Inside V ⊕ V, the subrepresentations isomorphic to V correspond to the lines in ℂ², so there are infinitely many, among them V ⊕ 0, 0 ⊕ V and the diagonal. Every one has the character χ_V. What tells them apart is how each is embedded in the whole, which is a relation to the larger structure. A194 §6 states the moral: “A group character classifies the representation up to equivalence. It does not generally individuate each vector.”
So a character can classify a kind. It cannot individuate a token or a position. That is why the owner’s two theses at U54 come apart. The character analogy is evidence for the kind claim (CLAIM-137), bounded by the results above. It is no evidence for the individuation claim (CLAIM-136), which needs positions and automorphism orbits instead.
An illustration
A198 §III gives a worked model of CASE-004’s scene: “two participants, a and b, and an interaction concerning overspending”. The basis is e₁ = a→b and e₂ = b→a, where the arrow “means only ‘occupies the active evaluative position relative to.’” The group C₂ swaps the participants, ρ(s) is the swap matrix, and χ = (2, 0), the sum of the trivial and the sign characters. The arithmetic is right.
But the character carries no communicative content. Any free action of C₂ on two configurations has character (2, 0), whatever the configurations mean, and the two “organizational types” are just the two irreducible representations of C₂. The contrast A198 wants sits elsewhere. As relational structures, the one-way reprimand {a→b} is moved by the swap, so its stabilizer is trivial, while the mutual configuration {a→b, b→a} is fixed by it, with stabilizer C₂. That is a difference of stabilizers, at the level of the G-set, which is the level A198 §4 says characters forget. So A198’s “This is an exact mathematical model of what a representation-independent communicative invariant could look like” overstates it. A198’s own hedge stands: “A reprimand is not mathematically equivalent to an antisymmetric representation”.
What survives approximately
A198 §VII Stage 3, A203 §21 and A218 Ch12.5 give an elementary bound. Let S be an invertible correspondence between two d-dimensional representations of the same group, and E_g = Sρ_A(g) − ρ_B(g)S its defect in intertwining. Then ρ_B(g) = Sρ_A(g)S⁻¹ − E_gS⁻¹, so
|χ_A(g) − χ_B(g)| = |Tr(E_gS⁻¹)| ≤ d‖E_gS⁻¹‖ ≤ d‖E_g‖‖S⁻¹‖,
in the operator norm. The derivation is correct, and A198 calls it “not a new deep representation-theoretic theorem”. Its use is the factor ‖S⁻¹‖. It puts CLAIM-011’s anti-collapse condition into a character setting: “a noninvertible or badly conditioned correspondence does not support the same argument” (A198). It has the same shape as CLAIM-063’s bound on commutators, and it needs an invertible S between representations of one group, which is not the translation case.
What it does not say
- That the analogy fails. It holds in the most favourable setting, for kinds, and up to isomorphism.
- That characters are useless to the programme. Within a fixed finite group in characteristic zero they are a complete invariant of linear representations.
- Anything about lossy stochastic channels, where “full character preservation is generally unavailable” (A198 §VII).