# Camacho–Sad index on toy singularities — 14 Nov 2026 (computation stretch 4)

Labels: **[known]** published with source; **[project]** our own check; **[speculative]** none here. No RH claim; no cross-domain analogies. Script `compute/camacho_sad.py`, results `compute/camacho-sad-results.json`, figure `figures/camacho-sad-toy.{svg,png}`.

## Convention (stated honestly)
**[known]** Brunella, *Birational Geometry of Foliations*, Ch. 3 (I am recalling this from memory, not re-reading the book this stretch; the primer §3.3 cites Ch. 1 for reduction and [CS] for the index). For an invariant curve S={f=0}, write the foliation 1-form ω = g·df + f·η near p. Then CS(F,S,p) = −Res_p(η/g |_S). For a compact invariant curve C: Σ_p CS(F,C,p) = C·C. [CS] Camacho–Sad 1982, Ann. Math. 115.
Sanity check of sign: the formulas below reproduce λ₂/λ₁ for {y=0} (the classical value) and the sum −1 = E·E on the exceptional divisor, so the sign convention is self-consistent with the sum rule. **To do:** confirm wording against the book's text.

## 1. X = λ₁x∂ₓ + λ₂y∂ᵧ, ω = λ₂y dx − λ₁x dy
**[project], exact (sympy):** CS({y=0}) = λ₂/λ₁, CS({x=0}) = λ₁/λ₂. Matches [CS]. (Symbolic residue, assertions pass.)

## 2. Blow-up of the origin
Chart x=u, y=uv: ω̃ = ω/u = (λ₂−λ₁)v du − λ₁u dv; E={u=0}, g=(λ₂−λ₁)v, η=−λ₁dv ⇒ **CS(E,p₁)=λ₁/(λ₂−λ₁)**. Chart x=uv, y=v: **CS(E,p₂)=λ₂/(λ₁−λ₂)**. Sum = **−1 = E·E** (exact sympy). Strict transform of {y=0}: CS = λ₂/λ₁ − 1 (drops by 1, as expected when blowing up a point of the curve).
Numerical support, [project]: (a) mpmath `iv` interval boxes (λ₁,λ₂ ranges incl. widths 1e-4) — the interval for CS₁+CS₂ always contains −1 (this verifies the algebra, not the residue theory, and the widening for boxes is dependency overestimate); (b) contour quadrature of −(1/2πi)∮η/g at complex λ=(1+0.5i, −0.7+2i) agrees with the formula to ~30 digits. These checks are *not* an independent proof of the CS theorem.

## 3. Resonances and Seidenberg (correction to the brief)
- **Ratio −1/n (λ₁=n, λ₂=−1):** resonant (Poincaré–Dulac monomials x(xyⁿ)ᵏ∂ₓ, y(xyⁿ)ᵏ∂ᵧ exist), but a *negative* ratio is **reduced** in Seidenberg's sense (reduced = ratio not in ℚ_{>0}, per primer §3.3 and [Bru]). So it does **not** need blow-up; CS indices −1/n and −n (computed n=1..4). It illustrates that resonance ≠ non-reduced. The brief's premise for this case was inaccurate.
- **Genuinely non-reduced: ratio n ∈ ℤ_{>0}, X = x∂ₓ + (ny + xⁿ)∂ᵧ** (Poincaré–Dulac normal form, not linearizable for n≥2). Sympy blow-up chain (chart y=ux): linear parts have ratio n → n−1 → … → 1 (x∂ₓ+(x+y)∂ᵧ, non-diagonalizable) → after n blow-ups X = x∂ₓ + ∂ᵧ, a regular point. Verified n=1..4. Only the chart containing the direction of the resonant term was followed; the other chart's singularity was not analysed here. CS of invariant {x=0} is 1/n for the whole family (resonant term invisible to CS). This is the sense in which non-reduced points need Seidenberg [Sei] 1968: the local normal form is not linear; repeated blow-ups are required to reach reduced models. This is an illustration of the mechanism, not a proof of Seidenberg's theorem.

## Not done
Trace-formula/Poisson check on the circle toy (alternative). No Arb certificate (mpmath iv used; python-flint's Arb available if wanted). No deploy.
