Two reflections in folded symplectic coordinates

A pair of sheet exchanges determines more than a pair of reflection lines. When both exchanges preserve the same folded two-form, their shared tangential direction is the characteristic line of that form. We use this fact to construct canonical spectator coordinates, isolate the remaining two-dimensional change, and normalize it by an actual smooth inverse of a cubic integral. The homogeneous version also changes the last position coordinate; its cubic correction is forced by preservation of the two-form.

The primary source is the reprint of Hörmander III, corrected second printing (1994), Theorems 21.4.4 and 21.4.10, printed pages 307–309 and 316–318, PDF pages 322–324 and 331–333 in this exact edition. We supply the commuting-flow details for the spectator construction and a separate construction of the last two homogeneous positions. Folded forms and symplectic target coordinates proves the folded Darboux theorem and its single-involution version. Simultaneous reflections and flat corrections proves full simultaneous smooth and homogeneous coordinates for the two involutions, including the flat correction.

Our convention is ιHaσ=−da\iota_{H_a}\sigma=-da, {a,b}=Hab\{a,b\}=H_ab. Hamiltonian fields are initially defined off the critical hypersurface, where σ\sigma is nondegenerate. Whenever we use a field on that hypersurface, we prove its smooth extension. All maps and flows are local germs, with neighborhoods shrunk for the finitely many required compositions.

The proof map gives exact current proof dependencies for every construction and solution. The earlier flow and bundle companion F0–F2 proves the smooth transverse and commuting-flow charts used here. The differential-form companion F0–F2 and phase-space lesson supply Cartan's formula, flow pullbacks and Hamiltonian commutator identities. The smooth-descent lesson supplies division with all smooth parameters. These existing proofs are reused directly.

1. The two reflection lines identify the characteristic line

Let MM have dimension 2n2n, and let σ\sigma be a closed two-form. Near cc, assume

σn=mμ,m(c)=0,dm(c)≠0,(σn−1)∣TcΓ≠0,Γ={m=0},(1.1) \sigma^n=m\mu,\qquad m(c)=0,\quad dm(c)\ne0,\qquad (\sigma^{n-1})|_{T_c\Gamma}\ne0,\qquad \Gamma=\{m=0\}, \tag{1.1}

with μ\mu nowhere zero. Shrink so that the restricted rank condition holds along Γ\Gamma. The preceding folded-form lesson proves

E=ker⁡σ∣Γ,dim⁡E=2,K=ker⁡σΓ,dim⁡K=1,E∩TΓ=K.(1.2) E=\ker\sigma|_\Gamma,\quad \dim E=2,\qquad K=\ker\sigma_\Gamma,\quad \dim K=1,\qquad E\cap T\Gamma=K. \tag{1.2}

For n=1n=1, the restricted exterior-power condition is automatic.

Let f,gf,g be smooth involutions, both with local fixed set Γ\Gamma, such that

f∗σ=σ,g∗σ=σ,Lf(c)≠Lg(c).(1.3) f^*\sigma=\sigma,\qquad g^*\sigma=\sigma,\qquad L_f(c)\ne L_g(c). \tag{1.3}

Here Lf,LgL_f,L_g are their minus-one eigenlines along Γ\Gamma. Each is transverse to Γ\Gamma. If v∈Lfv\in L_f, invariance gives σ(v,w)=0\sigma(v,w)=0 for every w∈TΓw\in T\Gamma; the same follows for every ambient ww, since vv and TΓT\Gamma span the tangent space. Thus Lf⊂EL_f\subset E, and similarly Lg⊂EL_g\subset E. Distinctness persists locally, and consequently

E=Lf+Lg,(Lf+Lg)∩TΓ=K.(1.4) E=L_f+L_g,\qquad (L_f+L_g)\cap T\Gamma=K. \tag{1.4}

If aa is invariant under both involutions, dada annihilates both reflection lines on Γ\Gamma. It therefore annihilates KK. This gives a useful differential restriction on common invariant functions before any canonical coordinate is constructed.

The full simultaneous coordinate theorem gives another chart

f(t,z,r)=(t,z,−r),g(t,z,r)=(t+r,z,−r).(1.5) f(t,z,r)=(t,z,-r),\qquad g(t,z,r)=(t+r,z,-r). \tag{1.5}

On Γ\Gamma, the tt direction is the intersection in (1.4), hence is KK. We will use zz to extend characteristic-constant functions as common invariants.

2. Smooth Hamiltonian fields and their commuting-flow sections

First use the single-involution folded Darboux theorem to choose ordinary coordinates (x,s,ξ′)(x,s,\xi'), centered at cc, with

σ=s ds∧dx1+∑j=2ndξj∧dxj,f(x,s,ξ′)=(x,−s,ξ′).(2.1) \sigma=s\,ds\wedge dx_1+\sum_{j=2}^n d\xi_j\wedge dx_j, \qquad f(x,s,\xi')=(x,-s,\xi'). \tag{2.1}

Thus K=R∂x1K=\mathbb R\partial_{x_1} on Γ={s=0}\Gamma=\{s=0\}.

For s≠0s\ne0,

Ha=1s(as∂x1−ax1∂s)+∑j=2n(aξj∂xj−axj∂ξj).(2.2) H_a=\frac1s(a_s\partial_{x_1}-a_{x_1}\partial_s) +\sum_{j=2}^n(a_{\xi_j}\partial_{x_j} -a_{x_j}\partial_{\xi_j}). \tag{2.2}

If aa is even in ss and ax1∣s=0=0a_{x_1}|_{s=0}=0, then

as=s ass∣s=0+O(s3),ax1=O(s2). a_s=s\,a_{ss}|_{s=0}+O(s^3),\qquad a_{x_1}=O(s^2).

Division in (2.2) is smooth, and the normal component vanishes on Γ\Gamma. For the full parameter statement, Taylor's integral identity gives as(y,s)/s=∫01ass(y,vs) dva_s(y,s)/s=\int_0^1 a_{ss}(y,vs)\,dv. Evenness gives ∂sax1(y,0)=0\partial_s a_{x_1}(y,0)=0, so ax1(y,s)/s=s∫01(1−v)ax1ss(y,vs) dva_{x_1}(y,s)/s=s\int_0^1(1-v)a_{x_1ss}(y,vs)\,dv. The hypotheses remove the two constant terms. Both integrals are smooth with every derivative in y,sy,s, by the compact-parameter differentiation proof already available. Thus the extension is smooth, rather than merely bounded. In particular HaH_a is tangent to Γ\Gamma, and there

Ha=ass∣Γ ∂x1+∑j=2n(aξj∂xj−axj∂ξj)∣Γ.(2.3) H_a=a_{ss}|_\Gamma\,\partial_{x_1} +\sum_{j=2}^n(a_{\xi_j}\partial_{x_j} -a_{x_j}\partial_{\xi_j})|_\Gamma. \tag{2.3}

Every common invariant function satisfies these conditions: it is ff-even and its restriction is characteristic-constant by Section 1. Its smooth Hamiltonian field is invariant under f,gf,g, since this is true off Γ\Gamma by the contraction equation and then everywhere by continuity.

We will repeatedly solve equations along commuting fields. Suppose A1,…,AkA_1,\ldots,A_k are commuting smooth fields, independent at the marked point, and a section NN of codimension kk is transverse to their span. The actual map

(u,z)⟼exp⁡(u1A1)⋯exp⁡(ukAk)z,z∈N,(2.4) (u,z)\longmapsto \exp(u_1A_1)\cdots\exp(u_kA_k)z,\qquad z\in N, \tag{2.4}

is a local diffeomorphism. Commutativity makes the order immaterial. A function constant along all fields is uniquely determined by its data on NN; equations Aia=biA_i a=b_i with constant compatible right-hand sides are solved by adding ∑ibiui\sum_i b_i u_i. This proves the local existence and uniqueness we use, rather than assuming a simultaneous differential system has a solution.

If an involution preserves the field distribution, the section, the initial data and the differential equations with their right-hand sides, uniqueness proves invariance. Preservation of the distribution alone would not suffice for a nonzero right-hand side. Antisymmetric data give an antisymmetric solution when the equations transform accordingly. The same argument controls degrees under dilation: apply a dilation to the function, divide by its prescribed degree factor, and check the identical equations and initial data. A distribution may be preserved even when its individual fields have different weights. More explicitly, if [R,Ai]=wiAi[R,A_i]=w_iA_i, then Dκ∗Ai=κ−wiAiD_{\kappa*}A_i=\kappa^{-w_i}A_i. For a proposed degree-δ\delta solution aa, its transformed function aκ=κ−δa∘Dκa_\kappa=\kappa^{-\delta}a\circ D_\kappa satisfies Aiaκ=κ−δ−wi(Aia)∘DκA_i a_\kappa=\kappa^{-\delta-w_i}(A_i a)\circ D_\kappa. Hence the equation is preserved when its right-hand side has degree δ+wi\delta+w_i, or is zero. Together with the homogeneous initial data this proves the stated degree by uniqueness. The commuting-flow and pushforward identities used here are proved in companion F1.

For a smooth canonical pair p,qp,q with {p,q}=1\{p,q\}=1, the fields Hp,HqH_p,H_q commute, are tangent to Γ\Gamma in our applications, and span a symplectic plane even on Γ\Gamma:

σ(Hp,Hq)=−dp(Hq)=1.(2.5) \sigma(H_p,H_q)=-dp(H_q)=1. \tag{2.5}

The section p=q=0p=q=0 is transverse to that plane. Its tangent is the σ\sigma-orthogonal of the plane. The map

(q,p,z)⟼exp⁡(qHp)exp⁡(−pHq)z(2.6) (q,p,z)\longmapsto \exp(qH_p)\exp(-pH_q)z \tag{2.6}

therefore splits the full form as dp∧dqdp\wedge dq plus its restriction to the section. Indeed the two flow derivatives are Hp,−HqH_p,-H_q, their mixed pairing is −1-1 in that order, their pairings with section tangents vanish, and the flows preserve σ\sigma. This argument needs no inverse of σ\sigma on Γ\Gamma.

The contractions ιHpσ=−dp\iota_{H_p}\sigma=-dp and ιHqσ=−dq\iota_{H_q}\sigma=-dq extend to the fold by continuity. Cartan's formula therefore proves that their actual smooth flows preserve σ\sigma everywhere. Their commutator is zero off the fold by the Hamiltonian identity, and zero on it by smoothness. A field tangent to Γ\Gamma preserves Γ\Gamma under its local flow: its restriction solves the same smooth initial-value problem and uniqueness applies.

For several canonical pairs, the corresponding fields all commute. Their common zero section and the product version of (2.6) give the same splitting. All these assertions follow on the full neighborhood, not just off the fold.

3. Construct all ordinary spectator pairs

The restrictions of xj,ξjx_j,\xi_j, j≥2j\ge2, to Γ\Gamma are constant along KK. In (1.5), they are functions of zz alone on r=0r=0. Extend those functions independently of t,rt,r, and call the resulting functions Xj,PjX_j,P_j. They are common invariants and satisfy

Xj∣Γ=xj,Pj∣Γ=ξj.(3.1) X_j|_\Gamma=x_j,\qquad P_j|_\Gamma=\xi_j. \tag{3.1}

Their full differentials agree with those of the old coordinates on Γ\Gamma: tangential agreement follows from (3.1), and both have zero normal derivative because they are ff-even.

We construct p2,q2,…,pn,qnp_2,q_2,\ldots,p_n,q_n in that order. Suppose the earlier pairs have been constructed, are common invariants, satisfy all canonical bracket relations, and agree with ξi,xi\xi_i,x_i on Γ\Gamma. Let

Nj−1={p2=q2=⋯=pj−1=qj−1=0}. N_{j-1}=\{p_2=q_2=\cdots=p_{j-1}=q_{j-1}=0\}.

For j=2j=2, this is the full neighborhood. Define pjp_j by extending Pj∣Nj−1P_j|_{N_{j-1}} constantly along the earlier Hamiltonian flows. Formula (2.4) and the canonical splitting prove existence, smoothness and

{pi,pj}={qi,pj}=0,i<j.(3.2) \{p_i,p_j\}=\{q_i,p_j\}=0,\qquad i<j. \tag{3.2}

The section and seed are common invariants, so is pjp_j.

On Γ\Gamma, each earlier field changes its own old spectator coordinate and possibly x1x_1, but leaves xj,ξjx_j,\xi_j unchanged when j>ij>i. This follows from (2.3) and the induction hypothesis. Hence propagation retains pj∣Γ=ξjp_j|_\Gamma=\xi_j. It also gives

Hpj∣Γ=∂xj+Cj∂x1.(3.3) H_{p_j}|_\Gamma=\partial_{x_j} +C_j\partial_{x_1}. \tag{3.3}

The field is smooth, tangent to Γ\Gamma, and tangent to Nj−1N_{j-1} by (3.2).

On that section, solve

Hpjqj=1,qj=0on Xj=0.(3.4) H_{p_j}q_j=1,\qquad q_j=0\quad\hbox{on }X_j=0. \tag{3.4}

The initial hypersurface is transverse: at Γ\Gamma, HpjXj=1H_{p_j}X_j=1. Extend the solution constantly along the earlier fields. Commutativity makes the equations compatible. Its equation and initial hypersurface are invariant under both involutions, so qjq_j is a common invariant. For the extension off Nj−1N_{j-1}, HpjH_{p_j} commutes with every earlier field by (3.2), is tangent to the section, and its equation has the constant right-hand side one. Consequently the product flow transports that equation to the full neighborhood. The earlier zero equations and the new equation therefore hold together; no unsolved simultaneous differential system is being assumed.

On Γ∩Nj−1\Gamma\cap N_{j-1}, the old function xjx_j solves (3.4), by (3.3), with the same data. Uniqueness and subsequent propagation show qj∣Γ=xjq_j|_\Gamma=x_j. Thus

{pj,qj}=1,Hqj∣Γ=−∂ξj+Dj∂x1.(3.5) \{p_j,q_j\}=1,\qquad H_{q_j}|_\Gamma=-\partial_{\xi_j}+D_j\partial_{x_1}. \tag{3.5}

The new field is again smooth and tangent to Γ\Gamma, and its canonical plane is transverse to the new zero section. This proves the induction, including all cross brackets, all symmetries and the full section transversality. In particular the resulting 2n−22n-2 spectator functions are independent.

4. Complete the ordinary theorem and normalize the remaining shift

Theorem 4.1 (ordinary simultaneous folded form). Under (1.1)–(1.3), there are coordinates (q,p)(q,p), centered at cc, in which

σ=p1 dp1∧dq1+∑j=2ndpj∧dqj,(4.1) \sigma=p_1\,dp_1\wedge dq_1+\sum_{j=2}^n dp_j\wedge dq_j, \tag{4.1}
f(q,p)=(q,−p1,p′),g(q,p)=(q1+p1,q′,−p1,p′).(4.2) f(q,p)=(q,-p_1,p'),\qquad g(q,p)=(q_1+p_1,q',-p_1,p'). \tag{4.2}

Proof. Let N={p2=q2=⋯=pn=qn=0}N=\{p_2=q_2=\cdots=p_n=q_n=0\}. In (1.5), the normal function rr is odd under both involutions, with nonzero normal differential. Extend r∣Nr|_N constantly along all spectator fields to obtain p1p_1. The section is invariant, so p1p_1 is odd under both exchanges. It commutes with every spectator function, vanishes precisely on Γ\Gamma, and has nonzero normal differential there: the spectator flows are diffeomorphisms preserving Γ\Gamma.

Write p1=s ap_1=s\,a, where aa is smooth, ff-even and nonzero. Although Hp1H_{p_1} is singular, the field

Y=p1Hp1(4.3) Y=p_1H_{p_1} \tag{4.3}

is smooth by (2.2). In fact its folded-plane coefficients are a(a+sas)a(a+s a_s) in the x1x_1 direction and −saax1-s a a_{x_1} in the ss direction. Its spectator coefficients have a factor s2s^2. This proves smoothness of every coefficient and derivative. At Γ\Gamma it is a2∂x1a^2\partial_{x_1}. It is invariant under both exchanges and commutes with all spectator fields; these facts first follow off Γ\Gamma and extend smoothly.

On the two-dimensional section NN, solve

Yq1=1,q1=0on x1=0,(4.4) Yq_1=1,\qquad q_1=0\quad\hbox{on }x_1=0, \tag{4.4}

and extend constantly along the spectator flows. The initial curve is transverse to YY, and is invariant under ff. Thus q1q_1 is smooth and ff-even. Its differential along the characteristic direction is nonzero. Together with the spectator coordinates and p1p_1, it is a full coordinate system near cc.

Off Γ\Gamma the brackets are exactly those of (4.1): the spectator brackets are canonical, all cross brackets vanish, and

{p1,q1}=1/p1. \{p_1,q_1\}=1/p_1.

Inverting this bracket matrix gives (4.1) there; both sides are smooth, so it holds on Γ\Gamma as well. When n=1n=1, the spectator construction is empty and the same two-dimensional argument applies directly.

The second exchange fixes every spectator coordinate and sends p1p_1 to −p1-p_1. Write its remaining position as q1+G(q,p)q_1+G(q,p). Preservation of (4.1) gives

p1 dp1∧dG=0. p_1\,dp_1\wedge dG=0.

Off Γ\Gamma, this forces every derivative of GG except its p1p_1 derivative to vanish. Continuity gives the same conclusion at Γ\Gamma; on a product neighborhood G=G(p1)G=G(p_1) is smooth. The fixed set and g2=Ig^2=I imply G(0)=0G(0)=0 and G(−u)=−G(u)G(-u)=-G(u). The reflection lines are distinct exactly when G′(0)≠0G'(0)\ne0.

Temporarily write the remaining position as tt and its normal momentum as uu. Make a map from new variables to old variables,

κ(t,τ)=(tA(τ),B(τ)).(4.5) \kappa(t,\tau)=(tA(\tau),B(\tau)). \tag{4.5}

It commutes with ff if AA is even and BB odd. It preserves the folded form if

BB′A=τ.(4.6) B B'A=\tau. \tag{4.6}

To conjugate the old exchange to (t+τ,−τ)(t+\tau,-\tau), we need

G(B)=τA=τ2BB′.(4.7) G(B)=\tau A=\frac{\tau^2}{BB'}. \tag{4.7}

Multiplication and integration give the exact equation

τ33=∫0B(τ)vG(v) dv.(4.8) \frac{\tau^3}{3}=\int_0^{B(\tau)}vG(v)\,dv. \tag{4.8}

Define

G1(b)=3∫01r2 G(br)br dr,(4.9) G_1(b)=3\int_0^1 r^2\,\frac{G(br)}{br}\,dr, \tag{4.9}

where the divided odd function is extended smoothly at zero. Then G1G_1 is smooth and even, G1(0)=G′(0)≠0G_1(0)=G'(0)\ne0, and the integral in (4.8) equals b3G1(b)/3b^3G_1(b)/3. The signed real cube root of G1G_1 is smooth near zero because G1G_1 has a fixed nonzero sign. The map

b⟼b G1(b)1/3 b\longmapsto b\,G_1(b)^{1/3}

is smooth, odd, with nonzero derivative at zero. Its actual local inverse is B(τ)B(\tau), smooth and odd with a simple zero.

Write B=τb(τ)B=\tau b(\tau) with bb even and nonzero. Then

A=1b(b+τb′)(4.10) A=\frac{1}{b(b+\tau b')} \tag{4.10}

is smooth, even and nonzero at zero. Differentiating (4.8) proves (4.7); (4.6) holds by definition. The full Jacobian of (4.5) is nonzero at τ=0\tau=0, since A(0)B′(0)≠0A(0)B'(0)\ne0. Thus it is an actual diffeomorphism and a full folded-form equivalence. Its inverse as a coordinate change completes (4.1)–(4.2). Negative G′(0)G'(0) is included by the signed cube root. □\square

5. Build the homogeneous canonical coordinates

Assume now that MM is conic with a nonzero radial field RR, that σ\sigma has degree one, that f,gf,g commute with dilation, and that

Rc,Lf(c),Lg(c)are linearly independent.(5.1) R_c,\quad L_f(c),\quad L_g(c) \quad\hbox{are linearly independent}. \tag{5.1}

Since RR is tangent to Γ\Gamma, (1.4) implies Rc∉KcR_c\notin K_c. The homogeneous single-involution theorem therefore applies, with n≥2n\ge2. Choose the chart (2.1) with

deg⁡xj=0,deg⁡s=12,deg⁡ξj=1 (j≥2),(x,s,ξ′)(c)=(0,0,en).(5.2) \deg x_j=0,\quad \deg s=\tfrac12,\quad \deg\xi_j=1\ (j\ge2),\qquad (x,s,\xi')(c)=(0,0,e_n). \tag{5.2}

The last momentum is positive after shrinking the conic neighborhood.

The homogeneous simultaneous involution theorem supplies coordinates (t,z,r,ρ)(t,z,r,\rho), where t,z,rt,z,r have degree zero and ρ>0\rho>0 has degree one, with

f(t,z,r,ρ)=(t,z,−r,ρ),g(t,z,r,ρ)=(t+r,z,−r,ρ).(5.3) f(t,z,r,\rho)=(t,z,-r,\rho),\qquad g(t,z,r,\rho)=(t+r,z,-r,\rho). \tag{5.3}

Again the tt direction on Γ\Gamma is KK. The restrictions of the old spectators are functions of z,ρz,\rho, independent of tt. Extend them independently of t,rt,r. The functions Xj,PjX_j,P_j in (3.1) are now common invariants with their exact degrees zero and one. The function

η=rρ(5.4) \eta=r\sqrt{\rho} \tag{5.4}

has degree 1/21/2, is odd under both exchanges and has a simple normal zero.

Construct p2,q2,…,pn−1,qn−1,pnp_2,q_2,\ldots,p_{n-1},q_{n-1},p_n by Section 3. The intermediate zero sections are conic. For a homogeneous Hamiltonian of degree dd,

[R,Ha]=(d−1)Ha.(5.5) [R,H_a]=(d-1)H_a. \tag{5.5}

This follows by dilating its contraction equation. Each distribution used in the construction is preserved by dilation, and the equations and initial data have their assigned degrees. Uniqueness in (2.4) gives

deg⁡pj=1,deg⁡qj=0. \deg p_j=1,\qquad \deg q_j=0.

In particular pn(c)=1p_n(c)=1; it commutes with all earlier pairs and is positive locally. Its field is smooth and

Hpn∣Γ=∂xn+Cn∂x1.(5.6) H_{p_n}|_\Gamma=\partial_{x_n}+C_n\partial_{x_1}. \tag{5.6}

We have not chosen qnq_n. Keeping pnp_n nonzero avoids imposing a nonconic zero level for that momentum.

Let

N0={p2=q2=⋯=pn−1=qn−1=0}, N_0=\{p_2=q_2=\cdots=p_{n-1}=q_{n-1}=0\},

a four-dimensional section; for n=2n=2, it is the whole neighborhood. The earlier fields together with HpnH_{p_n} commute. Their common transverse section is

T=N0∩{Xn=0}.(5.7) T=N_0\cap\{X_n=0\}. \tag{5.7}

The earlier-field transversality follows from the canonical pairs; the last direction is transverse by HpnXn=1H_{p_n}X_n=1 at Γ\Gamma. Both TT and the seed η∣T\eta|_T respect the involutions and dilation. Extend that seed constantly along all these flows, and call the result hh. Then hh has degree 1/21/2, is odd under f,gf,g, has a simple normal zero, and commutes with every earlier pair and pnp_n.

As in (4.3), Y=hHhY=hH_h is smooth, nonzero along KK on Γ\Gamma, invariant under both exchanges, and has degree zero as a vector field. It commutes with the earlier fields and HpnH_{p_n}. On N0N_0, solve

Yq1=1,Hpnq1=0,q1=0on Xn=x1=0.(5.8) Yq_1=1,\qquad H_{p_n}q_1=0,\qquad q_1=0\quad\hbox{on }X_n=x_1=0. \tag{5.8}

At Γ\Gamma, YY is a nonzero multiple of ∂x1\partial_{x_1}, YXn=0YX_n=0, and HpnXn=1H_{p_n}X_n=1. Thus the initial two-dimensional section is transverse to both fields. Their commuting product flow proves existence and uniqueness. Extend along the earlier flows. The initial section is conic and ff-invariant, so q1q_1 has degree zero and is ff-even. The seed is not required to be gg-invariant.

The remaining position needs one more smooth field. Although Hq1H_{q_1} is singular, evenness of q1q_1 and (2.2) show that

Z=hHq1(5.9) Z=hH_{q_1} \tag{5.9}

is smooth. To check its full extension, write h=sah=s a with smooth even nonzero aa. Formula (2.2) gives the folded-plane part a(q1,s∂x1−q1,x1∂s)a(q_{1,s}\partial_{x_1}-q_{1,x_1}\partial_s); all its spectator coefficients have a factor ss. Evenness gives q1,s∣Γ=0q_{1,s}|_\Gamma=0, while Yq1=1Yq_1=1 gives q1,x1∣Γ=a−2q_{1,x_1}|_\Gamma=a^{-2}. Thus Z∣Γ=−a−1∂sZ|_\Gamma=-a^{-1}\partial_s. On Γ\Gamma it is transverse to the fold. Off Γ\Gamma,

{h,q1}=1/h,Zh=−1,Zq1=0,Yh=0,Yq1=1.(5.10) \{h,q_1\}=1/h,\qquad Z h=-1,\quad Zq_1=0,\quad Yh=0,\quad Yq_1=1. \tag{5.10}

The singular bracket terms cancel:

[Y,Z]=h2H1/h−h Hq1(h)Hh=−Hh+Hh=0.(5.11) [Y,Z] =h^2H_{1/h}-h\,H_{q_1}(h)H_h =-H_h+H_h=0. \tag{5.11}

The other commutators with HpnH_{p_n} and the earlier fields are zero because all corresponding brackets vanish and those fields kill hh. These identities extend smoothly. We also have

[R,Z]=−12Z,f∗Z=−Z.(5.12) [R,Z]=-\tfrac12 Z,\qquad f_*Z=-Z. \tag{5.12}

The fields Y,Z,HpnY,Z,H_{p_n} are independent on N0N_0: they change q1,h,Xnq_1,h,X_n, respectively, with an invertible derivative matrix at the marked point. For ZXn=0Z X_n=0 there, use ff-evenness of XnX_n and the fact that the limiting field ZZ is in the normal ff-reflection line. Thus

B=N0∩{h=q1=Xn=0} B=N_0\cap\{h=q_1=X_n=0\}

is a transverse one-dimensional conic section, with pnp_n as a coordinate. Indeed RR is tangent to this section: all its defining functions have homogeneous zero values there. Since Rpn=pnRp_n=p_n and pn(c)=1p_n(c)=1, dpndp_n is nonzero on its tangent line. This proves the remaining coordinate assertion without introducing a zero level for the positive last momentum. Define qnq_n by the compatible equations

Yqn=0,Zqn=0,Hpnqn=1,qn∣B=0,(5.13) Yq_n=0,\qquad Zq_n=0,\qquad H_{p_n}q_n=1, \qquad q_n|_B=0, \tag{5.13}

then extend constantly along the earlier fields. The commuting-flow map (2.4) proves smooth existence and uniqueness. Dilation preserves the distribution and equations: Y,HpnY,H_{p_n} have weight zero and the equation for the weight −1/2-1/2 field ZZ has zero right-hand side. It follows that qnq_n has degree zero. The change Z↦−ZZ\mapsto-Z under ff likewise leaves (5.13) unchanged, so qnq_n is ff-even.

All brackets are now those of

σ=h dh∧dq1+∑j=2ndpj∧dqj.(5.14) \sigma=h\,dh\wedge dq_1+\sum_{j=2}^n dp_j\wedge dq_j. \tag{5.14}

For the last cross bracket, Zqn=0Zq_n=0 says {q1,qn}=0\{q_1,q_n\}=0 off the fold, while Yqn=0Yq_n=0 says {h,qn}=0\{h,q_n\}=0; the other equations give the remaining brackets. The full coordinate differential is nonzero at Γ\Gamma: the earlier pairs give their independent transverse directions, and on N0N_0 the coordinates h,q1,pn,qnh,q_1,p_n,q_n are independent by the three flow directions and dpn∣B≠0dp_n|_B\ne0. Inverting the bracket matrix off the fold proves (5.14), then smoothness proves it everywhere.

Finally set

xj=qj,ξj=pj (j≥2),ξ1=hpn/2.(5.15) x_j=q_j,\quad \xi_j=p_j\ (j\ge2),\qquad \xi_1=h\sqrt{p_n/2}. \tag{5.15}

This is a full smooth coordinate change near pn=1p_n=1. Every xjx_j has degree zero and every ξj\xi_j degree one, with marked values (0,en)(0,e_n). Since h2/2=ξ12/ξnh^2/2=\xi_1^2/\xi_n, the form becomes

σ=d(ξ12/ξn)∧dx1+∑j=2ndξj∧dxj.(5.16) \sigma=d(\xi_1^2/\xi_n)\wedge dx_1 +\sum_{j=2}^n d\xi_j\wedge dx_j. \tag{5.16}

The first exchange fixes all coordinates except the sign of ξ1\xi_1; the second fixes all momenta except that sign and all positions except possibly x1,xnx_1,x_n.

The conic domain can be made explicit. First perform the finitely many commuting-flow constructions on a common neighborhood of cc, using their nonzero transverse differentials and shrinking the flow boxes. The degree identities just proved hold wherever a point and its nearby dilation remain in that neighborhood. On the transverse section pn=1p_n=1, the full coordinate map is a local diffeomorphism after omitting the fixed coordinate pnp_n. Shrink this section once, then use the unique positive-ray product proved in the simultaneous-reflection lesson, §6. Extend each coordinate along those rays with its stated degree. The coordinate pn>0p_n>0 recovers the dilation factor, and the normalized coordinates recover the point of the section, so the extended map is a diffeomorphism onto its conic image. It agrees with the local construction by the degree identities. The form and involution identities extend to the whole positive saturation by homogeneity. This justifies a full conic neighborhood without requiring uniform unscaled flow times along an unbounded ray.

6. The homogeneous shift and its full normalization

Write the second exchange in the coordinates of Section 5 as

g(x,ξ)=(x1+v(x,ξ),x2,…,xn−1,xn+w(x,ξ),−ξ1,ξ2,…,ξn). g(x,\xi)=(x_1+v(x,\xi),x_2,\ldots,x_{n-1},x_n+w(x,\xi), -\xi_1,\xi_2,\ldots,\xi_n).

Preservation of (5.16) gives

d(ξ12/ξn)∧dv+dξn∧dw=0.(6.1) d(\xi_1^2/\xi_n)\wedge dv+d\xi_n\wedge dw=0. \tag{6.1}

For ξ1≠0\xi_1\ne0, the two displayed momentum differentials are independent. Comparing the coefficients involving each position differential and each spectator momentum differential forces v,wv,w to be independent of all positions and of ξ2,…,ξn−1\xi_2,\ldots,\xi_{n-1}. Continuity extends that conclusion to the fold. Homogeneity of degree zero then gives smooth functions of the ratio

t=ξ1/ξn,v=V(t),w=W(t). t=\xi_1/\xi_n,\qquad v=V(t),\quad w=W(t).

Involution and the fixed set give V,WV,W odd and zero at zero. Equation (6.1) becomes

t2V′(t)+W′(t)=0.(6.2) t^2 V'(t)+W'(t)=0. \tag{6.2}

Distinctness of reflection lines is V′(0)≠0V'(0)\ne0. If V=2tV=2t, (6.2) and W(0)=0W(0)=0 force W=−2t3/3W=-2t^3/3.

Theorem 6.1 (homogeneous simultaneous folded form). Under (1.1)–(1.3) and (5.1), there are homogeneous coordinates of degrees zero and one, marked (0,en)(0,e_n), in which (5.16) holds, ξn>0\xi_n>0, and

f(x,ξ)=(x,−ξ1,ξ′),g(x,ξ)=(x1+2t,x2,…,xn−1,xn−23t3,−ξ1,ξ′),t=ξ1/ξn.(6.3) f(x,\xi)=(x,-\xi_1,\xi'),\qquad g(x,\xi)= (x_1+2t,x_2,\ldots,x_{n-1},x_n-\tfrac23t^3, -\xi_1,\xi'),\quad t=\xi_1/\xi_n. \tag{6.3}

Proof. Keep the spectators and ξn\xi_n fixed. Define a map from new to old variables by

x1old=x1S(t),xnold=xn+x1T(t),ξ1old=ξnR(t),ξnold=ξn.(6.4) \begin{split} x_1^{\mathrm{old}}&=x_1 S(t),\\ x_n^{\mathrm{old}}&=x_n+x_1 T(t),\\ \xi_1^{\mathrm{old}}&=\xi_n R(t),\qquad \xi_n^{\mathrm{old}}=\xi_n . \end{split} \tag{6.4}

Here RR will be odd with a simple zero; S,TS,T will be even. It is homogeneous and commutes with ff. Put U=R2U=R^2. Preservation of the form is equivalent to

US+T=t2,US′+T′=0,U′S=2t.(6.5) US+T=t^2,\qquad US'+T'=0,\qquad U'S=2t. \tag{6.5}

One can see sufficiency even at the level of primitives:

λ=(ξ12/ξn)dx1+∑j=2nξjdxj. \lambda=(\xi_1^2/\xi_n)dx_1+\sum_{j=2}^n\xi_jdx_j.

Substitution of (6.4) changes its first and last terms to

ξn{(US+T)dx1+dxn+x1(US′+T′)dt}. \xi_n\{(US+T)dx_1+dx_n+x_1(US'+T')dt\}.

Thus the first two identities in (6.5) give the exact pullback of λ\lambda, and hence of σ=dλ\sigma=d\lambda. The third identity and the first imply the second by differentiation. Necessity follows from the same full expression: put C=US+T−t2C=US+T-t^2 and D=US′+T′D=US'+T'. The difference of the pulled-back and original primitives is ξn(C dx1+x1D dt)\xi_n(C\,dx_1+x_1D\,dt). Its exterior derivative has coefficient CC in dξn∧dx1d\xi_n\wedge dx_1, and coefficient ξn(C′−D)\xi_n(C'-D) in dt∧dx1dt\wedge dx_1. These independent coefficients must vanish when the two-forms agree. Hence C=D=0C=D=0; differentiating C=0C=0 then gives U′S=2tU'S=2t. This proves the asserted equivalence as well as sufficiency.

Choose

S=2tU′,T=t2−2tUU′.(6.6) S=\frac{2t}{U'},\qquad T=t^2-\frac{2tU}{U'}. \tag{6.6}

If R=t r(t)R=t\,r(t), with rr smooth, even and nonzero, these formulas are smooth and even across zero: U′=2t r(r+tr′)U'=2t\,r(r+tr'). They give S(0)=1/r(0)2≠0S(0)=1/r(0)^2\ne0 and T(0)=0T(0)=0. The full map (6.4) is a diffeomorphism near the marked point; its Jacobian in x1,xn,ξ1,ξnx_1,x_n,\xi_1,\xi_n is nonzero there because S(0)R′(0)≠0S(0)R'(0)\ne0.

To obtain first shift 2t2t, the conjugacy equation is

V(R)=2tS=2t2RR′. V(R)=2tS=\frac{2t^2}{RR'}.

It integrates exactly to

2t33=∫0R(t)vV(v) dv.(6.7) \frac{2t^3}{3}=\int_0^{R(t)}vV(v)\,dv. \tag{6.7}

Define V1(b)=3∫01r2V(br)/(br) drV_1(b)=3\int_0^1 r^2 V(br)/(br)\,dr. As in (4.9), it is smooth and even, with V1(0)=V′(0)≠0V_1(0)=V'(0)\ne0. Thus (6.7) is

t=R [V1(R)/2]1/3. t=R\,[V_1(R)/2]^{1/3}.

Its signed-root map has a smooth odd inverse R(t)R(t) with a simple zero. Formulas (6.6) now define a full smooth homogeneous equivalence preserving the primitive and commuting with ff.

The conjugated second exchange has first shift 2t2t, is still an involution, and still preserves the folded form. Its last shift is odd, zero at zero, and satisfies (6.2). It is consequently −2t3/3-2t^3/3. Equivalently direct conjugacy gives W(R)=Wnew+2tTW(R)=W_{\mathrm{new}}+2tT. This proves (6.3) on the full conic neighborhood. □\square

The last position correction in (6.3) is necessary. Keeping only the first shift 2t2t would change the form by 2t2 dξn∧dt2t^2\,d\xi_n\wedge dt. The term dξn∧d(−2t3/3)d\xi_n\wedge d(-2t^3/3) cancels it exactly.

7. Exact models, radial independence and the diagram

For the model (5.16), write ρ=ξn>0\rho=\xi_n>0, t=ξ1/ρt=\xi_1/\rho. On the fold, the ambient radical is

E=R∂x1+R∂ξ1,K=R∂x1. E=\mathbb R\partial_{x_1}+\mathbb R\partial_{\xi_1}, \qquad K=\mathbb R\partial_{x_1}.

The model has a simple top-power zero, since

σn=n!2ξ1ρ dξ1∧dx1∧⋀j=2n(dξj∧dxj).(7.1) \sigma^n=n!\frac{2\xi_1}{\rho}\, d\xi_1\wedge dx_1\wedge\bigwedge_{j=2}^n(d\xi_j\wedge dx_j). \tag{7.1}

The restricted form has rank 2n−22n-2. At (0,en)(0,e_n), the reflection lines of (6.3) are

Lf=R∂ξ1,Lg=R(∂ξ1−∂x1),Rc=∂ξn.(7.2) L_f=\mathbb R\partial_{\xi_1},\qquad L_g=\mathbb R(\partial_{\xi_1}-\partial_{x_1}), \qquad R_c=\partial_{\xi_n}. \tag{7.2}

They are linearly independent. The canonical one-form is precisely the primitive in Section 6. Its restricted value at the marked point is dxndx_n, so is nonzero.

Both model maps are involutions. Their product in the order f∘gf\circ g preserves t,ρt,\rho and sends

x1⟼x1+2t,xn⟼xn−23t3.(7.3) x_1\longmapsto x_1+2t,\qquad x_n\longmapsto x_n-\tfrac23t^3. \tag{7.3}

This product is a translation in the two position coordinates when the frequencies are fixed; it is a symplectic map for the folded form. The shifts use linear and cubic powers of the ratio; both shifts and both positions have degree zero under dilation.

The exact homogeneous shifts and their cancellation in the two-form

Figure 7.1. Spectators and the positive frequency ρ\rho are fixed. The upper panel plots the exact shifts V(t)=2tV(t)=2t and W(t)=−2t3/3W(t)=-2t^3/3 in the product (7.3). The lower panel plots the two contributions t2V′=2t2t^2V'=2t^2 and W′=−2t2W'=-2t^2 to (6.2); their sum is identically zero. These are coordinate shifts and differential-form coefficients, rather than source trajectories. At t=1/2t=1/2, the shifts are 11 and −1/12-1/12; the cancellation terms are 1/21/2 and −1/2-1/2. The formula locators are (6.1)–(6.3) and (7.3); Exercise 8.8 derives a nonlinear example before normalization.

The nonzero radial condition has a substantive role. On an ordinary folded plane times a symplectic spectator plane, take q>0q>0,

σ=u du∧dq+dρ∧dz,Dκ(q,z,u,ρ)=(κq,z,u,κρ).(7.4) \sigma=u\,du\wedge dq+d\rho\wedge dz,\qquad D_\kappa(q,z,u,\rho)=(\kappa q,z,u,\kappa\rho). \tag{7.4}

This form has degree one and satisfies the fold rank conditions. Put

f(q,z,u,ρ)=(q,z,−u,ρ),g(q,z,u,ρ)=(q(1+u)3,z,−u/(1+u),ρ).(7.5) f(q,z,u,\rho)=(q,z,-u,\rho),\qquad g(q,z,u,\rho)=(q(1+u)^3,z,-u/(1+u),\rho). \tag{7.5}

Both maps are homogeneous involutions preserving the form. Their fixed set is u=0u=0, and their reflection lines are distinct for q>0q>0. At (q,z,u,ρ)=(1,0,0,0)(q,z,u,\rho)=(1,0,0,0), however, the nonzero radial vector is ∂q\partial_q, in their span. The restricted one-form ιRσ∣TΓ=ρ dz\iota_R\sigma|_{T\Gamma}=\rho\,dz is zero there. The model of Theorem 6.1 has a nonzero restricted one-form, so no homogeneous equivalence can take this example to that marked model. The ordinary theorem still applies.

8. Exercises with complete solutions

Exercise 8.1 (introductory: the intrinsic direction). Under (1.1)–(1.3), prove (1.4) without using Poisson brackets. Show that a common invariant function is constant along the characteristic leaves on Γ\Gamma.

Solution. Invariance under dfdf, which is identity on TΓT\Gamma and minus identity on LfL_f, makes σ(v,w)=0\sigma(v,w)=0 for v∈Lf,w∈TΓv\in L_f,w\in T\Gamma. Alternation gives σ(v,v)=0\sigma(v,v)=0; these vectors span TMTM, so Lf⊂EL_f\subset E. The same holds for gg. Distinct lines span the two-dimensional EE, whose intersection with TΓT\Gamma is the one-dimensional KK. If a∘f=aa\circ f=a, then da(v)=da(df v)=−da(v)da(v)=da(df\,v)=-da(v) for v∈Lfv\in L_f, so da(v)=0da(v)=0, and similarly for LgL_g. Thus da∣K=0da|_K=0. Integrating that differential equation along any local nonzero field spanning KK proves constancy on its local leaves.

Exercise 8.2 (intermediate: the regularity condition). In the four-dimensional ordinary chart with form s ds∧dx+dρ∧dzs\,ds\wedge dx+d\rho\wedge dz, set a=s2(1+x2)+ρa=s^2(1+x^2)+\rho. Compute HaH_a and its restriction to the fold. Compare it with HbH_b for b=xb=x.

Solution. The derivatives are as=2s(1+x2)a_s=2s(1+x^2), ax=2xs2a_x=2xs^2, aρ=1a_\rho=1, az=0a_z=0. Formula (2.2) gives

Ha=2(1+x2)∂x−2xs∂s+∂z. H_a=2(1+x^2)\partial_x-2xs\partial_s+\partial_z.

It is smooth, tangent to s=0s=0, and there equals 2(1+x2)∂x+∂z2(1+x^2)\partial_x+\partial_z. The spectator translation remains nonzero while a characteristic component is allowed. For b=xb=x, Hb=−s−1∂sH_b=-s^{-1}\partial_s; it has no smooth extension. Evenness alone does not suffice: bx∣Γ=1b_x|_\Gamma=1. Multiplication by ss gives the smooth transverse field sHb=−∂ssH_b=-\partial_s, the mechanism later used for ZZ.

Exercise 8.3 (intermediate: canonical pairs by full flows). Let smooth fields Hp,HqH_p,H_q satisfy {p,q}=1\{p,q\}=1 in a folded neighborhood. With σ(Hp,Hq)=1\sigma(H_p,H_q)=1, verify the pullback identity in (2.6), including the sign, and explain why it is valid on the fold.

Solution. On the section p=q=0p=q=0, the two fields are transverse and the section tangent is their σ\sigma-orthogonal. Along the map, the derivatives in q,pq,p are Hp,−HqH_p,-H_q, because the fields commute. Their pairing is −1-1, equal to (dp∧dq)(∂q,∂p)(dp\wedge dq)(\partial_q,\partial_p). A transported section tangent pairs to zero with both fields because both flows preserve σ\sigma. The remaining section pairings give the restricted form. Hence the pullback is dp∧dq+σ∣Ndp\wedge dq+\sigma|_N. The derivative is invertible by transversality. All fields, flows and forms used are smooth on the full neighborhood; preservation follows from dιHpσ=−d2p=0d\iota_{H_p}\sigma=-d^2p=0, including on the fold by the extended contraction equation. No nondegenerate ambient inverse on the fold was used.

Exercise 8.4 (advanced: a negative cubic derivative). In the ordinary final normalization, take G(b)=−8bG(b)=-8b. Find B,AB,A exactly and verify both preservation of u du∧dtu\,du\wedge dt and conjugacy to the shift τ\tau.

Solution. The integral in (4.8) is −8B3/3-8B^3/3, so B=−τ/2B=-\tau/2. Then BB′=τ/4BB'=\tau/4, giving A=4A=4. The map is κ(t,τ)=(4t,−τ/2)\kappa(t,\tau)=(4t,-\tau/2), with nonzero Jacobian −2-2, and

B dB∧d(4t)=(−τ/2)(−dτ/2)∧4dt=τ dτ∧dt. B\,dB\wedge d(4t)=(-\tau/2)(-d\tau/2)\wedge4dt =\tau\,d\tau\wedge dt.

The old shift is G(B)=4τ=τAG(B)=4\tau=\tau A, so gκ=κg0g\kappa=\kappa g_0, with g0(t,τ)=(t+τ,−τ)g_0(t,\tau)=(t+\tau,-\tau). The real signed cube root covers this case smoothly; a choice of positive cube root would not solve the equation.

Exercise 8.5 (advanced: nonlinear ordinary normalization). Let G(b)=b+3b3G(b)=b+3b^3. Give the exact implicit normalizing coordinate and the first nontrivial terms of B(τ)B(\tau) and A(τ)A(\tau).

Solution. Equation (4.8) reads

τ3=B3+95B5,τ=B(1+95B2)1/3. \tau^3=B^3+\tfrac95 B^5,\qquad \tau=B(1+\tfrac95B^2)^{1/3}.

The latter map is smooth, odd, with derivative one at zero, so its local inverse is the full BB. Expansion gives τ=B+35B3+O(B5)\tau=B+\tfrac35B^3+O(B^5), hence

B=τ−35τ3+O(τ5),BB′=τ−125τ3+O(τ5),A=1+125τ2+O(τ4). B=\tau-\tfrac35\tau^3+O(\tau^5),\qquad BB'=\tau-\tfrac{12}{5}\tau^3+O(\tau^5),\qquad A=1+\tfrac{12}{5}\tau^2+O(\tau^4).

Differentiating the exact implicit equation, rather than just the truncated expansion, proves BB′A=τBB'A=\tau and G(B)=τAG(B)=\tau A for the full maps.

Exercise 8.6 (advanced: the transverse homogeneous field). Starting with {h,q1}=1/h\{h,q_1\}=1/h, derive (5.10)–(5.12). Why does the equation Zqn=0Zq_n=0 not obstruct degree zero for qnq_n?

Solution. Set Y=hHh,Z=hHq1Y=hH_h,Z=hH_{q_1}. Then Hhh=Hq1q1=0H_hh=H_{q_1}q_1=0 and Hq1h=−1/hH_{q_1}h=-1/h, giving Yh=0,Yq1=1,Zh=−1,Zq1=0Yh=0,Yq_1=1,Zh=-1,Zq_1=0. The identity [Hh,Hq1]=H1/h=−h−2Hh[H_h,H_{q_1}]=H_{1/h}=-h^{-2}H_h makes

[Y,Z]=h2(−h−2Hh)−h(−1/h)Hh=0. [Y,Z]=h^2(-h^{-2}H_h)-h(-1/h)H_h=0.

The fields are smooth by the division arguments in Sections 4–5, so the identities extend to Γ\Gamma. Since deg⁡h=1/2,deg⁡q1=0\deg h=1/2,\deg q_1=0, (5.5) gives [R,Y]=0[R,Y]=0 and [R,Z]=−Z/2[R,Z]=-Z/2. Under ff, hh is odd and q1q_1 even, so YY is invariant and ZZ changes sign. Dilation multiplies ZZ by its weight factor but leaves its zero equation unchanged. The other two fields in (5.13) have weight zero, the nonzero right-hand side is one, and the section and zero data are conic. Uniqueness therefore gives degree zero for qnq_n.

Exercise 8.7 (intermediate: the required last shift). In dimension four, with ρ>0\rho>0, consider the model primitive λ=ρt2dx+ρdz\lambda=\rho t^2dx+\rho dz. Compute the pullback by g0(x,z,t,ρ)=(x+2t,z,−t,ρ)g_0(x,z,t,\rho)=(x+2t,z,-t,\rho). Find the position correction that makes it preserve λ\lambda and verify that the corrected map is an involution.

Solution. The pullback is

g0∗λ=ρt2d(x+2t)+ρdz=λ+2ρt2dt. g_0^*\lambda=\rho t^2d(x+2t)+\rho dz =\lambda+2\rho t^2dt.

Its exterior derivative differs by 2t2dρ∧dt2t^2d\rho\wedge dt, so g0g_0 does not preserve the two-form. A shift z↦z+W(t)z\mapsto z+W(t) adds ρW′(t)dt\rho W'(t)dt. Thus W′=−2t2W'=-2t^2, and the fixed-set condition gives W=−2t3/3W=-2t^3/3. Both shifts are odd. Applying the corrected map twice cancels them and restores tt, so it is an involution. It preserves the primitive exactly, not only its derivative.

Exercise 8.8 (advanced: full nonlinear homogeneous normalization). Take V(b)=2b+4b3V(b)=2b+4b^3. Find W(b)W(b), the exact implicit R(t)R(t), and the first nontrivial terms of R,S,TR,S,T. Check the full conjugacy equations by differentiating identities.

Solution. Equation (6.2) gives W′=−2b2−12b4W'=-2b^2-12b^4, so

W(b)=−23b3−125b5. W(b)=-\tfrac23b^3-\tfrac{12}{5}b^5.

Equation (6.7) gives

t3=R3+65R5,t=R(1+65R2)1/3. t^3=R^3+\tfrac65 R^5,\qquad t=R(1+\tfrac65R^2)^{1/3}.

This is a smooth odd invertible map near zero. Its inverse has

R=t−25t3+O(t5),U=R2=t2−45t4+O(t6). R=t-\tfrac25t^3+O(t^5),\qquad U=R^2=t^2-\tfrac45t^4+O(t^6).

Consequently

S=1+85t2+O(t4),T=−45t4+O(t6). S=1+\tfrac85t^2+O(t^4),\qquad T=-\tfrac45t^4+O(t^6).

For the full functions, (6.6) gives U′S=2tU'S=2t, US+T=t2US+T=t^2, and hence US′+T′=0US'+T'=0; these are exact primitive-preservation identities. Differentiating the implicit integral gives V(R)RR′=2t2V(R)RR'=2t^2, equivalently V(R)=2tSV(R)=2tS. Finally let F=W(R)−2tTF=W(R)-2tT. Using W′(R)=−R2V′(R)W'(R)=-R^2V'(R), the derivative of V(R)=2tSV(R)=2tS, and the two identities for TT, one obtains

F′=−U(2S+2tS′)−2T−2tT′=−2(US+T)=−2t2. F'=-U(2S+2tS')-2T-2tT' =-2(US+T)=-2t^2.

As F(0)=0F(0)=0, F=−2t3/3F=-2t^3/3, proving the last full conjugacy equation. The expansions alone would not prove these full identities.

Exercise 8.9 (advanced: a true excluded radial example). Verify every assertion about (7.4)–(7.5), including the independent reflection lines and the obstruction to the homogeneous marked model.

Solution. The top power is 2u du∧dq∧dρ∧dz2u\,du\wedge dq\wedge d\rho\wedge dz, with simple zero u=0u=0, and the restricted form is dρ∧dzd\rho\wedge dz, of rank two. Dilation multiplies each term by κ\kappa, and its radial field is q∂q+ρ∂ρq\partial_q+\rho\partial_\rho, nonzero for q>0q>0. Both maps commute with dilation. Set B(u)=−u/(1+u)B(u)=-u/(1+u) and A(u)=(1+u)3A(u)=(1+u)^3. On a small neighborhood of zero, B(B(u))=uB(B(u))=u, A(B(u))A(u)=1A(B(u))A(u)=1, and the only fixed normal value is zero. Thus g2=Ig^2=I and its fixed set is Γ\Gamma. Since BB′A=uBB'A=u, its full folded-plane pullback is

B dB∧d(qA)=BB′A du∧dq=u du∧dq. B\,dB\wedge d(qA)=BB'A\,du\wedge dq =u\,du\wedge dq.

The spectator form is unchanged, so both involutions preserve σ\sigma. At u=0u=0, the reflection lines are R∂u\mathbb R\partial_u and R(∂u−32q∂q)\mathbb R(\partial_u-\tfrac32q\partial_q); they are distinct for q>0q>0. Their span contains the radial vector at the marked point (1,0,0,0)(1,0,0,0). The primitive is ιRσ=−qu du+ρ dz\iota_R\sigma=-qu\,du+\rho\,dz; its restricted value at that point is zero. A homogeneous form equivalence intertwines radial fields and preserves this restricted-one-form condition. The marked model has nonzero restricted value dxndx_n, so such an equivalence is impossible. All other hypotheses, including distinct reflection lines, hold for this excluded example.

Exercise 8.10 (intermediate: exact top power and product order). Derive (7.1)–(7.3), including all constants, and compute the shifts and cancellation coefficients at t=1/2t=1/2.

Solution. We have

d(ξ12/ρ)=2ξ1ρdξ1−ξ12ρ2dρ. d(\xi_1^2/\rho)=\frac{2\xi_1}{\rho}d\xi_1 -\frac{\xi_1^2}{\rho^2}d\rho.

In the top exterior power, the second term is killed by the spectator dρ∧dxnd\rho\wedge dx_n. Choosing the first folded factor and each of the n−1n-1 spectator factors gives n!n!, proving (7.1). At the marked point the differential of gg sends a variation in ξ1\xi_1 to 2∂x1−∂ξ12\partial_{x_1}-\partial_{\xi_1}; its minus-one line is therefore ∂ξ1−∂x1\partial_{\xi_1}-\partial_{x_1}. The cubic shift has zero differential there. The first reflection line and radial vector are as in (7.2). In f∘gf\circ g, gg makes the two shifts and flips tt; ff flips it back without another shift, giving (7.3). At t=1/2t=1/2, V=1,W=−1/12V=1,W=-1/12, and t2V′=1/2,W′=−1/2t^2V'=1/2,W'=-1/2. In the opposite product order g∘fg\circ f, both position shifts have the opposite signs.

The simultaneous folded-form proofs are local on the stated ordinary or conic neighborhoods. The complete canonical relation fold theorem, Airy representation and continuity results remain subsequent work.

References and component notices

Original lesson, exercises and coordinate artwork: GPT-6.1 Sol (OpenAI), Ultra, September 2026, CC0. Restoration, supporting details and exact programme prerequisite review: GPT-6 Astra (OpenAI), Ultra, 5 October 2026. The cited book and linked prerequisite components retain their own rights; no book text or file is included in this reader.