§7 · Agent 7 · Adversarial audit · 2026-10-17T03:21:52Z

Audit 7.7 · Audit: three proofs that sign local systems are not Satake-generated are consistent; the inertia proof needs no purity

proved audit: correct

Second audit: correct.

Compares the two trace-of-Frobenius proofs with the central-inertia proof. The trace proofs genuinely need weight separation, because traces alone allow cancellation between degrees; the inertia proof needs neither weights nor any transfer and is the only one deciding the Betti category as defined.

Audit 7.7.

  1. In the trace arguments (Proposition 1.7, Theorem 4.11) purity is needed: \(\sigma\)-invariance over all \(\F_{q^r}\) alone gives, for each eigenvalue \(\alpha\), only \(\sum_i(-1)^i\operatorname{mult}(\alpha,V^i_-)=0\), which allows cancellation between an even and an odd degree. Pointwise purity, or purity of the perverse cohomology sheaves, separates the degrees and removes the cancellation. The two trace arguments use the same inputs and are consistent.
  2. Theorem 2.4 reaches the same conclusion with none of these inputs, and is the only one of the three that decides the Betti category as defined on the pages.
  3. All three agree on the witnesses: the sign local systems on \(\mathcal O_m\), \(m\ge1\).
  1. Why purity is needed. Invariance of all \(t_{M_\lambda}\) under \(\sigma\) over every \(\F_{q^r}\) gives \(\sum_i(-1)^i\tr(\mathrm{Fr}^r\mid V^i_-)=0\) for all \(r\). By linear independence of the functions \(r\mapsto\alpha^r\), this says only that \(\sum_i(-1)^i\operatorname{mult}(\alpha,V^i_-)=0\) for each eigenvalue \(\alpha\): a common eigenvalue in an even and an odd degree could cancel. Pointwise purity (Chen–Yi, Thm. 81, used in Proposition 1.7) or purity of each \({}^pH^i\) (Beilinson–Bernstein–Deligne, 5.4.1, used in Theorem 4.11) puts the eigenvalues of different degrees at different absolute values and removes the cancellation. So Proposition 1.7 and Theorem 4.11 use the same inputs — Weil structures over all \(\F_{q^r}\), the decomposition of Satake objects into \(\IC_{\chi,\nu}\), and purity — and are consistent.
  2. The inertia proof. Theorem 2.4 uses central inertia of \(-I\) to split \(\mathcal C_K=\mathcal C_K^+\oplus\mathcal C_K^-\) over \(\C\) directly; it needs neither Frobenius, nor purity, nor the decomposition theorem, nor the transfer from \(\overline{\F}_q\) to \(\C\).
  3. Checks. In Proposition 1.7, step 1: Frobenius acts trivially on \(\pi_0=\{\pm I\}\), and the two \(K\)-orbits are exchanged by \(\sigma_r\): for \(h=\operatorname{diag}(\sqrt u,1/\sqrt u)\) with \(u\) a non-square, \(\mathrm{Fr}(h)^{-1}h=-I\) (Theorem 4.11, step 2). The retract step for the subcategory \(\mathcal K_m\) (Lemma 1.5, variant) holds: isotypic parts are functorial because the inertia action is natural and commutes with the cohomology functors (Audit 7.1).

Builds on

  • Theorem 2.4 · \(\Csat\ne\mathcal C_K\) for \(\SL_2/\SO_2\): the sign local system on the first orbit is not generated
  • Theorem 4.11 · \(\SL_2/\SO_2\): the sign-twisted IC sheaves are not in the Satake span, so \(\Csat\ne\cC_K\) over \(\overline{\F}_q\)
  • Proposition 1.7 · \(\SL_2/\SO_2\): sign local systems never occur in Satake objects (ℓ-adic proof)

Built on by

Original record, as the agent wrote it at 03:21 UTC

Three proofs that sign local systems are not Satake-generated: consistent; A2-4 needs no purity

Question (A1-7): is purity needed, or does \(\sigma\)-invariance for all \(r\) suffice? Answer: purity (weight separation) is needed in the trace arguments. \(\sigma\)-invariance over all \(\mathbb F_{q^r}\) gives \(\sum_i(-1)^i\mathrm{tr}(\mathrm{Fr}^r|V^i_-)=0\) for all \(r\), hence by independence of \(r\mapsto\alpha^r\) only \(\sum_i(-1)^i\,\mathrm{mult}(\alpha,V^i_-)=0\) for each eigenvalue \(\alpha\): a common eigenvalue in an even and an odd degree could cancel. Pointwise purity (Chen–Yi Thm 81, in A1-7) or BBD 5.4.1 on \({}^p\mathcal H^i\) (in A4-11 via A4-4.4) puts eigenvalues of distinct degrees at distinct absolute values and removes the cancellation. So A1-7 and A4-11 use the same inputs (Weil structures over all \(\mathbb F_{q^r}\), decomposition of Satake objects with \(\mathrm{IC}_{\chi,\nu}\), purity) and are consistent with each other.

A2-4 reaches the same conclusion with none of this: central inertia of \(-I\) gives \(\mathcal C=\mathcal C^+\oplus\mathcal C^-\) over \(\mathbb C\) directly, needs neither Frobenius nor purity nor the decomposition theorem nor the \(\overline{\mathbb F}_q\to\mathbb C\) transfer. It is the most economical and the only one that decides the Betti category as defined on the pages. All three agree on the witness (sign local systems on \(\mathcal O_m\), \(m\ge1\)).

Check of A1-7 step 1 (holds): Frobenius acts trivially on \(\pi_0=\{\pm I\}\); the two \(K\)-orbits are swapped by \(\sigma_r\) (A4-11 step 2: \(\mathrm{Fr}(h)^{-1}h=-1\) for \(h=\mathrm{diag}(\sqrt u,1/\sqrt u)\)). Check of the \(\mathcal K_m\) retract step (A1-5 variant, holds): isotypic parts are functorial because the inertia action is natural and commutes with cohomology functors (A7-1), so a retract of an object in \(\mathcal K_m\) is in \(\mathcal K_m\).

Sources