Tower Zero-Counter
One rung up from the single exponential: functions f(x) = p(eˣ, e^{eˣ}) — polynomials in y₀ = eˣ and y₁ = e^{eˣ}. That they have only finitely many zeros is proven, rolle-only, in MachLib. But the explicit constant grows exponentially in the tower degree — climb the presets and watch the actual count barely move while the bound's growth takes off.
Depth 1: no e^{eˣ}. The classic single-exp case — bound is polynomial. Interval [-3, 3]. Amber dots = numerically-found sign changes.
sorry, single analytic axiom rolle.y₀ = eˣ = 2 at x = ln 2 ≈ 0.69. This one is y₁-free — a pure single-exponential (depth 1).
Why the bound explodes up the tower
The constructive proof counts zeros by repeatedly reducing f to a simpler function and adding +1 per step (Rolle). Two facts about those steps — both machine-checked this session — set the growth:
- The reduce multiplier carries a
y₀ = eˣfactor, so each reduce raises the inner (y₀) degree by exactly 1 (degreeY0_chain2Reduce_le). - A
y₁-degree-dropping trim then switches to a lower coefficient those reduces have already grown — so the per-level work compounds across the D₁ tower levels.
The step-count recurrence S(j+1) ≤ (dₓ+2)·S(j) + c over the tower levels solves to exp(D₁). Contrast the single-exponential case (Khovanskii Counter), whose bound length + Σ deg is merely polynomial. The finiteness was recently lifted to any exponential-type Pfaffian chain — the write-up.