The finite-range claim in degree five #
The lowest degree, where the exponent (n-4)/2 is 0 + 1/2, so Sendov.R 5 α involves
√Q itself.
The informal plan treats this degree separately, on the ground that bound (O) — the
tangent line at q = 1 — is too wasteful here, and proposes instead the chord bound
Q(t) ≤ 1 - (1-B)t, the resulting formula for ∫ t³ √(1-(1-B)t) dt in half-integer powers
of B, and the substitution α = 3r²/(1-r²) needed to make that rational.
None of this is necessary. The waste in (O) comes entirely from the tangent line being
taken at q = 1, whereas the values of Q that matter here are small; taking the tangent
at q = w² for a smaller rational w fixes it. With w = 1/3, Sendov.integral_rpow_le
gives √Q ≤ 3Q/2 + 1/6 and hence
∫ t in 0..1, t³ √Q dt ≤ (3/2) (1/4 - 2c/5 + A/6) + 1/24,
which is already enough. So degree five uses exactly the same machinery as every other
odd degree, with only the parameter w changed, and no square roots survive.
With M 5 = 4, A 5 α = 1 - α/2 and c 5 α = (12 - α - α²)/(4(3+α)), the resulting upper
bound for R 5 α is 1 - P α / (96 (3+α)²) with
P α = -9α⁴ - 123α³ + 596α² + 30α + 234,
which is positive up to α = 3.912, while feasibility gives α ≤ 2 outright (this is the
one degree where Sendov.alpha_le_half_M is exactly the bound A 5 α ≥ 0). For
orientation: over the feasible range α ≤ 1.678 the exact maximum of R 5 α is about
0.716, against 0.737 for this upper bound — whereas w = 1 would give 1.063, above
1, and would prove nothing.