Spatial Cutoffs, Compact Affine Cores, and Localization Residuals

Good to continue from the layer schedule. We now replace the previous lesson’s globally defined oscillatory waves with objects that are physically admissible on ${\mathbb{R}}^2$: smooth, compactly supported fields. The price of this localization is not merely technical. It introduces specific residuals into both Boussinesq equations, and the construction has to ensure that those residuals never disturb the smaller regions in which later layers will operate.

The central design principle is:

$$\boxed{\text{A new layer is supported entirely inside regions where every older layer is exactly affine.}}$$

That is stronger than saying the new layer is “near the origin,” or merely inside the support of an older layer. This lesson separates the geometric regions involved, explains the nesting inequalities, and then tracks precisely what a spatial cutoff breaks.

---

Localizing a wave without destroying the local affine model

Recall the idealized layer from Module 1. It had a rapidly oscillating phase and was analysed on an affine background,

$${\theta }_{\mathrm{old}}(x,t)=G(t)\cdot x,u_{\mathrm{old}}(x,t)=D(t)x.$$

That description is useful locally, but neither $G\cdot x$ nor a plane wave such as $\sin (\lambda \zeta \cdot x)$ is compactly supported. A literal global plane wave carries infinite spatial extent, and an affine velocity grows with $|x|$. Neither can be the final finite-energy construction on ${\mathbb{R}}^2$.

The paper addresses this with two linked devices:

1. a periodic phase profile that agrees exactly with a linear function near phase $0$;
2. a smooth, compactly supported material envelope that confines the layer to a small moving region.

For one layer, suppress the layer index and introduce:

$$a=M(t{)}^{-1}x,s=\lambda \zeta (t)\cdot x.$$

Here:

• $a$ is the material coordinate: it labels a point as transported by the affine background flow;
• $s$ is the fast oscillatory phase;
• $M(t)$ is the deformation matrix of the affine flow;
• $\zeta (t)$ is the transported wavevector;
• $\lambda$ is the large frequency scale.

Let $F(s)$ be smooth, periodic, and mean zero, but chosen so that

$$F(s)=s\text{when}|s|<s_0.$$

Its periodic primitive $P$ then satisfies, in that same interval,

$$P(s)=P(0)+\frac{s^2}{2}.$$

This is an important compromise. Globally, $F$ remains a benign periodic profile; locally, near a selected phase cell, it behaves exactly like the linear function that produces the desired affine fields.

The envelope is a radial smooth cutoff in material space:

$$g(a)=f\left(\frac{a}{\ell }\right),$$

where $f=1$ on $B_{1/2}$ and $f=0$ outside $B_1$. Thus $g=1$ for $|a|\le \ell /2$, transitions smoothly in the annulus $\ell /2<|a|<\ell$, and vanishes for $|a|\ge \ell$.

Schematically, the leading temperature and streamfunction profiles are

$$V_0(s,a,t)=w(t)\Theta (t)F(s)g(a),$$ $${\Psi }_0(s,a,t)=\frac{w(t)\Omega (t)}{{\lambda }^2|\zeta (t){|}^2}P(s)g(a).$$

The physical fields are obtained by evaluating these at $s=\lambda \zeta \cdot x$ and $a=M^{-1}x$. Localizing the streamfunction, rather than simply cutting off the velocity, guarantees that the resulting velocity remains divergence free:

$$u={\nabla }^{\perp }\psi .$$

That is a structural requirement, not a cosmetic choice.

---

Support, plateau, and affine core: three genuinely different sets

It is tempting to say “the cutoff is one near the origin,” and stop there. But the construction needs three nested notions, each serving a separate purpose.

For layer $q$, the paper defines:

$$T_q(t)=M_q(t)B_{{\ell }_q},$$ $$P_q(t)=M_q(t)B_{{\ell }_q/2},$$ $$A_q(t)=P_q(t)\cap \left\{x:|{\lambda }_q{\zeta }_q(t)\cdot x|<s_0\right\}.$$

Their roles are as follows.

Set	Definition in words	What is true there?	Why it matters
$T_q$	Full support of layer $q$	The layer may be nonzero; cutoff derivatives may be nonzero	All localization errors live here
$P_q$	Plateau of the envelope	$g_q=1$, so envelope derivatives vanish	Higher corrections vanish here
$A_q$	Plateau intersected with the linear phase cell	$g_q=1$ and $F(s)=s$	The layer is exactly affine here

The implication structure is

$$A_q(t)\subset P_q(t)\subset T_q(t).$$

But the inclusions should not be read as mere bookkeeping. The plateau $P_q$ is not necessarily an affine zone. The cutoff is constant there, but the phase $s={\lambda }_q{\zeta }_q\cdot x$ may range through many periods of $F$. The field remains oscillatory unless the additional phase constraint $|s|<s_0$ is imposed.

Inside the affine region $A_q$, however, the wave profile becomes linear:

$$F({\lambda }_q{\zeta }_q\cdot x)={\lambda }_q{\zeta }_q\cdot x.$$

Consequently the leading fields have the desired local form:

$${\theta }_q(x,t)=w_q{\Theta }_q{\lambda }_q{\zeta }_q\cdot x,$$ $$u_q(x,t)=D_q(t)x,{\omega }_q(x,t)=w_q{\Omega }_q.$$

The exact matrix $D_q$ is a trace-free affine strain/rotation matrix determined by the layer’s orientation and vorticity amplitude. The important point is that it is independent of $x$ within $A_q$.

So, locally, a compact wave layer impersonates the global affine-wave model used in the amplitude analysis. Outside the affine core it is a genuine oscillatory, compactly supported field; inside it, it is exactly the simpler field needed to initiate and control later layers.

---

Why the cutoff is placed in material coordinates

The envelope is $g_q(M_q^{-1}x)$, not a fixed spatial cutoff $g_q(x)$. This keeps the support attached to the deformation generated by the affine background.

For a field written as

$$A^{♮}(x,t)=A(\lambda \zeta (t)\cdot x,M(t{)}^{-1}x,t),$$

the affine transport operator has the useful identity

$$({\partial }_t+D(t)x\cdot {\nabla }_x)A^{♮}=({\partial }_{t}A{)}^{♮}.$$

In particular, if $g(a)$ has no explicit time dependence, then it does not create an additional leading transport residual under the affine background. The envelope moves with the fluid described by $D(t)x$.

This is why the paper calls it a material envelope. A fixed spatial bump would be continually sheared across by the affine flow, producing large and unnecessary transport errors. In material coordinates, the bump is stationary; its physical image moves and deforms with $M(t)$.

There is another useful consequence. Since the background flow is incompressible,

$$\det M(t)=1.$$

The support may stretch in one direction and compress in another, but its area is preserved under the affine deformation. What must be controlled is its maximal physical radius, which motivates a bound of the form

$$\Vert M_j(t)\Vert ,\Vert M_j(t{)}^{-1}\Vert \le K_j.$$

The factor $K_j$ measures how much an older layer’s material geometry can distort during its remaining lifetime.

---

Nested affine cores: the condition that allows iteration

Suppose layer $q$ is newly activated, while layers $1,\dots ,q-1$ already exist. The entire support of the new layer must lie inside the affine region of each older layer:

$$T_q(t)\subset A_j(t)\text{for every }j<q,$$

throughout the full lifetime of layer $q$.

This condition achieves several things at once:

• Each older temperature field is literally linear throughout $T_q$.
• Each older velocity field is literally affine throughout $T_q$.
• Each older vorticity is spatially constant throughout $T_q$.
• The new layer therefore sees the exact affine coefficients assumed in the one-layer amplitude system.
• No uncontrolled Taylor-expansion remainder from older fields enters the next amplification calculation.

To see the two constraints behind this inclusion, take $x\in T_q(t)$. The physical support radius obeys

$$|x|\le K_q{\ell }_q.$$

For $x$ to lie in the plateau of older layer $j$, its old material coordinate must obey

$$|M_j^{-1}x|<\frac{{\ell }_j}{2}.$$

Using $\Vert M_j^{-1}\Vert \le K_j$, it is sufficient that

$$K_j{K}_q{\ell }_q<\frac{{\ell }_j}{2}.$$

This is the material-radius condition. It says that even after the older deformation is undone, the support of the new layer fits well within the older plateau.

For $x$ to lie in the older linear phase cell, one also needs

$$|{\lambda }_j{\zeta }_j\cdot x|<s_0.$$

Since $|{\zeta }_j|$ is bounded by a constant $C_{\zeta }$, it suffices that

$$C_{\zeta }{\lambda }_j{K}_q{\ell }_q<s_0.$$

This is the phase-linearity condition. It is more restrictive than mere spatial containment because an older layer may have a very high frequency: even a small physical region can traverse many of its phase oscillations.

The construction’s scale separation is designed so that ${\ell }_q$ shrinks drastically as $q$ grows. It must shrink fast enough to beat both:

1. the possible growth of the older deformation factors $K_j$;
2. the older frequency factors ${\lambda }_j$.

In the paper’s schedule, later frequency scales are separated so aggressively that these inequalities hold simultaneously for every earlier $j<q$, not merely for the immediately preceding layer.

This explains the word nested in a precise sense. Layer $q$ is not simply localized near the origin. Its whole support lies in the common intersection of the affine cores of every older layer that remains dynamically relevant.

---

A local architecture with a protected central region

It helps to picture one layer radially in material coordinates:

Material location	Envelope $g$	Phase requirement	Role
$	a	<\ell/2$,with$	s
$	a	<\ell/2$,butoutside$	s
$\ell/2<	a	<\ell$	transitions from $1$ to $0$
$	a	\geq\ell$	$0$

The later correction fields are constructed to vanish not only on the smaller affine region, but on the whole plateau $|a|<\ell /2$. That stronger property is essential. A future layer’s support only needs to fit inside the older affine core, but requiring corrections to vanish throughout the plateau gives a buffer: the complicated correction machinery remains separated from the protected central state.

The paper also uses compactly supported streamfunctions for this separation. In two dimensions, an arbitrary vorticity distribution outside a central ball can still induce velocity inside that ball through the nonlocal Biot–Savart law. But if a correction is built from a compact streamfunction that itself vanishes on that ball, its induced velocity vanishes there exactly. This lets the correction annulus remain dynamically silent in the core.

---

What localization breaks

Before localization, the leading temperature and streamfunction profiles were chosen so that the principal terms in the Boussinesq equations cancel once the amplitude pair satisfies its designed ODE. Multiplying those profiles by $g(a)$ breaks that exact cancellation.

The basic reason is simply the product rule. For example,

$${\nabla }_a(F(s)g(a))=F(s){\nabla }_{a}g(a)$$

when differentiating with respect to the slow material variable. Likewise, differentiating the localized streamfunction to form vorticity produces terms involving

$${\nabla }_{a}g,{\nabla }_a^{2}g.$$

These are zero on the plateau and outside the support, but are nonzero in the transition annulus. Thus all genuinely spatial-localization residuals are confined to that annulus.

There are several related kinds of residual.

1. Envelope-gradient terms in the temperature equation

The temperature equation involves transport by the localized velocity. When the streamfunction and temperature each carry the envelope $g$, their interaction generates terms with factors such as

$$g{\nabla }_{a}g.$$

These arise in the nonlinear advective bracket. They vanish where $g$ is constant, but not where the layer is being smoothly switched off in space.

2. Envelope-derivative terms in the vorticity equation

Vorticity is obtained from the Laplacian of the streamfunction. The Laplacian sees both the fast phase and the slow envelope:

$${\Delta }_x{\Psi }^{♮}=({\lambda }^2\kappa {\partial }_s^2\Psi +\lambda D_1\Psi +{\Delta }_M\Psi {)}^{♮}.$$

The first term is the dominant fast-phase contribution familiar from the plane-wave calculation. The other terms mix phase differentiation with material-envelope differentiation or differentiate the envelope twice. They are the unavoidable elliptic cost of compact support.

3. Nonlinear interactions involving those cutoff terms

Once localized vorticity contains envelope-gradient pieces, nonlinear vorticity transport generates further terms. The construction therefore cannot cancel a single cutoff error and declare success: correcting one residual can create higher-order residuals. This is why the paper uses a finite hierarchy of corrections.

4. Nonoscillatory phase averages

Some residuals oscillate in the fast phase and can be targeted with a phase antiderivative. Others survive phase averaging:

$$⟨R⟩(a,t)=\frac{1}{2\pi }{\int }_0^{2\pi }R(s,a,t)ds.$$

Such an average is no longer oscillatory in $s$, but it need not be spatially constant: it can still depend on the slow material variable $a$. The paper does not invert these terms using the oscillatory phase mechanism. It retains them as part of the external force and estimates them directly.

A key scaling benefit is that differentiating a phase average in physical space does not produce a large $\lambda$ factor from the fast phase. That does not make the average harmless automatically, but it makes it much easier to control than an uncancelled oscillatory error at frequency $\lambda$.

5. Activation residuals: related, but conceptually distinct

The temporal activation factor $w_q(t)$ smoothly turns a layer on and off. Its derivative produces terms involving ${\dot{w}}_q$. These are residuals too, but they are not caused by spatial localization. Unlike cutoff-gradient residuals, activation terms need not vanish in the central core during the short activation period.

This distinction is worth retaining:

Residual source	Supported in cutoff annulus?	Vanishes on plateau?
Derivatives of $g$	Yes	Yes
Interactions containing ${\nabla }_{a}g$	Yes	Yes
Phase averages of localization interactions	Yes	Yes
Temporal activation terms involving $\dot{w}$	Not necessarily	Not necessarily

---

---

How corrections preserve the nesting geometry

The correction scheme must satisfy two competing goals:

• It must cancel enough of the cutoff residual to make the final forcing smooth.
• It must not alter the affine state that later layers rely on.

The solution is to solve for correction profiles in $(s,a)$ coordinates and to use an inverse only in the periodic phase variable $s$. At each finite correction level, the nonzero phase modes of the preceding residual become a source for another coupled temperature-streamfunction correction.

Because phase inversion acts at fixed $a$, it does not spread material support. Because the source terms contain derivatives of the envelope and earlier corrections, they vanish throughout the plateau. Therefore the corrections themselves vanish there as well.

The resulting hierarchy has the schematic form

$$\text{localized leading layer}+\text{finite corrections}=\text{controlled state}+\text{residual force}.$$

This should not be interpreted as a literal temporal sequence; all the profiles are assembled as one designed layer. The point is algebraic: each additional correction removes a further order of oscillatory localization error.

After finitely many correction levels, the residual force consists of:

$$\begin{array}{rl}& \text{activation terms}\\ & +\text{nonoscillatory phase averages}\\ & +\text{high-order oscillatory remainder}.\end{array}$$

The next lesson addresses why, by increasing the correction depth as frequency increases, the complete sum of these residual forces can converge smoothly even though the layers themselves reach increasingly high frequencies.

---

A practical paper-reading checklist

When navigating the construction, use the following test whenever the paper introduces a set, a cutoff, or an error term.

1. Which coordinate is the cutoff expressed in?
   Here it is material coordinate $a=M^{-1}x$, so the envelope travels with the affine flow.

2. Is the point in the full support, plateau, or affine region?
   Only the last gives the exact affine formula; “cutoff equals one” alone is not enough.

3. Does the expression contain a derivative of $g$?
   If so, it is confined to the material transition annulus and vanishes on the plateau.

4. Does the term have nonzero fast-phase frequency?
   If yes, it is a candidate for cancellation by the phase-based correction recursion.

5. Does it survive phase averaging?
   If yes, it is retained in the external forcing rather than removed through phase inversion.

6. Could it change the state in the core used by future layers?
   The answer must be no for every correction. This is ensured by their vanishing on the plateau and by the nesting $T_q\subset A_j$.

---

Takeaways

A compact layer is built from a periodic wave profile and a material cutoff. The cutoff gives finite spatial support while the phase profile supplies a smaller region in which the wave is exactly linear and the induced velocity is exactly affine.

The geometric hierarchy is essential:

$$\text{full support}\supset \text{plateau}\supset \text{affine core}.$$

By making successive supports extremely small, the construction ensures that each new layer remains, for its whole lifetime, in the affine core of every older layer. That is what makes the previously derived amplitude mechanism applicable repeatedly without accumulating unknown background errors.

Spatial localization inevitably introduces derivatives of the envelope into the temperature, vorticity, and nonlinear interaction terms. These localization residuals live in a surrounding annulus, vanish on the plateau, and are handled by a finite correction hierarchy. Oscillatory phase modes are successively canceled; activation terms, phase averages, and sufficiently high-order remainders are left for the forcing estimates.

Next, we will examine the derivative bookkeeping behind that final step: why taking more correction levels as the frequency rises makes the total external forcing converge in every required mixed space-time derivative.