Slit-domain and quadratic bottleneck coordinates
DefinitionErdosProblems_Erdos1041_Counterexample_Bottleneckdegree-seven-counterexampleerdos-1041polynomial-lemniscate
Defines the slit domain, projection, quotient, quadratic coordinate, and near-critical region used in the bottleneck estimate for paths and connected sets. Covering and lower-bound properties are separate theorems with their stated hypotheses.
Definition code
import Definitions.Def_ErdosProblems_Erdos1041_Counterexample_Defs
import Mathlib
import Mathlib.Algebra.Polynomial.Derivative
import Mathlib.Algebra.Polynomial.Div
import Mathlib.Algebra.Polynomial.Roots
import Mathlib.Analysis.SpecialFunctions.Complex.Log
import Mathlib.Analysis.SpecialFunctions.Pow.Real
import Mathlib.Analysis.SpecialFunctions.Trigonometric.Basic
import Mathlib.Tactic.NormNum
import Mathlib.Tactic.Ring
import Mathlib.Topology.Connected.LocallyConnected
import Mathlib.Topology.EMetricSpace.BoundedVariation
/-! External source: ani, erdosproblems.com forum thread 1041, 7 Sept 2026.
Formalisation of Lemma 2.3 and Corollary 2.4 from `ani_degree7_counterexample.tex`. -/
/-!
This module owns slice S3 and declares `s3_bottleneck_length` at the interface
name and statement, with no `sorry`.
The proof isolates two topological hypotheses. The covering hypothesis is
`s3_bottleneck_isCoveringMap`. The
connectedness hypothesis is replaced by the sharper
`bottleneck_slit_preimage_near`: the holomorphic square root of `𝒜/â` supplies
the paper's quadratic coordinate `ψ`, its two branches `ψ = ±η` give two
preimages of each slit point next to `cc`, `hzeros` makes the covering
two-sheeted, and a third preimage anywhere in the component would produce a
third preimage of a nearby point of the slit disc. So every slit preimage in
the component is one of the two local branches, hence inside the disk-criterion
radius. Neither Rouché's theorem nor the argument principle is used; see
`bottleneck_ball_subset_image` and `bottleneck_sublevel_isPreconnected`.
-/
noncomputable section
open Topology
namespace Erdos1041.Counterexample
/-! ### Two `ℝ`-on-`ℂ` scalar-action instances
`StarConvex ℝ (0 : ℂ)` elaborates its `SMul ℝ ℂ` to `Algebra.toSMul` (through
`NormedAlgebra ℝ ℂ`), and for that particular instance path Mathlib v4.29.1 does
not resolve `ContinuousSMul ℝ ℂ` or `NormSMulClass ℝ ℂ` by typeclass search.
Both are true and are proved here from `Complex.real_smul`, which rewrites the
action to multiplication by the real coercion. They are `private`: nothing
outside this file needs them. -/
private instance instContinuousSMulRealComplex : ContinuousSMul ℝ ℂ where
continuous_smul := by
have hfun : (fun p : ℝ × ℂ => p.1 • p.2) = fun p : ℝ × ℂ => (p.1 : ℂ) * p.2 := by
funext p
exact Complex.real_smul
rw [hfun]
exact (Complex.continuous_ofReal.comp continuous_fst).mul continuous_snd
private instance instNormSMulClassRealComplex : NormSMulClass ℝ ℂ where
norm_smul r z := by
rw [show r • z = (r : ℂ) * z from Complex.real_smul, norm_mul]
simp
/-- The displacement parametrisation of the paper's radial slit. -/
def bottleneckSlit (v : ℂ) : Set ℂ :=
{w | ∃ t : ℝ, 0 ≤ t ∧ t < 1 - ‖v‖ ∧
w = v + (t : ℂ) * (v / (‖v‖ : ℂ))}
/-- The slit disc, as a subtype suitable for covering-space lifting. -/
def bottleneckSlitBase (v : ℂ) : Set ℂ :=
{w | ‖w‖ < 1 ∧ w ∉ bottleneckSlit v}
/-- The component with the slit preimage removed. -/
def bottleneckSlitDomain (p : Polynomial ℂ) (cc : ℂ) : Set ℂ :=
{z | z ∈ connectedComponentIn (Omega p) cc ∧
p.eval z ∉ bottleneckSlit (p.eval cc)}
/-- The polynomial restriction used for L2. -/
def bottleneckSlitProjection (p : Polynomial ℂ) (cc : ℂ) :
bottleneckSlitDomain p cc → bottleneckSlitBase (p.eval cc) :=
fun z => ⟨p.eval z, ⟨connectedComponentIn_subset (Omega p) cc z.2.1, z.2.2⟩⟩
section Probe
end Probe
/-! ## Two complex-analysis tools
Mathlib v4.29.1 has neither Rouché's theorem nor the argument principle, and
`DiffContOnCl.ball_subset_image_closedBall` loses a factor of two that the
constants of Corollary 2.4 cannot afford. Both tools are proved here instead
from the open mapping theorem and the maximum modulus principle. -/
/-! ## The quadratic coordinate `ψ` of Corollary 2.4
The paper's inverse quadratic coordinate `χ` is built from the holomorphic
square root of `𝒜(z)/â`, which exists because the disk criterion `hdisk` puts
that quotient in `closedBall 1 (1/4)`, well inside `Complex.slitPlane`. -/
/-- The normalised quadratic factor `𝒜(z)/â`. -/
def bottleneckQuot (p : Polynomial ℂ) (cc aHat z : ℂ) : ℂ :=
(shiftQuad p cc).eval z / aHat
/-- The paper's quadratic coordinate `ψ(z) = z √(𝒜(z)/â)`. It satisfies
`p (cc + z) - p cc = â ψ(z)²`, and on the disk-criterion radius it is comparable
to `z`, so it separates the two branches of `p` at the critical point. -/
def bottleneckPsi (p : Polynomial ℂ) (cc aHat z : ℂ) : ℂ :=
z * Complex.sqrt (bottleneckQuot p cc aHat z)
/-- The open set on which `ψ` is holomorphic. -/
def bottleneckPsiDomain (p : Polynomial ℂ) (cc aHat : ℂ) : Set ℂ :=
bottleneckQuot p cc aHat ⁻¹' Complex.slitPlane
/-! ## The two local preimages lie in the distinguished component
Both preimages produced by `bottleneck_two_preimages` sit inside the sublevel
set `‖ψ‖ < s`, on which `‖p - p cc‖ = ‖â‖ ‖ψ‖² < δ`, so that sublevel set lies
inside the lemniscate. It is preconnected by the minimum-modulus tool, and it
contains `0`; so its translate by `cc` is a preconnected subset of `Ω(p)`
through `cc`, hence inside the component. -/
/-- The sublevel set of `‖ψ‖` that joins the two local preimages to `cc`. -/
def bottleneckNear (p : Polynomial ℂ) (cc aHat : ℂ) (h s : ℝ) : Set ℂ :=
{z | z ∈ Metric.ball (0 : ℂ) h ∧ ‖bottleneckPsi p cc aHat z‖ < s}
/-! ## The covering hypothesis, discharged
`hcover` in `bottleneck_length_of_covering_and_preconnected` is the statement
that `p` restricted to the slit complement inside the distinguished component is
a covering map over the slit disc. `PROOF.md` §2 gives the argument: the
restriction is proper (a polynomial is a proper map, the component is relatively
closed in the lemniscate) and a local homeomorphism (`huniq` puts the only
critical point at `cc`, whose value `p cc` is removed with the slit), and a
proper local homeomorphism out of a Hausdorff space is a covering map, which is
slice S2's `s2_isCoveringMap_of_isProperMap_of_isLocalHomeomorph`. -/
/-- The slit-complement domain sits inside the preimage of the slit disc. -/
theorem bottleneckSlitDomain_subset (p : Polynomial ℂ) (cc : ℂ) :
bottleneckSlitDomain p cc ⊆
(fun z => p.eval z) ⁻¹' bottleneckSlitBase (p.eval cc) := by
rintro z ⟨hU, hJ⟩
exact ⟨connectedComponentIn_subset (Omega p) cc hU, hJ⟩
/-! ## The fibres over the slit have at most two points
`hzeros` pins the fibre of the covering over `0` to `{b₁, b₂}`, so the covering
over the simply connected slit disc has exactly two sheets: the sections through
`b₁` and through `b₂`. Points of the slit itself are limits of the slit disc,
so a third preimage there would produce a third preimage of a nearby point of
the slit disc through its inverse chart. -/
/-! ## Lemma 2.3 and Corollary 2.4
Both topological inputs of the conditional reduction are now discharged: the
covering by `s3_bottleneck_isCoveringMap`, and the localisation of the slit
preimage by `bottleneck_slit_preimage_near`, which replaces preconnectedness of
the whole slit preimage by the sharper statement that the two branches at `cc`
already exhaust each slit fibre. -/
end Erdos1041.Counterexample
Source
Lean source: https://github.com/wcook04/plectis-erdos-lean/blob/cc7e541cf2081c6fef5a5e377d52e365e33b01eb/ErdosProblems/Erdos1041/Counterexample/Bottleneck.lean#L1-L1585
Construction by ani: https://www.erdosproblems.com/forum/thread/1041#post-8861
Related paper and provenance: https://github.com/wcook04/plectis-erdos/blob/551bae6dc6e732cf85172d66323c8d2bc77ba962/paper/1041/erdos-1041-lemniscate-newton-flow.tex#L25-L99
Paper prior-art bibliography: https://github.com/wcook04/plectis-erdos/blob/551bae6dc6e732cf85172d66323c8d2bc77ba962/paper/1041/erdos-1041-lemniscate-newton-flow.tex#L1662-L1772
AI-assisted formalization in Will Cook's project; ani is credited for the degree-seven construction. Independent correspondence of the 1958 Problem 5 wording to this modern formulation is unrecorded.