Principia Orthogona · Book IV · Higher Dimensions Contents · ← Ch 23
Book IV · Chapter 24 · The Algebraic Turn

The Bifurcation Set Is a Discriminant, or What the Search Was Never For

Arc · Higher Dimensions, rung 23
Prerequisite · Ch 23 (discriminants, dual varieties); Ch 3 (System 3.1)
Status · one elementary theorem; two limits stated as limits
Verification · ch24-verify.py · 9 blocks, all passing
Chapter 23 closed on one equation wearing three names: 4a³ + 27b² is the dual of the cuspidal cubic, the discriminant of x³ + ax + b, and the bifurcation set of the cusp catastrophe. This chapter takes the third name seriously. For a family of planar radial fields the bifurcation set is the discriminant of the radial polynomial — not by analogy, by identity — and the Floquet multiplier is exactly 1 on it, because a multiple root is exactly where f and f′ vanish together. Deformed inside its own odd family, the dm³ field folds at ε = 1/4, at radius √2. That value is an exact rational obtained by one resultant, without integrating anything. Sixty steps of bisection on the flow return 0.2499966. The search was never necessary. Along the way the chapter records what the dm³ attractor is, how far the planar reduction of System (3.1) may be pushed, and how many digits of the basin figure may be quoted.

Contents

  1. Reeb versus Legendrian: what the attractor is
  2. The theorem
  3. The dm³ field, deformed inside its own family
  4. Numerics: the cycles and the multiplier
  5. εc by search, εc by discriminant
  6. The e−z coupling, and what the reduction is
  7. The basin figure, and how much of it may be quoted
  8. The same curve, one chapter earlier
  9. Honest inventory
§ 24.1

Reeb versus Legendrian: what the attractor is

The contact form carried through Book IV is α = dz − r²dθ on M = ℝ²+ × ℝ, with dα = −2r dr ∧ dθ. The Reeb field is the unique R with ιRdα = 0 and α(R) = 1. Writing R = a ∂r + b ∂θ + c ∂z:

ιR dα = −2r ( a dθ − b dr ) = 0  ⇒  a = b = 0  (r ≠ 0)
α(R) = c − r²b = c = 1 ⇒ R = ∂z, uniquely

The dm³ attractor is γ*(t) = (1, t, t) in (r, θ, z). Its tangent is γ̇ = (0, 1, 1), and

α(γ̇) = ż − r²θ̇ = 1 − r² = 0 at r = 1

So γ* lies in ker α. It is Legendrian. The Reeb field is transverse to ker α and a Legendrian curve lies inside it; a curve cannot be both, and the two conditions — α(γ̇) = 1 against α(γ̇) = 0 — are complementary rather than merely different.

Chapter 3 carries this already: its text reads “It is Legendrian, not Reeb”, with the tangent and the pairing written out. What is added here is not the correction but the computation behind it. Block [1] solves for R symbolically from the two defining conditions and then evaluates α(γ̇), rather than asserting either.

The same block records a second thing, easily lost. z(t) = t is unbounded, so γ* is a helix in M and not a closed curve. T* = 2π is the period of the (r, θ) reduction, not of γ*. Every period quoted in this chapter is that one, and no statement below is a statement about a closed orbit in M.

§ 24.2

The theorem

Take a planar system in polar form with the angular speed normalised:

ṙ = f(r; λ),   θ̇ = 1 f polynomial in r, λ the parameters

A limit cycle is a positive root r* of f. Along it the transverse variational equation is d(δr)/dt = f′(r*) δr, integrated over T = 2π, so the Floquet multiplier is m = exp(2π f′(r*)). The cycle is non-hyperbolic exactly when m = 1, which is exactly when f′(r*) = 0 — and since r* is already a root, exactly when r* is a multiple root of f.

For the family ṙ = f(r; λ), θ̇ = 1 with f polynomial in r, the bifurcation set in parameter space is { λ : discr f(·; λ) = 0 }, and the Floquet multiplier equals 1 at every point of it.

Proof

The discriminant of a polynomial vanishes precisely when it has a multiple root; a multiple root of f is precisely a common root of f and f′; a common root is precisely a cycle with f′(r*) = 0, hence m = e0 = 1. Each step is an equivalence, so the three descriptions cut out the same set. □

Nothing dynamical survives in the statement. The left-hand side is defined by the behaviour of a flow; the right-hand side is computed by elimination in exact arithmetic, from the coefficients, without integrating the flow once.

§ 24.3

The dm³ field, deformed inside its own family

The dm³ radial field is ṙ = r(1 − r²), whose positive root is r* = 1. Deform it by the next odd term, so the deformation stays inside the family the original belongs to:

f(r) = r(1 − r² + εr⁴) ε = 0 recovers dm³ exactly
u = r²,  f = r g(u),  g(u) = 1 − u + εu²
discu(g) = 1 − 4ε block [3]

The discriminant is linear. It vanishes at εc = 1/4 and nowhere else, the double root there is u* = 2, and so the fold sits at r* = √2. The multiplier check is immediate: f′(r) = 1 − 3r² + 5εr⁴, and at (r, ε) = (√2, 1/4),

f′(√2; 1/4) = 1 − 3(2) + 5(1/4)(4) = 1 − 6 + 5 = 0
m = exp(2π · 0) = 1 exactly, as Theorem 1 requires

Two cycles exist for 0 < ε < 1/4 — the stable one continuing the dm³ cycle from r = 1, and an unstable one descending from infinity — and they annihilate at the fold. Above εc there is no positive root at all: g has no real roots, and every trajectory escapes.

0 0.05 0.10 0.15 0.20 0.25 0 1 2 3 ε r* stable cycle unstable cycle fold: ε = 1/4, r* = √2 dm³: ε = 0, r* = 1 no cycle →
The fold, drawn from the roots and not from the flow. Both branches are r = √u with u = (1 ± √(1 − 4ε)) / 2ε; the discriminant 1 − 4ε under the radical is the whole diagram. The two branches meet where it vanishes, at ε = 1/4, u = 2, r* = √2 — and there, and only there, the multiplier is 1. To the right of the dashed line g has no real root and no cycle exists.
§ 24.4

Numerics: the cycles and the multiplier

The algebraic claims above are checked against integration. Block [4] runs the radial equation from r(0) = 0.5 to t = 400 at rtol 10−12 and compares the endpoint with √u:

εr* by integrationr* algebraic|error|
0.0001.0000000000001.0000000001.92 × 10−13
0.1001.0616104058421.0616104061.59 × 10−13
0.2001.1755705045851.1755705059.75 × 10−14
0.2401.2909944487351.2909944495.14 × 10−13
0.2491.3715075740571.3715075741.43 × 10−13

Block [5] does the same for the multiplier, integrating the variational equation over one period T = 2π and comparing with exp(2π f′(r*)):

εr*m by variational integrationexp(2π f′(r*))
0.00001.0000000003.4873423564 × 10−63.4873423562 × 10−6
0.20001.1755705054.2371726419 × 10−44.2371726419 × 10−4
0.24001.2909944491.5164619865 × 10−21.5164619865 × 10−2
0.25001.4142135621.00000000001.0000000000

The last row is the theorem. The multiplier climbs across four orders of magnitude as ε approaches the fold — from e−4π at the dm³ cycle, where contraction is fierce, to 1 — and the exact place it arrives at 1 was known before the first integration was run.

§ 24.5

εc by search, εc by discriminant

Block [6] does it the other way: brackets the fold between ε = 0.20, where a cycle exists, and ε = 0.30, where none does, and bisects sixty times, deciding at each step by integrating to t = 300 and asking whether the trajectory settled on a finite root.

εc by bisection : 0.249996601832319
εc = 1/4 exactly: 0.250000000000000 gap 3.40 × 10−6

Sixty bisection steps, each one an initial-value problem integrated to t = 300, and the answer is correct to six figures. The discriminant is correct to all of them, and it is a rational number, and it took one resultant. The gap is not an indictment of the search — near a fold the surviving cycle’s basin narrows and its approach slows, so any decision rule based on a finite horizon degrades exactly where the answer matters most. That is a property of the method, not of the integrator.

This is the chapter in one line. The bifurcation set was never a thing to be found by looking. It is the zero set of a polynomial in the coefficients, and it was available before the model was integrated once.

§ 24.6

The e−z coupling, and what the reduction is

Everything above is planar. System (3.1) is not: it carries a coupling that ties the radial equation to z.

ṙ = r(1 − r² + εr⁴) + 2(r − 1)e−z
θ̇ = 1
ż = r² − 2(r − 1)²e−z

Block [7] integrates the full three-dimensional system from (0.9, 0, 0) to t = 400 and compares the radius reached with the planar algebraic root:

ε3-D r(400)planar r*|difference|z(400)
0.0001.0000000000001.0000000000004.49 × 10−13400
0.2001.1755705045861.1755705045851.05 × 10−12552
0.2401.2909944487361.2909944487363.96 × 10−13666

The coupling decays like e−z, and z grows without bound on the attractor. So the planar reduction is asymptotic: exact in the limit z → ∞, and not at any finite z. The table is a measurement at one horizon, not a bound. What it licenses is the statement that the three-dimensional system inherits the planar bifurcation set in the limit; it does not license an error estimate at finite time, and none is proved here.

§ 24.7

The basin figure, and how much of it may be quoted

Block [8] regenerates the inner basin boundary at z = 0 for ε = 0 — the figure printed as r* ≈ 0.776, refined in Chapter 10 — by bisecting on whether the trajectory converges to the attractor. It is a plain float bisection driven by an integrator at rtol 10−11, atol 10−13, so its error floor is around 10−13 and digits past the twelfth are integrator noise. The rigorous value lives in certify_rstar_rigorous.py, which brackets it by mpmath centre, Jacobian-linearised error transport, and an interval-Hessian Lagrange remainder:

certified : [0.775940575501953167, 0.775940575502343743] width 3.906 × 10−13
block [8] : 0.775940575502539698 above the upper bound by 1.96 × 10−13

The float value lies outside the certified bracket. It is not inside it and the check does not pretend otherwise: the twelve leading digits are identical — 775940575502 — and the two differ at the thirteenth, by an amount comparable to the width of the bracket itself. Eleven significant figures agree under rounding; that is the whole of what may be claimed.

Block [8] passes, and what it asserts is worth stating exactly, because a stronger assertion was available and would have been false. It checks that the float bisection sits within its own error floor of the certificate, and that the first eleven significant figures match. A check demanding that the float land inside the rigorous bracket would fail — and it would be the check that was wrong, not the number.

Chapter 3 quotes ≈ 0.776 and Chapter 10 establishes it. Both are safe at that precision. Anything past eleven significant figures must be cited from the certificate and not from this block.

§ 24.8

The same curve, one chapter earlier

Block [9] closes the loop with Chapter 23. The discriminant of the depressed cubic and the bifurcation set of the quartic potential are the same curve:

discx(x³ + ax + b) = −(4a³ + 27b²)
V(x) = x⁴/4 + ax²/2 + bx,  Res(V′, V″) = 4a³ + 27b² the cusp catastrophe

The sign is a convention on the discriminant and does not move the zero locus. Chapter 23 obtained 4a³ + 27b²c as the dual of the cuspidal cubic, and 4a³ + 27b² in the chart c = 1; its block [12] noted the coincidence and pointed here. The elimination performed there on a family of plane curves and the elimination performed here on a family of vector fields are the same elimination. Dual variety, discriminant, bifurcation set: what changes between the three names is what the parameters are called, and nothing else.

That is the sense in which the ladder has become algebraic. A question posed about the long-run behaviour of a flow has been answered by a resultant, and the answer is a rational number rather than a converged search.

§ 24.9

Honest inventory

ClaimStatusWhat it rests on
Reeb field of α = dz − r²dθ is ∂zComputedblock [1], solved from the two defining conditions
the dm³ attractor is Legendrian, not a Reeb orbitComputedblock [1]; α(γ̇) = 1 − r² = 0 at r = 1
γ* is a helix; T* = 2π is the reduction’s periodComputedblock [1]; z(t) = t is unbounded
bifurcation set = { discr f = 0 }, multiplier 1 on itProved§24.2; three equivalences, elementary
m = exp(2π f′(r*))ClassicalFloquet theory for the transverse variational equation
discu(1 − u + εu²) = 1 − 4εComputedblock [3]
εc = 1/4, u* = 2, r* = √2Computedblock [3], exact rational
f′(√2; 1/4) = 0, so m = 1 exactly at the foldComputedblock [3]
cycle radii agree with the roots to 10−9Computedblock [4], five values of ε
multipliers agree to 10−6 relativeComputedblock [5], four values of ε
bisection lands 3.40 × 10−6 from 1/4Computedblock [6], 60 steps
the 3-D system inherits the planar root at t = 400Computedblock [7]; a horizon, not a bound
the planar reduction is exact at finite zFALSEcoupling is e−z; exact only as z → ∞
a finite-z error bound for the reductionOPENnot attempted here; block [7] measures, it does not bound
basin boundary ≈ 0.776 at z = 0Computedblock [8]; Ch 10 refines, Ch 3 quotes
block [8]’s float value lies inside the certified bracketFALSEabove it by 1.96 × 10−13; eleven figures is the limit
4a³ + 27b² is dual, discriminant and bifurcation setComputedblock [9]; Ch 23 block [12]
the dm³ flow algebraisesOPENcarried from Ch 23: the helix z = θ is transcendental in these coordinates

Every row marked Computed is regenerated by ch24-verify.py, which sits beside this chapter and exits non-zero on any failure. Two rows are marked FALSE deliberately. They are claims a reader might reasonably assume the chapter is making, and it is not: the planar reduction is a limit, and the basin figure has eleven honest significant figures rather than eighteen. A verification script that reported neither would still exit zero, and would be worth less for it.

FloquetThe multiplier of a hyperbolic limit cycle as the exponential of the integrated transverse divergence is standard; see Perko, Differential Equations and Dynamical Systems, §3.5. In polar form with θ̇ = 1 the integral collapses to 2π f′(r*).
DiscriminantsGelfand–Kapranov–Zelevinsky, Discriminants, Resultants and Multidimensional Determinants. Ch. 1 for the identification used in Ch 23; the parameter-space reading here is the A-discriminant of the coefficient family.
CatastropheThe cusp is Thom’s; Arnold, Catastrophe Theory, gives the bifurcation set of x⁴/4 + ax²/2 + bx as 4a³ + 27b² = 0 directly. Block [9] recomputes it as a resultant so that the identity with Ch 23’s dual variety is machine-checked and not quoted.
The certificatecertify_rstar_rigorous.py is the only citable source for the basin boundary past eleven significant figures. Block [8] is a float computation and says so in its own output.
Why fold, not HopfThe deformation here is radial and keeps θ̇ = 1, so the origin’s stability never changes and no Hopf bifurcation occurs. What folds is a pair of cycles against each other — a saddle-node of periodic orbits — which is why the discriminant of f, and not the linearisation at a fixed point, is the object that detects it.
Where this goes nextThe two open rows are different in kind. A finite-z bound for the reduction is a Grönwall estimate someone can sit down and do. Whether the dm³ flow algebraises is not: the structure algebraises (Ch 23, §23.7), and the attractor is Legendrian, but the helix z = θ is transcendental in these coordinates and no change of coordinates that fixes both has been exhibited.
Principia Orthogona · Book IV · Chapter 24
Pablo Nogueira Grossi · G6 LLC, Newark, New Jersey · ORCID 0009-0000-6496-2186
ISBN 979-8-9954416-8-7 · doi 10.5281/zenodo.19117400
← Ch 23 · Duality and the Discriminant · Contents
Proved · kernel-checked
discriminant book21/Spiral.lean:75 Each name above is declared in this repository at the line shown and appears in an axiom report with no sorryAx. A clean axiom report is not a reading of the statement: per R20, a theorem can assume its conclusion and still report clean. Follow the link before citing one as evidence.