Skip to content
Shadow Theory

Section 2 9 October 2026

A compact stock-preserving baker module

Reading position 3 of 16

2 A compact stock-preserving baker module

The preparation map uses the same two scalar positions repeatedly: a scratch or writer coordinate xx and an archive coordinate yy. In the effective electromagnetic realization they are two Cartesian coordinates of one charged planar carrier. This restriction matters: an arbitrary vector potential on a many-particle configuration space is not automatically a local electromagnetic field.

Use dimensionless time τ=ωt\tau=\omega t and positions in the common Gaussian standard deviation σ\sigma, where σ2=ℏ/(2mω)\sigma^2=\hbar/(2m\omega). Write

φ(x)=(2π)−1/4e−x2/4,ϕ(x)=∣φ(x)∣2,C(x,y)=(Φ(x),Φ(y))=(u,v), \varphi(x)=(2\pi)^{-1/4}e^{-x^2/4},\qquad \phi(x)=|\varphi(x)|^2,\qquad C(x,y)=(\Phi(x),\Phi(y))=(u,v),

with Φ\Phi the standard Gaussian CDF. The complete active stock has amplitude φ(x)φ(y)\varphi(x)\varphi(y) and density Γ(x,y)=ϕ(x)ϕ(y)\Gamma(x,y)=\phi(x)\phi(y). The CDF coordinates are used to design and analyze deterministic controls; no actual coordinate is sampled from that reference density by definition.

The discontinuous reference baker is

B(u,v)=(2u−s,v+s2),s=⌊2u⌋. B(u,v)=\left(2u-s,\frac{v+s}{2}\right), \qquad s=\lfloor2u\rfloor . (2.1)

We construct a smooth compact Hamiltonian flow whose completed map agrees with this entire affine map on two large rectangles. It is unnecessary and impossible to treat the discontinuous baker itself as a globally smooth physical flow.

2.1 Smooth rounded squares with an area clock

For 0<h≤1/160<h\leq1/16 choose the even integer p=2⌈2/h⌉p=2\lceil2/h\rceil, so p≥64p\geq64 and ph≥4ph\geq4. Let

Θ(s)=e−1/se−1/s+e−1/(1−s)(0<s<1),Θ(s)=0 (s≤0),Θ(s)=1 (s≥1). \Theta(s)= \frac{e^{-1/s}}{e^{-1/s}+e^{-1/(1-s)}}\quad(0<s<1), \qquad \Theta(s)=0\ (s\leq0),\quad\Theta(s)=1\ (s\geq1).

It is smooth with flat endpoint joins. We shall use

0≤Θ′≤2,∣Θ′′∣<200. 0\leq\Theta'\leq2,\qquad |\Theta''|<200. (2.2)

For completeness, set ℓ=1/(1−s)−1/s\ell=1/(1-s)-1/s. Then Θ′=ℓ′Θ(1−Θ)\Theta'=\ell'\Theta(1-\Theta) and ∣Θ′′∣≤(∣ℓ′′∣+ℓ′2)e−∣ℓ∣|\Theta''|\leq(|\ell''|+\ell'^2)e^{-|\ell|}. Writing z=1/min⁡(s,1−s)≥2z=1/\min(s,1-s)\geq2 bounds the latter by e2(z4+2z3+8z2+32)e−ze^2(z^4+2z^3+8z^2+32)e^{-z}. The maxima of zke−zz^ke^{-z}, together with e2<9e^2<9, e4>256/5e^4>256/5, e3>27/2e^3>27/2 and e2>32/5e^2>32/5, bound it by 9(5+4+5+5)=171<2009(5+4+5+5)=171<200. For the first derivative, with a=∣2s−1∣a=|2s-1| its exact formula is

Θ′=2(1+a2)(1−a2)2sech⁡2 ⁣(2a1−a2)≤2, \Theta'= \frac{2(1+a^2)}{(1-a^2)^2} \operatorname{sech}^2\!\left(\frac{2a}{1-a^2}\right) \leq2,

using cosh⁡2z≥1+z2\cosh^2 z\geq1+z^2.

Define

Rp(θ)=(cos⁡pθ+sin⁡pθ)−1/p,χ(r)=Θ(4r−1),R(r,θ)=1+χ(r)(Rp(θ)−1),Q(r,θ)=rR(r,θ)(cos⁡θ,sin⁡θ),K=R+rRr,w=RK.\begin{align}R_p(\theta)&=(\cos^p\theta+\sin^p\theta)^{-1/p},\notag\\ \chi(r)&=\Theta(4r-1),\qquad R(r,\theta)=1+\chi(r)(R_p(\theta)-1),\notag\\ Q(r,\theta)&=rR(r,\theta)(\cos\theta,\sin\theta),\qquad K=R+rR_r,\quad w=RK. \tag{2.3}\end{align}

These curves are circles for r≤1/4r\leq1/4 and homogeneous rounded squares for r≥1/2r\geq1/2. They are strictly nested since K≥1K\geq1. The intermediate curves need not be convex.

Lemma 2.1 (Area-normalized quarter-turn)

The chart in (2.3) gives a unique smooth radius r(q)>0r(q)>0 for every q≠0q\ne0. The function

A(r)=r22∫02πR(r,θ)2 dθ,H∗(q)=A(r(q)) A(r)=\frac{r^2}{2}\int_0^{2\pi}R(r,\theta)^2\,d\theta, \qquad H_*(q)=A(r(q)) (2.4)

is smooth also at the origin. Its Hamiltonian vector field is J∇H∗=(−∂q2H∗,∂q1H∗)J\nabla H_*=(-\partial_{q_2}H_*,\partial_{q_1}H_*). Every positive level has period one; its time 1/41/4 map is exactly the counterclockwise geometric quarter-turn, at every point on the level. The inverse quarter-turn is obtained at time −1/4-1/4. All intermediate points stay inside the square with axis intercept rr.

Proof

In the first octant, t=tan⁡θ∈[0,1]t=\tan\theta\in[0,1] gives

∂θlog⁡Rp=t−tp−11+tp∈[0,1]. \partial_\theta\log R_p =\frac{t-t^{p-1}}{1+t^p}\in[0,1].

Symmetry implies 1≤Rp≤21\leq R_p\leq\sqrt2 and ∣Rp′∣≤Rp|R'_p|\leq R_p globally. Thus K≥1K\geq1 gives radial invertibility, and the inverse function theorem gives smoothness off the origin. Near the origin the chart is ordinary polar coordinates and H∗=π∣q∣2H_*=\pi|q|^2, so there is no origin singularity. Flat joins in χ\chi make the intermediate patch smooth. Since R≤RpR\leq R_p and Rp∣cos⁡θ∣,Rp∣sin⁡θ∣≤1R_p|\cos\theta|,R_p|\sin\theta|\leq1, the level stays in its axis-intercept square.

Let Z(r)=∫02πw(r,θ) dθZ(r)=\int_0^{2\pi}w(r,\theta)\,d\theta. Differentiating (2.4) gives Ar=rZA_r=rZ. The chart Jacobian is det⁡(Qr,Qθ)=rRK=rw\det(Q_r,Q_\theta)=rRK=rw. Since the Hamiltonian is constant on each level, its equations are r˙=0\dot r=0 and θ˙=Ar/(rw)=Z/w\dot\theta=A_r/(rw)=Z/w. The lifted angular clock

T(r,θ)=1Z(r)∫0θw(r,s) ds \mathcal T(r,\theta)=\frac1{Z(r)} \int_0^\theta w(r,s)\,ds (2.5)

therefore satisfies T˙=1\dot{\mathcal T}=1. Fourfold symmetry gives T(r,θ+π/2)=T(r,θ)+1/4\mathcal T(r,\theta+\pi/2)=\mathcal T(r,\theta)+1/4, proving the exact endpoint and period claims. Each path remains on its level, which proves confinement. For r≥1/2r\geq1/2 the enclosed area is the full rounded-square area Apr2A_p r^2: changing its inner foliation subtracts no area offset.

□

Let

ψh(r)=1−Θ ⁣(r−(1−h)h/2),Hc(q)=∫0r(q)As(s)ψh(s) ds−∫01−h/2As(s)ψh(s) ds. \psi_h(r)=1-\Theta\!\left(\frac{r-(1-h)}{h/2}\right), \qquad H_c(q)=\int_0^{r(q)}A_s(s)\psi_h(s)\,ds -\int_0^{1-h/2}A_s(s)\psi_h(s)\,ds .

The stream HcH_c is smooth and zero outside a compact subset of (−1,1)2(-1,1)^2. Its vector field is exactly ψh(r)J∇H∗\psi_h(r)J\nabla H_*, and ∣ψh′∣≤4/h|\psi'_h|\leq4/h. The cutoff is applied to the complete stream derivative; merely cutting a velocity or omitting a scalar compensation would not have the same conservation property.

Theorem 2.2 (Complete compact baker isotopy)

Define

Lh=[h,12−h]×[h,1−h],Rh=[12+h,1−h]×[h,1−h]. L_h=[h,\tfrac12-h]\times[h,1-h],\qquad R_h=[\tfrac12+h,1-h]\times[h,1-h].

There is a smooth area-preserving isotopy of the open unit square, compactly supported inside it and flat at its temporal endpoints, whose time-one map ThT_h agrees with BB on Lh∪RhL_h\cup R_h. The excluded set has exact reference area

βh=1−(1−4h)(1−2h)=6h−8h2. \beta_h=1-(1-4h)(1-2h)=6h-8h^2. (2.6)

The complete flow stays inside the square at every intermediate time and is the identity near its boundary.

Proof

First the whole inner square [−1+2h,1−2h]2[-1+2h,1-2h]^2 is contained in the unchanged region r<1−hr<1-h. In the outer homogeneous region, its rounded-square radius is at most 21/p(1−2h)2^{1/p}(1-2h), and

2(1−2h1−h)p≤2e−ph≤2e−4<1. 2\left(\frac{1-2h}{1-h}\right)^p \leq2e^{-ph}\leq2e^{-4}<1.

Points whose homogeneous radius is below 1/21/2 lie inside the outer half-radius curve, so their actual nested radius is also below 1/2<1−h1/2<1-h. This covers the patched center as well.

The first quantile Hamiltonian is Hg(u,v)=Hc(2u−1,2v−1)/4\mathcal H_g(u,v)=H_c(2u-1,2v-1)/4. Its time 1/41/4 map is (u,v)↦(1−v,u)(u,v)\mapsto(1-v,u) on [h,1−h]2[h,1-h]^2. The divisor four is the coordinate Jacobian. The second Hamiltonian is the sum of two disjoint terms

Hb(u,v)=−Hc(2u−1,4v−1)8−Hc(2u−1,4v−3)8. \mathcal H_b(u,v)= -\frac{H_c(2u-1,4v-1)}8 -\frac{H_c(2u-1,4v-3)}8 . (2.7)

Each support is strictly inside its own horizontal half-square, with a positive gap between them. The divisor eight is its coordinate Jacobian, and the negative sign gives clockwise motion. On their unchanged regions the maps are respectively (x,y)↦(2y,(1−x)/2)(x,y)\mapsto(2y,(1-x)/2) and (x,y)↦(2y−1,1−x/2)(x,y)\mapsto(2y-1,1-x/2).

The global quarter-turn sends LhL_h into the lower unchanged region and RhR_h into the upper one. In normalized coordinates their first coordinate is 1−2v1-2v, bounded by 1−2h1-2h in absolute value, and their second is 4u−14u-1 or 4u−34u-3, bounded by 1−4h1-4h. Their composed maps are exactly (2u,v/2)(2u,v/2) and (2u−1,(v+1)/2)(2u-1,(v+1)/2). Every intermediate point follows a confined complete level; ordinary rigid rotation of square corners, which would leave the square, has not been used.

For each of two consecutive stages of duration 1/21/2, use the nonnegative rate g(τ)=Θ′(2τ)/2g(\tau)=\Theta'(2\tau)/2 in its local stage time. It integrates to 1/41/4, has maximum at most one, and is flat at both ends. Multiplying the corresponding Hamiltonian by this rate preserves its autonomous endpoint by time reparameterization. Smooth compact Hamiltonian fields have complete diffeomorphic area-preserving flows and smooth joins. The two rectangles have combined area (1−4h)(1−2h)(1-4h)(1-2h), proving (2.6).

□
Proposition 2.3 (Exact inverse decoder)

Let

Eh=B(Lh∪Rh)=[2h,1−2h]×([h2,1−h2]∪[1+h2,1−h2]). E_h=B(L_h\cup R_h)= [2h,1-2h]\times \left([\tfrac h2,\tfrac{1-h}2] \cup[\tfrac{1+h}2,1-\tfrac h2]\right).

Reversing the two stages with reversed signs gives a smooth duration-one inverse module. On EhE_h it has the exact map

s=⌊2v⌋,v+=2v−s,u+=u+s2. s=\lfloor2v\rfloor,\qquad v^+=2v-s,\qquad u^+=\frac{u+s}{2}. (2.8)

In particular the decoded bit does not require a Born or independent actual scratch coordinate uu. The exceptional reference area is again βh\beta_h.

Proof

The inverse flow is realized by time reversal of the compact Hamiltonian path. The baker is a bijection almost everywhere with inverse (2.8); direct substitution verifies both branches. Area preservation maps the two original rectangles to EhE_h with unchanged total area.

□

2.2 What the first-flow-derivative estimate controls

Theorem 2.4 (Whole-time first jets and one-module labels)

For the uncut level flow Φs\Phi_s of H∗H_*,

∥DΦs∥op, ∥DΦs−1∥op≤384875<400000 \|D\Phi_s\|_{\rm op},\ \|D\Phi_s^{-1}\|_{\rm op} \leq384875<400000 (2.9)

for every ss, independently of p,hp,h. For the cut flow, uniformly for ∣s∣≤1/4|s|\leq1/4,

∥DΦs∥op, ∥DΦs−1∥op≤Dh:=400000+160/h. \|D\Phi_s\|_{\rm op},\ \|D\Phi_s^{-1}\|_{\rm op} \leq D_h:=400000+160/h. (2.10)

On the guarded unchanged tube of one forward or inverse baker module, its quantile prefix and inverse prefix derivatives have norm below 800000800000. For its moving physical inverse label ητ=Fτ−1∘C\eta_\tau=F_\tau^{-1}\circ C this gives the sufficient Euclidean-to-ℓ1\ell^1 bound

∥Dητ∥2→1<106=:Bgood. \|D\eta_\tau\|_{2\to1}<10^6=:B_{\rm good}. (2.11)

For an inverse module one may enter from E2hE_{2h} and retain this bound until the virtual entrance label changes by h/8h/8 in ℓ1\ell^1. Its original entrance exclusion has area β2h=12h−32h2\beta_{2h}=12h-32h^2. Without that good-tube restriction, 2Dh22D_h^2 is a safe one-module quantile prefix bound.

In physical standardized coordinates, the generator has amplitude at most

∣V∣≤400/h |V|\leq400/h (2.12)

per unit pulse rate; it therefore holds for the duration-one protocol above. These are first-flow-derivative and amplitude bounds, not uniform bounds on all spatial derivatives of the fields or on a cumulative many-cycle inverse.

Proof

On the fixed central annulus, rχ′≤4r\chi'\leq4 and r2∣χ′′∣≤800r^2|\chi''|\leq800. With Rp−1<1/2R_p-1<1/2, the chart quantities satisfy

R≤32,K≤72,1≤w≤214,∣rRr∣≤2,∣rKr∣≤404,∣rwr∣≤613<616. R\leq\tfrac32,\quad K\leq\tfrac72,\quad 1\leq w\leq\tfrac{21}4,\quad |rR_r|\leq2,\quad |rK_r|\leq404,\quad |rw_r|\leq613<616 .

For example rKr=2rRr+r2RrrrK_r=2rR_r+r^2R_{rr} and rwr=(rRr)K+R(rKr)rw_r=(rR_r)K+R(rK_r). These bounds involve only the first angular derivative of RpR_p and the fixed radial step jets.

The inverse-radius gradient has radial and angular components 1/K1/K and −Rθ/(RK)-R_\theta/(RK), so ∣∇r∣≤2<2|\nabla r|\leq\sqrt2<2. For the clock (2.5),

12π(21/4)≤Tθ≤21/42π,∣rTr∣≤1232,∣∇T∣≤2465/r. \frac1{2\pi(21/4)}\leq\mathcal T_\theta \leq\frac{21/4}{2\pi},\qquad |r\mathcal T_r|\leq1232,\qquad |\nabla\mathcal T|\leq2465/r .

The two quotient terms in Tr\mathcal T_r each cost at most 616616; one can take θ∈[0,2π]\theta\in[0,2\pi]. Other angular lifts add an integer to T\mathcal T and have the same derivatives. The last bound follows from ∇T=Tr∇r+Tθ∇θ\nabla\mathcal T=\mathcal T_r\nabla r+ \mathcal T_\theta\nabla\theta and ∣∇θ∣≤1/r|\nabla\theta|\leq1/r.

Regard QQ now as the inverse chart in (r,T)(r,\mathcal T). Since ∣Qθ∣≤(9/4)r|Q_\theta|\leq(9/4)r,

∣rθr∣T≤1232⋅2π(21/4)≤40656, |r\theta_r|_{\mathcal T}\leq1232\cdot2\pi(21/4) \leq40656 ,

using π<22/7\pi<22/7. It follows that ∣Qr∣T<100000|Q_r|_{\mathcal T}<100000 and ∣QT∣<75r|Q_{\mathcal T}|<75r. The full uncut flow is Q(r,T+s)Q(r,\mathcal T+s); differentiation therefore gives 100000⋅2+75r(2465/r)=384875100000\cdot2+75r(2465/r)=384875. The inverse uses −s-s and the same estimate. The circular origin supplies its continuous smooth extension.

For the cut flow replace T+s\mathcal T+s by T+sψh(r)\mathcal T+s\psi_h(r). Its extra derivative is at most 75r(1/4)(4/h)2≤150/h75r(1/4)(4/h)2\leq150/h on its support. Outside the support the map is the identity. This proves (2.10).

The global square conjugation has condition number one; the half-square conjugations have condition number two. On a good completed stage, the map is exactly an affine quarter-turn, so it costs only one or two, respectively. At most one incomplete stage incurs the uncut bound. Thus any prefix or inverse prefix of one module costs less than 800000800000 on that tube. Since ∥DC∥op≤ϕ(0)<2/5\|DC\|_{\rm op}\leq\phi(0)<2/5, conversion to ℓ1\ell^1 gives 2⋅800000⋅ϕ(0)<480000<106\sqrt2\cdot800000\cdot\phi(0)<480000<10^6. The inverse-label estimate involves the forward CDF derivative. The set E2hE_{2h} is at least h/2h/2 in ℓ1\ell^1 distance from the complement of EhE_h. The stipulated h/8h/8 drift therefore keeps the virtual entrance label in EhE_h, where the preceding good-level proof applies. The corresponding forward version uses L2h∪R2hL_{2h}\cup R_{2h}. The excluded sets have the stated areas, by the same rectangle calculation. Using cut derivatives for both stages instead gives 2Dh22D_h^2.

Finally Ar=rZ≤(21π/2)rA_r=rZ\leq(21\pi/2)r and ∣∇r∣<2|\nabla r|<2, so ∣J∇H∗∣≤21πr<66|J\nabla H_*|\leq21\pi r<66 on the compact support. An inverse linear coordinate conjugation costs at most 1/21/2, giving quantile speed at most 3333 per unit rate. Every support quantile is at least h/8h/8 from zero and one. The function I(u)=ϕ(Φ−1(u))I(u)=\phi(\Phi^{-1}(u)) is concave, since I′′=−1/I<0I''=-1/I<0, and symmetric. Its chords imply I(u)≥2ϕ(0)min⁡(u,1−u)I(u)\geq2\phi(0)\min(u,1-u). Hence ∥DC−1∥≤42π/h<12/h\|DC^{-1}\|\leq4\sqrt{2\pi}/h<12/h on the support. The physical standardized speed is below 33⋅12/h=396/h<400/h33\cdot12/h=396/h<400/h.

□

On each fixed good cell of an nn-cycle baker, DB−n=diag⁡(2−n,2n)DB^{-n}=\operatorname{diag}(2^{-n},2^n). Thus BgoodB_{\rm good} is a one-module bound; applying it to the cumulative inverse would be incorrect. Likewise the angular and cutoff higher derivatives can grow with pp and 1/h1/h. The estimate has removed an unnecessary first-flow-jet penalty without making all physical field jets or controller resources small.