Prescribed canonical coordinates and isotropic fibers

Darboux coordinates can retain functions already known to be canonical. On a conic symplectic manifold they can also retain their degrees and specified values. This flexibility lets us straighten conic isotropic submanifolds even when their canonical one-form vanishes. We prove the coordinate extension as an actual function construction, then identify the resulting fiber model and the conormal geometry of regular base projections.

We use the symplectic and Hamiltonian conventions of Phase space and generating families. The characteristic geometry and the zero-valued momentum construction in Homogeneous submanifold normal forms provide context; its nonzero restricted one-form theorem does not supply the zero-one-form result proved here. The exact flow proofs are F0–F1, tangent flows, commuting fields and dilation conjugation, with the complete finite-coordinate flow construction. The stationary-phase foundation proves the inverse and implicit function theorems P2–P3 and the parameter fundamental theorem of calculus; the proof map binds their exact locators and earlier inputs. Exterior differentiation, pullback and Cartan's formula are proved in the differential-form foundation. The map constant-rank argument needed here is proved explicitly in Theorem 5.1 below.

The primary source is the reprint of Hörmander III, second edition (1994), Theorems 21.1.6, 21.1.9, 21.2.8 and 21.2.9 (printed 273–274, 279–281 and 287–288). We spell out the partial symplectic linear algebra, all flow equations, the exceptional final homogeneous coordinate, both isotropic induction steps, and an explicit homogeneous map from the intermediate model to a fiber. No pseudodifferential calculus is needed in this lesson.

1. Extend an isometry of alternating subspaces

Our convention on a symplectic manifold is

ιHfω=−df,{f,g}=Hfg,ω=∑j=1ndpj∧dqj.(1.1) \iota_{H_f}\omega=-df,\qquad \{f,g\}=H_fg,\qquad \omega=\sum_{j=1}^n dp_j\wedge dq_j. \tag{1.1}

Thus Hpj=∂qjH_{p_j}=\partial_{q_j}, Hqj=−∂pjH_{q_j}=-\partial_{p_j}, and {pk,qj}=δkj\{p_k,q_j\}=\delta_{kj}. The Jacobi identity follows from dω=0d\omega=0, and gives [Hf,Hg]=H{f,g}[H_f,H_g]=H_{\{f,g\}}.

We need a linear extension that works when a prescribed subspace is degenerate for the restricted form.

Lemma 1.1 (alternating-subspace extension). Let W,W′W,W' be symplectic vector spaces of the same dimension. If U⊂WU\subset W, U′⊂W′U'\subset W' and L:U→U′L:U\to U' is a linear isomorphism preserving the restricted alternating forms, then LL extends to a symplectic isomorphism W→W′W\to W'.

Proof. Let K=U∩UωK=U\cap U^\omega, the radical of the restricted form. Its image is the corresponding radical K′K'. Choose a complement SS to KK in UU. The restriction to SS is nondegenerate: a vector of SS orthogonal to SS is also orthogonal to KK, hence to UU, and lies in K∩S=0K\cap S=0. Its image S′=L(S)S'=L(S) is similarly symplectic.

Work in the symplectic space SωS^\omega, which contains KK. For a basis k1,…,kdk_1,\ldots,k_d of KK, choose lj∈Sωl_j\in S^\omega such that

ω(ki,lj)=δij.(1.2) \omega(k_i,l_j)=\delta_{ij}. \tag{1.2}

These choices exist because the functionals ω(ki,⋅)\omega(k_i,\cdot) are independent on SωS^\omega: nondegeneracy identifies its vectors with covectors, and the kik_i are independent. Put Aij=ω(li,lj)A_{ij}=\omega(l_i,l_j) and

l~i=li−12∑jAijkj.(1.3) \widetilde l_i=l_i-\frac12\sum_j A_{ij}k_j. \tag{1.3}

Then ω(ki,l~j)=δij\omega(k_i,\widetilde l_j)=\delta_{ij}, while ω(l~i,l~j)=Aij−Aij/2+Aji/2=0\omega(\widetilde l_i,\widetilde l_j)=A_{ij}-A_{ij}/2+A_{ji}/2=0. Thus K+span⁡{l~i}K+\operatorname{span}\{\widetilde l_i\} is a nondegenerate symplectic block orthogonal to SS.

Perform the same construction in W′W', starting with L(ki)L(k_i) and S′S'. Extend LL by mapping each l~i\widetilde l_i to its constructed counterpart. This is an isomorphism of the resulting nondegenerate blocks. Their symplectic orthogonal complements have the same dimension; choose symplectic bases there and map one basis to the other. The direct sum of these maps is the required extension. Empty radicals and empty complements are allowed. □\square

In particular, independent vectors with the pairings of any indexed subset of a standard symplectic basis can be completed to a full indexed basis. Apply the lemma to their span and the span of the corresponding standard vectors. It also lets us include a prescribed radial vector, with its given pairings, in the extension.

2. Complete ordinary canonical functions without changing them

Theorem 2.1 (ordinary prescribed coordinates). Let SS be a smooth symplectic manifold of dimension 2n2n, c∈Sc\in S, and A,B⊂{1,…,n}A,B\subset\{1,\ldots,n\}. Suppose smooth functions qjq_j, j∈Aj\in A, and pkp_k, k∈Bk\in B, have independent differentials at cc and satisfy, on a neighborhood,

{qi,qj}=0,{pk,pl}=0,{pk,qj}=δkj(2.1) \{q_i,q_j\}=0,\qquad \{p_k,p_l\}=0,\qquad \{p_k,q_j\}=\delta_{kj} \tag{2.1}

for every indicated prescribed index. Then they extend to a full canonical coordinate system (q,p)(q,p) near cc, with the prescribed functions retained identically.

One may also require the coordinate differential at cc to be any compatible symplectic isomorphism extending their given differentials, and assign arbitrary marked values to the newly constructed functions.

Proof. The prescribed Hamilton vectors −Hqj(c)-H_{q_j}(c), Hpk(c)H_{p_k}(c) are independent and have the standard indexed pairings ω(−Hqj,Hpk)=δjk\omega(-H_{q_j},H_{p_k})=\delta_{jk}. Lemma 1.1 completes them. Equivalently, choose a full canonical cotangent basis at cc extending the given differentials. Fix that basis as the desired differential of our eventual chart.

At any intermediate stage the Hamilton fields of the current prescribed functions commute. Indeed their Poisson brackets in (2.1) are constants, and the Hamilton field of a constant is zero. Their values remain independent on a sufficiently small neighborhood, because their desired values at cc are a subset of the chosen full basis.

Here is the required smooth equation solver. For independent commuting fields V1,…,VdV_1,\ldots,V_d, take a small transverse section Σ\Sigma through cc. Their flows define

Φ(t,w)=exp⁡(t1V1)⋯exp⁡(tdVd)(w),w∈Σ.(2.2) \Phi(t,w)=\exp(t_1V_1)\cdots\exp(t_dV_d)(w), \qquad w\in\Sigma. \tag{2.2}

Its differential at (0,c)(0,c) is invertible, so it is a coordinate chart after shrinking. Commutativity gives Vi=∂tiV_i=\partial_{t_i} there. Thus equations Viu=aiV_i u=a_i, with constant aia_i, have the exact solution

u(Φ(t,w))=g(w)+∑iaiti.(2.3) u(\Phi(t,w))=g(w)+\sum_i a_it_i. \tag{2.3}

Choose g(c)g(c) as desired and dg(c)dg(c) as the desired differential restricted to TcΣT_c\Sigma. Explicitly, choose section coordinates ww with w(c)=0w(c)=0, express that covector as ∑αdα dwα\sum_\alpha d_\alpha\,dw_\alpha, and set g(w)=g(c)+∑αdαwαg(w)=g(c)+\sum_\alpha d_\alpha w_\alpha. This also covers a zero-dimensional section by an empty sum. Together with the equations on the Vi(c)V_i(c), this fixes the whole desired differential.

To add qJq_J, J∉AJ\notin A, solve

HqjqJ=0 (j∈A),HpkqJ=δkJ (k∈B).(2.4) H_{q_j}q_J=0\ (j\in A),\qquad H_{p_k}q_J=\delta_{kJ}\ (k\in B). \tag{2.4}

To add pKp_K, K∉BK\notin B, solve

HqjpK=−δjK (j∈A),HpkpK=0 (k∈B).(2.5) H_{q_j}p_K=-\delta_{jK}\ (j\in A),\qquad H_{p_k}p_K=0\ (k\in B). \tag{2.5}

The signs are those of (1.1). The desired tangent covector has exactly these evaluations, so (2.3) supplies it and the chosen marked value. Add the new index to AA or BB. All canonical relations involving the new function hold on the neighborhood, and none of the earlier functions has been altered.

There are finitely many additions. Shrink at each step to preserve independence and then use the common final neighborhood. The full 2n2n differentials at cc are the chosen basis, so the resulting functions are coordinates. Their complete Poisson matrix is the standard matrix by (2.4)–(2.5), including their own zero brackets. Since an invertible Poisson matrix determines the symplectic form, ω=∑dpj∧dqj\omega=\sum dp_j\wedge dq_j in this chart. □\square

Starting with empty A,BA,B gives ordinary Darboux coordinates. Starting with already known functions is stronger: a subsequent coordinate construction can preserve their exact level surfaces and flows, not just a tangent plane.

For example, on T∗R2T^*\mathbb R^2 the functions

q1=x1,p1=ξ1+x22(2.6) q_1=x_1,\qquad p_1=\xi_1+x_2^2 \tag{2.6}

are a prescribed canonical pair. An exact completion is

q2=x2,p2=ξ2+2x1x2.(2.7) q_2=x_2,\qquad p_2=\xi_2+2x_1x_2. \tag{2.7}

The two extra derivatives in {p1,p2}\{p_1,p_2\} cancel. Indeed

p1dq1+p2dq2=ξ1dx1+ξ2dx2+d(x1x22).(2.8) p_1dq_1+p_2dq_2 =\xi_1dx_1+\xi_2dx_2+d(x_1x_2^2). \tag{2.8}

Differentiation proves the symplectic identity. This example is ordinary; its added momentum terms have degree zero in the original covectors and therefore are not homogeneous momentum functions.

3. Retain degrees, marked values and the radial vector

A conic symplectic manifold has a free local dilation action MtM_t, t>0t>0, with Mt∗ω=tωM_t^*\omega=t\omega. Let RR generate the action. Then

LRω=ω,λ=ιRω,dλ=ω.(3.1) \mathcal L_R\omega=\omega,\qquad \lambda=\iota_R\omega,\qquad d\lambda=\omega. \tag{3.1}

A degree-μ\mu function satisfies Rf=μfRf=\mu f. Cartan's formula and (1.1) give

[R,Hf]=HRf−f=(μ−1)Hf.(3.2) [R,H_f]=H_{Rf-f}=(\mu-1)H_f. \tag{3.2}

Consequently degree-zero positions have [R,Hq]=−Hq[R,H_q]=-H_q, while degree-one momenta have [R,Hp]=0[R,H_p]=0.

Theorem 3.1 (homogeneous prescribed coordinates). Let qjq_j, j∈Aj\in A, and pkp_k, k∈Bk\in B, satisfy (2.1) on a conic neighborhood of cc, with degrees zero and one respectively. Suppose

{Hqj(c):j∈A},{Hpk(c):k∈B},R(c)(3.3) \{H_{q_j}(c):j\in A\},\quad \{H_{p_k}(c):k\in B\},\quad R(c) \tag{3.3}

are jointly independent. Choose marked target values a1,…,ana_1,\ldots,a_n, b1,…,bnb_1,\ldots,b_n, agreeing with every prescribed function at cc, and assume

bJ≠0for some J∉A.(3.4) b_J\ne0\quad\text{for some }J\notin A. \tag{3.4}

Then all the missing functions can be constructed so that their degrees, all canonical relations, and all marked values hold. They define a homogeneous symplectomorphism of conic neighborhoods of cc and (a,b)∈T∗Rn∖0(a,b)\in T^*\mathbb R^n\setminus0, retaining every prescribed function identically.

For the completed system, all momentum Hamilton fields, all position Hamilton fields except HqJH_{q_J}, and RR are jointly independent at cc.

Proof, first step: the marked linear completion. Put

εj=−Hqj(c),ek=Hpk(c),r=R(c). \varepsilon_j=-H_{q_j}(c),\qquad e_k=H_{p_k}(c),\qquad r=R(c).

Their additional radial pairings are

ω(r,εj)=0,ω(r,ek)=bk.(3.5) \omega(r,\varepsilon_j)=0,\qquad \omega(r,e_k)=b_k. \tag{3.5}

Use the standard indexed vectors εj0=∂pj\varepsilon_j^0=\partial_{p_j}, ek0=∂qke_k^0=\partial_{q_k} in the target tangent space, and

r0=∑j=1nbjεj0.(3.6) r_0=\sum_{j=1}^n b_j\varepsilon_j^0. \tag{3.6}

Condition (3.4) makes r0r_0 independent of the prescribed target vectors: its εJ0\varepsilon_J^0 coefficient cannot be supplied by their span. The map taking the known vectors and rr to these standard vectors and r0r_0 is an isomorphism of their spans and preserves every pairing, by (2.1) and (3.5). Lemma 1.1 extends it to a full symplectic isomorphism LL. This is our desired coordinate differential at cc.

For this full tangent system, all ek0e_k^0, all εj0\varepsilon_j^0 with j≠Jj\ne J, and r0r_0 are independent. Every intermediate subset of them is independent as well. We will retain these exact tangent values when adding functions.

Second step: solve the homogeneous equations for all coordinates except qJq_J. At the current stage the fields Hqj,Hpk,RH_{q_j},H_{p_k},R are independent, have the commuting Hamilton relations, and satisfy (3.2). Take a transverse section Σ\Sigma through cc for their joint span. The map

Φ(τ,z,ℓ,w)=(∏j∈Aexp⁡(τjHqj))(∏k∈Bexp⁡(zkHpk))Meℓ(w)(3.7) \Phi(\tau,z,\ell,w)= \left(\prod_{j\in A}\exp(\tau_jH_{q_j})\right) \left(\prod_{k\in B}\exp(z_kH_{p_k})\right) M_{e^\ell}(w) \tag{3.7}

is a coordinate chart after shrinking, by its independent differential at the origin. In these coordinates,

Hqj=∂τj,Hpk=∂zk,R=∂ℓ+∑j∈Aτj∂τj.(3.8) H_{q_j}=\partial_{\tau_j},\qquad H_{p_k}=\partial_{z_k},\qquad R=\partial_\ell+\sum_{j\in A}\tau_j\partial_{\tau_j}. \tag{3.8}

F1 of the flow prerequisite proves the variational identity and conjugation formula for these exact brackets. For the last identity, dilation conjugates a degree-zero Hamilton flow with parameter τj\tau_j to one with parameter ehτje^h\tau_j, and preserves the degree-one Hamilton flow parameter zkz_k. This follows from (3.2), or directly from dilation of a cotangent translation. Thus the plus sign in (3.8) is fixed.

To add a position qiq_i, i∉Ai\notin A, i≠Ji\ne J, set

qi={zi+g(w),i∈B,g(w),i∉B.(3.9) q_i= \begin{cases} z_i+g(w),&i\in B,\\ g(w),&i\notin B. \end{cases} \tag{3.9}

This satisfies (2.4) and Rqi=0Rq_i=0. Choose g(c)=aig(c)=a_i and its section differential to match the covector L∗dqi0L^*dq_i^0. That covector has the required evaluations on the Hamilton fields and value zero on R(c)R(c), so all its components are matched.

To add a momentum pKp_K, K∉BK\notin B, set

pK={−τK+eℓg(w),K∈A,eℓg(w),K∉A.(3.10) p_K= \begin{cases} -\tau_K+e^\ell g(w),&K\in A,\\ e^\ell g(w),&K\notin A. \end{cases} \tag{3.10}

This satisfies (2.5) and RpK=pKRp_K=p_K. Choose g(c)=bKg(c)=b_K and its section differential to match L∗dpK0L^*dp_K^0. Its radial evaluation is bKb_K, exactly the derivative of (3.10) in ℓ\ell at the origin. Every marked tangent component is therefore retained.

After either addition the enlarged Hamilton/radial list has the independent marked tangent system already chosen. Shrink so it remains independent nearby. Repeat until all momenta and all positions except qJq_J are constructed. The data have not required translating any homogeneous momentum by a constant.

Third step: construct the exceptional position. Theorem 2.1 completes this almost full system by an ordinary position xJx_J, retaining the desired differential and value aJa_J at cc. Write xi=qix_i=q_i for i≠Ji\ne J. In these ordinary canonical coordinates,

R=∑ipi∂pi+h(x,p)∂xJ,(3.11) R=\sum_i p_i\partial_{p_i}+h(x,p)\partial_{x_J}, \tag{3.11}

because Rpi=piRp_i=p_i and Rxi=0Rx_i=0 for i≠Ji\ne J. Contracting ω\omega gives

ιRω=∑ipi dxi−h dpJ. \iota_R\omega=\sum_i p_i\,dx_i-h\,dp_J.

Equation (3.1) then implies dh∧dpJ=0dh\wedge dp_J=0. Since these are full coordinates, every partial derivative of hh except possibly its pJp_J derivative vanishes. On a small coordinate product h=h(pJ)h=h(p_J). Moreover h(bJ)=RxJ(c)=0h(b_J)=Rx_J(c)=0, by the chosen radial-compatible differential.

Shrink to an interval of pJp_J values avoiding zero, and define

qJ=xJ−∫bJpJh(v)v dv.(3.12) q_J=x_J-\int_{b_J}^{p_J}\frac{h(v)}v\,dv. \tag{3.12}

It has the required value aJa_J, the same differential at cc, and

RqJ=h(pJ)−pJ h(pJ)/pJ=0. Rq_J=h(p_J)-p_J\,h(p_J)/p_J=0.

Adding a function of pJp_J to xJx_J preserves every canonical bracket: it has zero bracket with all momenta, and with each qiq_i, i≠Ji\ne J. Thus the full coordinate system is canonical, with the stated Euler degrees.

Fourth step: pass from a local chart to a conic chart. The nonzero pJp_J gives a transverse level section pJ=bJp_J=b_J. In a sufficiently small ray neighborhood, choose its piece so that each ray meets it once, using the defining local conic chart. Extend the degree-zero functions constantly along each ray and the degree-one functions linearly with the dilation factor. The Euler equations prove agreement with the constructed functions wherever both are defined. The originally prescribed functions already have these same homogeneous extensions, so they remain identical.

The local canonical relations persist under this extension: a bracket of degrees μ,ν\mu,\nu has degree μ+ν−1\mu+\nu-1; the position-position and momentum-momentum brackets are zero, and the mixed brackets have degree zero and equal the specified constants. On the transverse level section the coordinate map is a local diffeomorphism onto a piece of pJ=bJp_J=b_J. Saturating that piece and its source section by positive dilation gives a diffeomorphism of conic neighborhoods. It is symplectic and homogeneous by construction, and the tangent independence follows from (3.6). □\square

The exceptional nonzero value is necessary for this conclusion with the stated independence. In any completed homogeneous chart,

R=∑ipi∂pi=−∑ipiHqi.(3.13) R=\sum_i p_i\partial_{p_i}=-\sum_i p_iH_{q_i}. \tag{3.13}

If every bib_i outside AA were zero, R(c)R(c) would be in the span of the already prescribed position Hamilton fields, contradicting (3.3). In particular AA has at most n−1n-1 elements. A homogeneous symplectomorphism also preserves the primitive itself, because it preserves both ω\omega and RR:

Φ∗(∑ipi dqi)=λ.(3.14) \Phi^*\left(\sum_i p_i\,dq_i\right)=\lambda. \tag{3.14}

4. Straighten a conic isotropic submanifold with zero primitive

Let VV be a conic embedded submanifold of a conic symplectic manifold SS. The radial field is tangent to VV. Therefore

V isotropic⟹λ∣TV=ω(R,⋅)∣TV=0.(4.1) V\text{ isotropic}\quad\Longrightarrow\quad \lambda|_{TV}=\omega(R,\cdot)|_{TV}=0. \tag{4.1}

Conversely, if the pullback of λ\lambda to VV is identically zero, its exterior derivative is the pullback of ω\omega, so VV is isotropic. Vanishing at a single point is not this differential identity.

Theorem 4.1 (conic isotropic fiber model). Let V⊂SV\subset S be a conic isotropic submanifold of dimension kk, with dim⁡S=2n\dim S=2n. Near any c∈Vc\in V, there are homogeneous symplectic coordinates (Q,P)(Q,P) in which

V={Q1=⋯=Qn=0,Pk+1=⋯=Pn=0}.(4.2) V=\{Q_1=\cdots=Q_n=0,\quad P_{k+1}=\cdots=P_n=0\}. \tag{4.2}

The assertion is an equality of conic germs, with the first kk momenta jointly nonzero. Necessarily 1≤k≤n1\leq k\leq n. In particular every conic Lagrangian is locally a nonzero cotangent fiber in suitable homogeneous canonical coordinates.

Proof. Isotropy gives k≤nk\leq n, and the nonzero tangent radial field gives k≥1k\geq1. First we obtain the auxiliary model

q1=0,p2=⋯=pk=0,qj=pj=0(k<j≤n),p1>0.(4.3) q_1=0,\quad p_2=\cdots=p_k=0,\quad q_j=p_j=0\quad(k<j\leq n), \qquad p_1>0. \tag{4.3}

We work by induction in ambient dimension.

Degree-one defining functions with independent normal differentials exist near the marked ray. To see this, choose a positive degree-one ray coordinate ρ\rho and a transverse section ρ=1\rho=1. The conic submanifold is the saturation of an embedded submanifold of this section. Take independent defining functions on that section, extend them constantly along rays, and multiply them by ρ\rho. On VV their differentials are ρ\rho times the independent section normal differentials, so they define the full conormal space.

Case k<nk<n: remove a symplectic normal pair. The Hamilton vectors of those defining functions span (TcV)ω(T_cV)^\omega. Since TcVT_cV is isotropic, the radical of the restricted form on (TcV)ω(T_cV)^\omega is TcVT_cV, and its rank is 2(n−k)>02(n-k)>0. Hence two degree-one defining functions f,gf,g have {f,g}(c)≠0\{f,g\}(c)\ne0. Both vanish on VV. If Hf(c)H_f(c) were proportional to R(c)R(c), then Hfg(c)H_fg(c) would be proportional to Rg(c)=g(c)=0Rg(c)=g(c)=0, a contradiction.

Apply Theorem 3.1 with ff as the prescribed momentum pnp_n, marked value zero, and choose a different marked momentum p1=1p_1=1. This is possible because n≥2n\geq2, Hf(c)H_f(c) and R(c)R(c) are independent, and no position has been prescribed. In its resulting chart, ∂qng={f,g}≠0\partial_{q_n}g=\{f,g\}\ne0. The implicit function theorem expresses g=0g=0 as

qn=a(q1,…,qn−1,p).(4.4) q_n=a(q_1,\ldots,q_{n-1},p). \tag{4.4}

Homogeneity and uniqueness of the solution show that aa has degree zero. The functions r=qn−ar=q_n-a and pn=fp_n=f satisfy {pn,r}=1\{p_n,r\}=1, have degrees zero and one, and both vanish on VV.

Their two Hamilton fields span a nondegenerate symplectic plane. At cc, RR is orthogonal to that plane, since Rr=0Rr=0 and Rpn=pn=0Rp_n=p_n=0. Thus the two fields and RR are independent. Apply Theorem 3.1 again, preserving this actual pair as the last position and momentum and choosing the positive marked momentum at index one. Now VV is contained in the full symplectic section qn=pn=0q_n=p_n=0. This section is conic, of dimension 2(n−1)2(n-1), and contains the same kk-dimensional isotropic VV; its radial field stays nonzero because p1>0p_1>0.

Induct in that section. Extend its homogeneous canonical change to the full neighborhood by leaving the removed pair unchanged. This is a symplectic product extension: the reduced functions depend only on the retained pairs, their pullback preserves the retained sum of dpj∧dqjdp_j\wedge dq_j, and the unchanged pair preserves the last summand. Its inverse is the reduced inverse times the identity. The reduced homogeneous functions keep their degrees under full simultaneous dilation. Shrink so the remaining covector components stay nonzero. Each such step removes one pair with both coordinates zero on VV.

Case k=n>1k=n>1: remove a position cylinder. Here VV is Lagrangian. Its conormal Hamilton span is TcVT_cV, of dimension n>1n>1; choose a degree-one defining function ff whose Hamilton vector is not proportional to R(c)R(c). Normalize it as pn=fp_n=f, with p1(c)=1p_1(c)=1, exactly as above.

Because ff vanishes on VV, its Hamilton vector lies in (TV)ω=TV(TV)^\omega=TV along VV. Thus the coordinate field Hpn=∂qnH_{p_n}=\partial_{q_n} is tangent to VV. Local flow uniqueness shows that VV is a position cylinder:

V={small qn interval}×V0,pn=0,(4.5) V=\{\text{small }q_n\text{ interval}\}\times V_0, \qquad p_n=0, \tag{4.5}

where V0=V∩{qn=pn=0}V_0=V\cap\{q_n=p_n=0\}. Choose the degree-zero position origin so qn(c)=0q_n(c)=0. The section is transverse to this tangent coordinate field, so V0V_0 has dimension n−1n-1; its restricted symplectic form is zero. It is conic, and its radial field remains nonzero in the positive first momentum. Induct on the reduced conic Lagrangian V0V_0. Extend the reduced chart by the unchanged cylinder pair. Each such step supplies one free position and its zero momentum.

Base n=k=1n=k=1. Use a homogeneous canonical chart with positive momentum. On the one-dimensional VV, (4.1) reads p dq=0p\,dq=0, with p>0p>0. Hence qq is constant on its local connected germ. Subtract that constant from this degree-zero position. The remaining momentum is a local coordinate on VV, since the radial tangent has Rp=p≠0Rp=p\ne0. Thus V={q=0,p>0}V=\{q=0,p>0\} as a germ.

The normal-pair steps and cylinder steps, with a permutation of indices, give exactly (4.3). The first momentum stays positive in the marked conic neighborhood.

It remains to turn (4.3) into (4.2) by an actual homogeneous symplectomorphism. On p1>0p_1>0, define

Q1=q1+∑j=2kqjpj/p1,P1=p1,Qj=pj/p1,Pj=−qjp1(2≤j≤k),Qj=qj,Pj=pj(j>k).(4.6) \begin{aligned} Q_1&=q_1+\sum_{j=2}^k q_jp_j/p_1,& P_1&=p_1,\\ Q_j&=p_j/p_1,&P_j&=-q_jp_1\quad(2\leq j\leq k),\\ Q_j&=q_j,&P_j&=p_j\quad(j>k). \end{aligned} \tag{4.6}

Every QjQ_j has degree zero and every PjP_j degree one. Direct differentiation gives the exact primitive identity

P1 dQ1+∑j=2kPj dQj=p1 dq1+∑j=2kpj dqj.(4.7) P_1\,dQ_1+\sum_{j=2}^k P_j\,dQ_j =p_1\,dq_1+\sum_{j=2}^k p_j\,dq_j. \tag{4.7}

The unchanged spectator terms complete it to ∑PjdQj=∑pjdqj\sum P_jdQ_j=\sum p_jdq_j. The inverse is

p1=P1,qj=−Pj/P1,pj=QjP1(2≤j≤k),q1=Q1+∑j=2kPjQj/P1,(4.8) \begin{split} p_1&=P_1,\qquad q_j=-P_j/P_1,\quad p_j=Q_jP_1\quad(2\leq j\leq k),\\ q_1&=Q_1+\sum_{j=2}^k P_jQ_j/P_1, \end{split} \tag{4.8}

with unchanged spectators. Thus this is a full homogeneous symplectic coordinate change, not just a parametrization of VV. Substitution maps (4.3) to Q=0,Pj=0Q=0,P_j=0 for j>kj>k, and its inverse maps that entire local model back to (4.3). Empty sums when k=1k=1 give the identity. This proves (4.2). □\square

A literal ordinary exchange qj↔pjq_j\leftrightarrow p_j would interchange degrees zero and one. The divisions and products by p1p_1 in (4.6) supply the correct homogeneous replacement, on a cone where that denominator is nonzero.

5. Recognize conormal germs where the base projection has constant rank

For an embedded submanifold Y⊂XY\subset X, its nonzero conormal is

N∗Y∖0={(x,ξ):x∈Y, ξ∣TxY=0, ξ≠0}.(5.1) N^*Y\setminus0 =\{(x,\xi):x\in Y,\ \xi|_{T_xY}=0,\ \xi\ne0\}. \tag{5.1}

At each nonzero conormal point it is conic Lagrangian. In adapted base coordinates Y={xr+1=⋯=xn=0}Y=\{x_{r+1}=\cdots=x_n=0\}, it is defined by those equations and ξ1=⋯=ξr=0\xi_1=\cdots=\xi_r=0. Its local dimension is nn, and the pullback of ξ⋅dx\xi\cdot dx is zero. Differentiating gives isotropy, hence the Lagrangian property. If YY is open in XX, there are no nonzero conormal points.

Theorem 5.1 (constant-rank conormal recognition). Let Λ⊂T∗X∖0\Lambda\subset T^*X\setminus0 be conic Lagrangian, dim⁡X=n\dim X=n. If the base projection π:Λ→X\pi:\Lambda\to X has constant rank rr near (x0,ξ0)(x_0,\xi_0), then there is an embedded base germ YY of dimension rr near x0x_0 such that

Λ=N∗Y(5.2) \Lambda=N^*Y \tag{5.2}

as germs near (x0,ξ0)(x_0,\xi_0). Necessarily r<nr<n. This does not assert equality with every conormal direction or with the whole global conormal bundle.

Proof. Here is the required constant-rank map argument. In local coordinates on Λ\Lambda and XX, select a nonzero rr-by-rr minor of dπd\pi at the marked point, and shrink so it stays nonzero. After reordering coordinates, use the first rr components of π\pi, together with the remaining domain coordinates, as new domain coordinates (u,v)(u,v). The inverse function theorem applies because the Jacobian of this change is block triangular with that invertible minor. In these coordinates π(u,v)=(u,h(u,v))\pi(u,v)=(u,h(u,v)). Its derivative has first block (I,0)(I,0); constant rank rr therefore forces every entry of ∂vh\partial_v h to vanish. On a small coordinate product the fundamental theorem of calculus along line segments gives h(u,v)=h(u,v0)h(u,v)=h(u,v_0). Consequently its local image is the embedded graph Y={(u,h(u,v0))}Y=\{(u,h(u,v_0))\}, and dπ(TΛ)=TYd\pi(T\Lambda)=TY. For r=0r=0, all derivatives of π\pi vanish and the same line-segment argument makes the map constant; its image is the zero-dimensional embedded germ. This proves the required conclusion in every rank without importing an unproved constant-rank theorem. The primitive vanishes on TΛT\Lambda by (4.1). Therefore for each (x,ξ)∈Λ(x,\xi)\in\Lambda in this neighborhood and each v∈TxYv\in T_xY, choose w∈T(x,ξ)Λw\in T_{(x,\xi)}\Lambda with dπ(w)=vd\pi(w)=v. Then

ξ(v)=λ(w)=0. \xi(v)=\lambda(w)=0.

Thus Λ⊂N∗Y∖0\Lambda\subset N^*Y\setminus0 locally. Both are embedded nn-dimensional manifolds; the inclusion has injective differential and is a local diffeomorphism by the inverse function theorem. Its image is open in N∗YN^*Y, proving the germ equality after shrinking. Finally the nonzero radial field is tangent to Λ\Lambda and lies in ker⁡dπ\ker d\pi, so r≤n−1r\leq n-1. □\square

The constant-rank locus is open and dense in Λ\Lambda. For density, inside any nonempty open patch choose the largest integer rank attained there. A nonzero minor at a point of that rank stays nonzero nearby; maximality in the patch prevents a larger rank, so this smaller neighborhood has constant rank. Openness is built into the definition of this locus. Conormal descriptions therefore cover an open dense subset, although the remaining base caustics can be singular.

Here is an original deformed-cusp example. On T∗R2T^*\mathbb R^2, take ∣r∣<1/4|r|<1/4, ρ>0\rho>0, and

x=(r2,−23r3−12r4),ξ=((r+r2)ρ,ρ).(5.3) x=\left(r^2,-\frac23r^3-\frac12r^4\right),\qquad \xi=((r+r^2)\rho,\rho). \tag{5.3}

The exact cancellation is

ξ⋅dx=ρ(r+r2)(2r dr)+ρ(−2r2−2r3) dr=0.(5.4) \xi\cdot dx =\rho(r+r^2)(2r\,dr) +\rho(-2r^2-2r^3)\,dr=0. \tag{5.4}

The parametrization is an embedding in this range: ρ=ξ2\rho=\xi_2, and ξ1/ξ2=r+r2\xi_1/\xi_2=r+r^2 uniquely determines rr, since 1+2r>01+2r>0. Its differential has rank two even at r=0r=0. Thus it is a smooth conic Lagrangian with a singular base image.

The base projection has rank one for r≠0r\ne0 and rank zero at zero. Its image satisfies

(x2+12x12)2=49x13.(5.5) \left(x_2+\frac12x_1^2\right)^2=\frac49x_1^3. \tag{5.5}

The smooth base change y1=x1, y2=x2+x12/2y_1=x_1,\ y_2=x_2+x_1^2/2 reveals the ordinary semicubical cusp. Theorem 5.1 gives a conormal germ at each regular point. It cannot give a conormal germ at the cusp: the base projection of a conormal bundle has locally constant rank equal to the dimension of its base submanifold, whereas the rank of (5.3) changes in every neighborhood of r=0r=0.

Two projections of a smooth conic Lagrangian with a cusp in its base image

Figure 5.1. Both panels use (5.3) on the slice ρ=1\rho=1, with −0.24≤r≤0.24-0.24\leq r\leq0.24. The upper projection keeps (x1,x2)(x_1,x_2) and develops a cusp. The lower keeps (ξ1/ξ2,x1)=(r+r2,r2)(\xi_1/\xi_2,x_1)=(r+r^2,r^2), whose derivative at zero is (1,0)(1,0). That projection already detects immersion of the lifted curve; the full conic Lagrangian also retains x2x_2 and the free positive radial variable. The curves are numerical samples of the exact formulas, not evidence replacing (5.4) or the embedding argument.

6. Exercises with complete solutions

Exercise 6.1 (dual-vector correction; foundational). In Lemma 1.1 verify (1.3), including its sign. What would happen with the opposite sign?

Solution. Since ω(ki,kj)=0\omega(k_i,k_j)=0, the dual pairings in (1.2) are unchanged. In the pairing of the corrected li,ljl_i,l_j, the correction in the first slot contributes −Aij/2-A_{ij}/2; the correction in the second contributes Aji/2=−Aij/2A_{ji}/2=-A_{ij}/2, because ω(li,kj)=−δij\omega(l_i,k_j)=-\delta_{ij}. The total is zero. With the opposite sign the two corrections add AijA_{ij}, leaving 2Aij2A_{ij}, which need not vanish.

Exercise 6.2 (ordinary shear; foundational). Compute all canonical brackets for (2.6)–(2.7), and explain why the transformation fails the homogeneous momentum condition.

Solution. The positions commute. The bracket {p1,q1}\{p_1,q_1\} equals one, {p2,q2}\{p_2,q_2\} equals one, and both other position-momentum brackets vanish. For the remaining bracket, using {f,g}=fξ⋅gx−fx⋅gξ\{f,g\}=f_\xi\cdot g_x-f_x\cdot g_\xi,

{p1,p2}=2x2−2x2=0. \{p_1,p_2\}=2x_2-2x_2=0.

Under ξ↦tξ\xi\mapsto t\xi, p1p_1 becomes tξ1+x22t\xi_1+x_2^2, which differs from tp1t p_1 unless x22=0x_2^2=0 or t=1t=1. The same issue occurs for p2p_2. Thus it is a full ordinary canonical map, with exact primitive difference (2.8), but not a homogeneous one.

Exercise 6.3 (flow signs; intermediate). Derive both (2.4) and (2.5) from the desired canonical brackets. Why do the field equations have constant compatible right sides?

Solution. For a new qJq_J, the equations are {qj,qJ}=0\{q_j,q_J\}=0 and {pk,qJ}=δkJ\{p_k,q_J\}=\delta_{kJ}, hence (2.4). For a new pKp_K, antisymmetry turns {pK,qj}=δKj\{p_K,q_j\}=\delta_{Kj} into {qj,pK}=−δjK\{q_j,p_K\}=-\delta_{jK}; all old momentum brackets are zero, giving (2.5). Every old pair has constant bracket, so their Hamilton fields commute by Jacobi. Constant right sides have zero derivatives along those fields, which gives the commuting compatibility needed for (2.3).

Exercise 6.4 (radial sign in a flow chart; intermediate). On a two-dimensional cotangent chart take Hq=−∂pH_q=-\partial_p, R=p∂pR=p\partial_p, and initial momentum b≠0b\ne0. Write the momentum after first dilating by eℓe^\ell and then flowing by τHq\tau H_q. Verify the radial field in these parameters.

Solution. The momentum is p=eℓb−τp=e^\ell b-\tau. A further dilation by ehe^h transforms it to eℓ+hb−ehτe^{\ell+h}b-e^h\tau, so (τ,ℓ)↦(ehτ,ℓ+h)(\tau,\ell)\mapsto(e^h\tau,\ell+h). Its generator is ∂ℓ+τ∂τ\partial_\ell+\tau\partial_\tau, which applied to pp gives eℓb−τ=pe^\ell b-\tau=p. The negative sign instead would give eℓb+τe^\ell b+\tau, so it would fail the momentum Euler equation.

Exercise 6.5 (exceptional position; intermediate). In a canonical chart suppose R=∑pi∂pi+h(pJ)∂xJR=\sum p_i\partial_{p_i}+h(p_J)\partial_{x_J}, with bJ≠0b_J\ne0 and h(bJ)=0h(b_J)=0. Verify all claimed properties of (3.12).

Solution. The integral is smooth on an interval avoiding zero and vanishes at pJ=bJp_J=b_J. Its derivative there is h(bJ)/bJ=0h(b_J)/b_J=0, so the marked value and differential of xJx_J are retained. Euler differentiation gives RqJ=h(pJ)−pJh(pJ)/pJ=0Rq_J=h(p_J)-p_J h(p_J)/p_J=0. Since only a function of pJp_J is added to xJx_J, all brackets with the pip_i stay unchanged. For i≠Ji\ne J, its bracket with qi=xiq_i=x_i is zero because {xi,pJ}=0\{x_i,p_J\}=0. The coordinate change is invertible by adding the integral back, so all canonical identities are preserved.

Exercise 6.6 (necessary marked momentum; intermediate). Explain why, under (3.3), the completed marked momentum vector cannot be supported only in the set AA. Can the given independent Hamilton fields include every position field?

Solution. In any homogeneous canonical chart (3.13) holds. If bi=0b_i=0 outside AA, then R(c)=−∑i∈AbiHqi(c)R(c)=-\sum_{i\in A}b_iH_{q_i}(c), contradicting its independence from the prescribed position fields. Thus at least one nonzero marked momentum lies outside AA. If AA contained all positions, this contradiction would be unavoidable. Therefore ∣A∣≤n−1|A|\leq n-1; the completed independence must omit one position Hamilton field with nonzero corresponding momentum.

Exercise 6.7 (homogeneous rotation; advanced). For k=2k=2, verify the primitive identity and inverse of (4.6). Identify the image of the auxiliary model, retaining spectator coordinates.

Solution. Here Q1=q1+q2p2/p1Q_1=q_1+q_2p_2/p_1, Q2=p2/p1Q_2=p_2/p_1, P1=p1P_1=p_1, P2=−q2p1P_2=-q_2p_1. Thus

P1dQ1+P2dQ2=p1dq1+p2dq2+q2p1d(p2/p1)−q2p1d(p2/p1). P_1dQ_1+P_2dQ_2 =p_1dq_1+p_2dq_2+q_2p_1d(p_2/p_1)-q_2p_1d(p_2/p_1).

The last terms cancel. Solving gives q2=−P2/P1q_2=-P_2/P_1, p2=Q2P1p_2=Q_2P_1, q1=Q1+P2Q2/P1q_1=Q_1+P_2Q_2/P_1. For the auxiliary model q1=p2=0q_1=p_2=0, qj=pj=0q_j=p_j=0 for j>2j>2, all QjQ_j vanish and all PjP_j for j>2j>2 vanish, while P1>0P_1>0 and P2P_2 are free locally. This is the exact fiber model of dimension two, with unchanged spectators.

Exercise 6.8 (zero primitive as a germ; advanced). Prove both directions of (4.1) for a conic submanifold. Why does vanishing of λ\lambda only at one tangent space not suffice to prove isotropy on a neighborhood?

Solution. If VV is isotropic, both RR and each v∈TVv\in TV are tangent, so λ(v)=ω(R,v)=0\lambda(v)=\omega(R,v)=0. Conversely, let i:V→Si:V\to S be inclusion and assume i∗λ=0i^*\lambda=0 as a one-form on VV. Then i∗ω=i∗dλ=d(i∗λ)=0i^*\omega=i^*d\lambda=d(i^*\lambda)=0, which is isotropy. A value of a one-form at one point determines none of its first derivatives; its exterior derivative there may be nonzero. The converse therefore needs an identically zero pullback on the germ, not just a zero value at the marked tangent space.

Exercise 6.9 (deformed cusp; advanced). Verify the two ranks in (5.3), the base equation (5.5), and the nonzero tangent of the lower panel in Figure 5.1.

Solution. The base derivative in rr is (2r,−2r2−2r3)(2r,-2r^2-2r^3), while the base derivative in ρ\rho is zero. Thus the base rank is one for r≠0r\ne0 and zero at zero. At r=0r=0, the full rr derivative is (0,0,ρ,0)(0,0,\rho,0) and the full ρ\rho derivative is (0,0,0,1)(0,0,0,1), which are independent because ρ>0\rho>0. Away from zero the inverse using ρ\rho and r+r2r+r^2 proves full rank two throughout the specified interval. Since x1=r2x_1=r^2, x2+x12/2=−2r3/3x_2+x_1^2/2=-2r^3/3, squaring gives (5.5). The lower projection derivative is (1+2r,2r)(1+2r,2r), equal to (1,0)(1,0) at zero.

Exercise 6.10 (the germ qualification; advanced). Let Y={x2=0}⊂R2Y=\{x_2=0\}\subset\mathbb R^2. Describe a conic Lagrangian germ equal to a piece of N∗YN^*Y while omitting an entire sign of conormal directions. State its base rank and compare it with the cusp.

Solution. Take Λ={x2=0,ξ1=0,ξ2>0}\Lambda=\{x_2=0,\xi_1=0,\xi_2>0\}. Its parameters are (x1,ξ2)(x_1,\xi_2), it has dimension two, and its primitive is zero. It is an open conic subset of the nonzero conormal of YY, whose negative ξ2\xi_2 directions it omits. At every marked point of Λ\Lambda, Theorem 5.1 gives equality of the germ with N∗YN^*Y, with base rank one. This is not global equality with both signs of that conormal. Its rank is constant; the cusp relation has rank zero at the caustic and rank one arbitrarily close, so no such smooth conormal germ exists at that caustic.

7. Scope and author checks

The proof completes ordinary prescribed canonical functions, their full homogeneous extension with arbitrary compatible marked values, the zero-primitive conic isotropic normal form, and constant-rank conormal recognition. It preserves actual functions, exact Poisson signs, degree zero and one, the exceptional nonzero momentum, all induction dimensions, and local conic-germ qualifications. Ten original graded exercises have complete solutions.

The preserved proof and its exact current programme dependencies were reviewed for this restoration; independent human mathematical review remains pending. Bounded exact checks verify the alternating dual correction, an ordinary nonlinear canonical shear, semidirect radial flow signs and homogeneous equations, the final position correction, the primitive-preserving rotation in several dimensions, and the deformed cusp's ranks and primitive. The reproducible figure was inspected as two projections of a stated radial slice. These checks do not certify arbitrary smooth flows or substitute for the general proofs.

Clean conic Lagrangian pair normal forms, stable phase equivalence, Maslov topology, broader symbol classes, propagation, hyperbolic and mixed problems, glancing and complex-phase mathematics, and every other unfinished assigned course strand remain active.

Sources and restoration

Original lesson, ten solutions and illustration: GPT-6.1 Sol (OpenAI), Ultra, September 2026, CC0. Restoration and exact prerequisite review: GPT-6 Astra (OpenAI), Ultra, 5 October 2026. Original additions here are CC0. Linked components retain their individual licences. No book file or text is included.