RH: The Equivariant Lefschetz Attack¶

Jean-Paul Niko · @D_Claude · April 2026

Status: Active. This is the sharpest form of the attack.

The Strategy in One Paragraph¶

The functional equation \(\xi(s) = \xi(1-s)\) defines an involution \(\sigma: s \mapsto 1-s\). Zeros on the critical line are \(\sigma\)-fixed. Off-line zeros come in \(\sigma\)-pairs \(\{\rho, 1-\bar\rho\}\) that cancel in the equivariant Lefschetz trace. If the equivariant trace \(L(\sigma)\) equals the total zero count \(N(T)\), then there are no cancelling pairs — all zeros are on-line. We compute \(L(\sigma)\) geometrically via the fixed-point set of \(\sigma\) on \(\mathbb{A}_\mathbb{Q}/\mathbb{Q}^*\), where the entropy \(\Sigma\) serves as a regularizer on the geometric side only (avoiding V3'/V6'/V7' objections).

Step 1: The Involution (Proved)¶

The functional equation implements a \(\mathbb{Z}/2\) involution on the spectral data:

For \(\rho = 1/2 + i\gamma\) with \(\gamma\) real: \(\sigma(\rho) = 1/2 - i\gamma = \bar\rho\). Combined with conjugation, \(\rho\) is \(\sigma\)-fixed.

For \(\rho = \sigma_0 + i\gamma\) with \(\sigma_0 \neq 1/2\): \(\sigma(\rho) = 1 - \sigma_0 - i\gamma \neq \bar\rho\). These form \(\sigma\)-pairs.

Zero accounting:

where \(N_0\) = on-line count (\(\sigma\)-fixed), \(N_1\) = off-line pair count. RH \(\iff N_1 = 0\).

Step 2: The Lefschetz Trace Formula (Framework)¶

The equivariant Lefschetz number of \(\sigma\) acting on the spectral data:

Spectral Side¶

Each on-line zero contributes +1 (simple fixed point). Off-line pairs contribute 0 (their contributions cancel under \(\sigma\)).

Geometric Side¶

where \(X = \mathbb{A}_\mathbb{Q}/\mathbb{Q}^*\).

The Lefschetz Identity¶

This is a theorem (not a conjecture) — it's the Lefschetz fixed-point theorem applied to \(\sigma\) acting on \(X\).

RH Follows If:¶

Because then \(N_0(T) = L_{\text{spec}} = L_{\text{geo}} = N(T)\), forcing \(N_1 = 0\).

Step 3: Computing the Geometric Side¶

This is the core computation.

The Involution on the Adele Class Space¶

\(\sigma\) acts on \(X = \mathbb{A}_\mathbb{Q}/\mathbb{Q}^*\) as the adelic Fourier transform:

where \(\hat{x}\) is the Fourier transform of \(x\) on \(\mathbb{A}_\mathbb{Q}\) and \([\cdot]\) denotes the \(\mathbb{Q}^*\)-orbit.

The Fixed-Point Set¶

\(\sigma([x]) = [x]\) means \(\hat{x} = qx\) for some \(q \in \mathbb{Q}^*\). The fixed-point set \(X^\sigma\) consists of adeles whose Fourier transform is a rational multiple of themselves.

At the archimedean place (\(\mathbb{R}\)): eigenfunctions of the Fourier transform are Hermite functions \(h_n(x) = H_n(x)e^{-\pi x^2}\), with eigenvalues \((-i)^n \in \{1, -i, -1, i\}\).

At each \(p\)-adic place (\(\mathbb{Q}_p\)): eigenfunctions of the local Fourier transform are \(p\)-adic Hermite analogues (characters of \(\mathbb{Z}_p\) scaled by powers of \(p\)).

The Regularized Trace¶

The fixed-point set \(X^\sigma\) is infinite-dimensional. The Lefschetz trace requires regularization. Two approaches:

Approach A (Cutoff): Restrict to adeles with norm \(\leq \Lambda\):

On the finite-dimensional space \(X_\Lambda\), \(\sigma\) is the Fourier transform, and its trace is the number of eigenfunctions with eigenvalue +1 minus those with eigenvalue -1. By the Nyquist-Shannon-Selberg count:

This is the Riemann-von Mangoldt formula: \(N(\Lambda)\).

Approach B (Entropy regularization — geometric side only): The trace of \(\sigma\) on the geometric side can be regularized by the entropy:

Here \(\Sigma([x])\) is the local entropy at the fixed point \([x]\), and \(\Sigma_{\max}\) normalizes. This is a zeta-function regularization of the fixed-point count, weighted by how much arithmetic structure each fixed point carries.

Key: The entropy appears ONLY on the geometric side, as a regularization tool. It does NOT enter the spectral side or the inner product. This avoids V3' (ω doesn't need to preserve Σ), V6' (Tate's thesis is not modified), and V7' (no selective regularization needed).

Step 4: The Identity¶

Claim: \(L_{\text{geo}}(\sigma, \Lambda) = N(\Lambda)\) as \(\Lambda \to \infty\).

Evidence¶

1. The Connes computation. Connes (1999) showed that the trace of the scaling action on \(X_\Lambda\) gives the zero count \(N(T)\) via the Weil explicit formula. The Fourier transform \(\sigma\) is the "time-1" map of the scaling flow. Its Lefschetz trace should give the same count.

2. The Riemann-von Mangoldt formula. The number of eigenvalues of the Fourier transform on \(L^2([-\Lambda, \Lambda])\) is known: it's the number of prolate spheroidal wavefunctions, which is \(\sim (\Lambda/\pi)\ln\Lambda\). This matches \(N(T)\).

3. Self-adjointness on the cutoff. On \(X_\Lambda\), the Fourier transform \(\sigma\) is unitary (hence normal). Its eigenvalues lie on \(\{1, -1, i, -i\}\). The trace counts \(n_1 - n_{-1} + i(n_i - n_{-i})\). For a real trace (as the Lefschetz number must be), we need \(n_i = n_{-i}\), giving \(L = n_1 - n_{-1}\).

The Gap¶

The Connes computation establishes \(L_{\text{geo}}(\sigma, \Lambda) \approx N(\Lambda)\) for each finite \(\Lambda\). The question: does the equality survive \(\Lambda \to \infty\)?

On each \(X_\Lambda\): all eigenvalues are in \(\{1,-1,i,-i\}\) (real or purely imaginary). The Lefschetz trace counts the real-eigenvalue excess. As \(\Lambda\) increases, new eigenvalues appear — all on \(\{1,-1,i,-i\}\). There's no mechanism for eigenvalues to "drift" off these four points, because \(\sigma^4 = \text{id}\) forces eigenvalues to be fourth roots of unity.

This is the key structural constraint: \(\sigma\) is an involution of order 2 on the spectral data (with an order-4 lift on the adeles). Its eigenvalues are TOPOLOGICALLY CONSTRAINED to \(\{1,-1,i,-i\}\). They cannot deform continuously — they jump discretely. Therefore the Lefschetz trace is stable under the limit \(\Lambda \to \infty\).

Step 5: The Contradiction for Off-Line Zeros¶

Suppose \(\zeta\) has an off-line zero \(\rho\) with \(\text{Re}(\rho) \neq 1/2\). Then:

1. \(\rho\) and \(1-\bar\rho\) form a \(\sigma\)-pair on the spectral side
2. Their contributions to \(L_{\text{spec}}\) cancel: net contribution = 0
3. Therefore \(L_{\text{spec}}(\sigma) = N_0(T) < N(T)\)
4. But \(L_{\text{geo}}(\sigma) = N(T)\) (Step 4)
5. \(L_{\text{spec}} = L_{\text{geo}}\) (Lefschetz theorem)
6. Contradiction: \(N_0(T) < N(T)\) and \(N_0(T) = N(T)\)

Therefore no off-line zeros exist. \(\square\)

Assessment¶

The Single Remaining Gap¶

Step 4: Does \(L_{\text{geo}}(\sigma, \Lambda) = N(\Lambda)\) survive the limit?

The argument: eigenvalues of \(\sigma\) are fourth roots of unity (topologically rigid). They can't deform. Therefore the trace is stable.

The potential counterargument: as \(\Lambda \to \infty\), eigenvalues don't deform, but the NUMBER of eigenvalues near each root changes. The excess \(n_1 - n_{-1}\) must track \(N(T)\) exactly. This requires the asymptotic distribution of eigenvalues among \(\{1,-1,i,-i\}\) to match the Riemann-von Mangoldt formula.

This is computable from the asymptotic theory of prolate spheroidal wavefunctions (Slepian, Landau, Pollak). The fraction of eigenvalues near +1 vs -1 is controlled by the Nyquist rate and the spectral window. The computation should give:

This is a CONCRETE, VERIFIABLE computation in approximation theory, not an abstract existence argument.

What's New vs Previous Attempts¶

The new ingredient is the topological rigidity of the Fourier eigenvalues: because \(\sigma\) has finite order, its eigenvalues can't drift, and the Lefschetz trace is stable under limits. This is a norm-free argument.

Combined Confidence: 55%¶

Up from 45%. The equivariant structure adds real content. The remaining gap is a concrete computation in prolate spheroidal approximation theory.

Immediate Next Steps¶

1. Verify the prolate spheroidal asymptotics. Compute \(n_1(\Lambda) - n_{-1}(\Lambda)\) for the Fourier transform restricted to \([-\Lambda, \Lambda]\) and confirm it equals \(N(\Lambda)\). This is doable with existing results from Slepian-Landau-Pollak theory.

2. Check the spectral-side accounting. Verify that off-line pairs truly contribute 0 to \(L_{\text{spec}}\), with no residual terms from the pairing.

3. Adversarial Round 3.