Explanation of the manuscript's Section 5: correcting the base flow to every order (pages 45 to 62)

Section 5, explained: Correcting the base flow to every order (pages 45 to 62)

The problem this section is handed

Section 4 leaves a leading axisymmetric field (u(0), p(0)) built from profiles E, U, V0, Π of the similarity variables X = r²/(2q) and η, where the concentration scale q is comparable to τ = 1 - t in the core. It satisfies two balances exactly: incompressibility, and the leading radial balance in which pressure supplies the centripetal force [p. 7, p. 8]. In the azimuthal and axial momentum equations with axial viscosity deleted, its residual is minus the cylindrical divergence of a stress T that vanishes outside the annulus Xa < X < Xb [p. 9, p. 10, p. 45].

Two effects were left out: axial viscosity in the tangential equations, and, in the radial equation, everything except pressure and centrifugal force (the time derivative, transport and viscosity of the radial velocity) [p. 45, p. 46]. At fixed similarity coordinates each is smaller than the terms kept by one factor of q^{2h}, the squared ratio of radial to axial length [p. 8, p. 46], but only relatively: by our arithmetic from the exponents on p. 46, axial viscosity on the leading swirl has size q^{-3/2+h}, unbounded, and since h < 1/100 [p. 4], one factor of q^{2h} is a weak gain (at q = 10^{-10} it still exceeds 0.6).

The plan hands later sections residual of exactly two shapes [p. 7]. One is a stress with two tangential components inside the annulus, which the waves of Section 7 supply [p. 9, p. 12]. The other is flat: every Cartesian derivative bounded by every power of q [p. 11], the form that lets the final force be smooth with all derivatives controlled [p. 7, p. 16], carried separately by later stages [p. 14]. So Section 5 must build an exactly divergence-free axisymmetric background whose residual is minus the divergence of an annular stress plus a flat remainder EB, leave the heat exterior untouched, and keep that stress close enough to T for the waves' positive representation to apply [p. 11, p. 45, p. 60].

What a person would try first, and why it fails

Hand the leftovers to the waves, or to the force. The omitted terms act near the axis [p. 45], in the core where the stress must vanish and no wave is placed, and the radial defect has no slot in the two-component wave target [p. 9]. Nor can an unbounded residual serve as a smooth force, as the paper notes for the annular residual [p. 3]. Near the axis, the flow itself must cancel these terms.

Solve the full equations for one better self-similar profile. Our reading: this does not close. The full axisymmetric equations are not invariant under the anisotropic scaling; axial viscosity and the remaining radial terms carry an explicit q^{2h} relative to the rest [p. 46], so a q-independent profile balances only the terms sharing the leading power. The anisotropy itself (slender core, unbounded swirl Reynolds number q^{-h}) is what the construction is built on [p. 3, p. 4, p. 8].

Expand in q^{2h}, solve each order, sum the series. This is the right start, but the naive version breaks three times. The coefficient bounds may grow too fast for the series to converge, and the paper never claims convergence [p. 45, p. 56]. An outward solve from the axis that loses radius at each order would stop short of the annulus at high orders; our reading is that this is why the paper insists on one radial interval for every order [p. 47]. And cutting a solved order off in radius leaves the radially integrated fields without compact support: the pressure can keep an exterior constant and the stress an exterior tail [p. 49].

The idea, motivated

Count powers. At fixed (X, η), each term the leading profile balances in the tangential equations (time derivative, radial and axial transport, radial viscosity) multiplies the field by q^{-1}; for axial transport this uses A + D = 1 [p. 8, p. 47]. Two axial derivatives instead cost q^{-2D} = q^{-1}q^{2h}, and in r times the radial equation the non-centrifugal terms sit at q^{-1} = q^{-2A}q^{2h} [p. 46]. Both defects are late by exactly one factor of q^{2h}, so one expansion can treat both. That is the ansatz (5.1): order n carries the extra power q^{λ_n}, λ_n = 2nh, order zero is the leading profile, and the swirl is written through a coefficient ϕ_n that is smooth at the axis [p. 45, p. 46] (M5.1).

Substitute and collect. Collecting the coefficient of q^{2nh} gives the order-n system (5.2) to (5.6) [p. 46] (M5.2). What order n - 1 omitted returns as a known source at order n: axial viscosity of order n - 1 in the tangential equations, and the radial remainder Ω_{n-1} in the pressure equation [p. 46, p. 47]. The system is also linear, because in the quadratic sums over i + j = n the order-n unknown meets only the fixed leading profile [p. 47]. Pressure enters the axial equation, so it is solved jointly with ϕ_n and U_n [p. 46]. A warning sign: the operators contain η-derivatives of the unknowns, so each order is a PDE in (X, η), not an ODE in X [p. 46, p. 48].

Solve exactly near the axis, on one interval for every order. The stress must stay zero near the axis, so each order is solved exactly there, with zero axis values so the leading axis traces never move [p. 45, p. 47]. The interval must reach into the inner collar of the annulus [p. 47]; in our reading, an interval shrinking with n would let high orders leave stress in the core, where nothing absorbs it. Lemma 5.1 writes the order-n problem as a first-order system (5.7) in ξ = √X (proportional to r), adding the unknown K_n = A_X(U_n) - U_n so the radial average becomes a local equation; an explicit integral operator started at the axis inverts the singular diagonal part [p. 47, p. 48] (M5.3). Each Picard step integrates once in ξ but may differentiate once in η, and each η-derivative shrinks the complex η-neighborhood [p. 48]. The saving observation: the coefficient A1 of ∂_η maps the first four unknowns into the last two and kills the last two, so no two derivative factors are adjacent and k steps carry at most ⌈k/2⌉ derivatives [p. 48] (M5.4). The factorial from the integration simplex then beats the derivative loss, the k-th root of (5.8) tends to zero, and the series converges on any finite interval whatever the constant C_n [p. 48]. Our reading: the system is second order in the radius and first order in η, a heat equation run sideways from the axis, which is why holomorphy in η is the natural hypothesis. Parity finishes the lemma: the profiles are even in ξ, hence smooth in X, hence smooth across the axis [p. 49] (M5.5).

Extend, and see what the cutoff leaves. The obvious extension is a radial cutoff of the inner solution [p. 50]. But the streamfunction, radial velocity, pressure and stress are radial integrals from the axis, and such an integral vanishes past its source exactly when its total integral vanishes; the stress primitives (5.9) use the integrating factors R² and R [p. 49, p. 51]. Hence five total moments, functions of η: axial flux m_{n,1}, angular momentum m_{n,2}, pressure increment m_{n,3}, axial transport of angular momentum m_{n,4}, and axial momentum flux with pressure m_{n,5} [p. 49, p. 52, p. 53] (M5.6). The first and third close the streamfunction and the pressure [p. 50, p. 52]. For the stress, integrate the conservative forms of the two tangential residuals against r² dr and r dr: every radial flux drops out, leaving time and axial derivatives of m_{n,1}, m_{n,2}, m_{n,4}, m_{n,5} and of order n - 1 moments that already vanish [p. 53] (M5.9). Since the power of q in a product does not depend on how i + j = n splits, each integral vanishes identically before those derivatives act [p. 53]. At order one the leading angular momentum integral diverges (our reading, from the power-law swirl tail on p. 8), so the paper subtracts a z-independent pure power that axial viscosity cannot see [p. 53].

Impose the moments where they are cheap. Five bumps, two added to U_n and three to E_n, sit on a patch Ipos reserved in Section 4, where the leading profile has no axial velocity and a pure-power swirl [p. 50, p. 51] (M5.7). There the moments are affine in the five coefficients and split into 2 × 2 and 3 × 3 matrices of distinct powers against disjoint ordered bumps, invertible by Lemma A.1, so discrepancies of any size are solvable [p. 51, p. 52]. All integrated fields then vanish past a radius X₊ independent of n, placed beyond the leading axial perturbation because the order-zero radial remainder can be nonzero past the positive-order patches [p. 50, p. 52] (M5.8). The velocity on the patch reserved for later mean corrections stays exactly leading [p. 52]. For n ≥ 2 the stress lies in [X₋, X₊]; the order-one stress reaches the outer edge Xb but is bounded there by the edge weight ζ times inverse powers of the log distance δ [p. 50, p. 54] (M5.10).

Truncate. Each finite sum is divergence-free, satisfies the radial balance, and has tangential residual exactly minus a stress divergence at every retained order, leaving an error O(q^{2h(N+1)-K_m}) with K_m independent of N [p. 55] (M5.11). The loss is order independent because a physical derivative costs a fixed power of q whichever order it hits; differentiating q^{λ_n} only feeds the constants [p. 55, p. 56].

Sum without convergence. Multiply the n-th term by χ(c_n q) with c_n growing fast [p. 56] (M5.13). Cutting the velocity directly would break incompressibility, since q depends on z and t (our reading); cutting the Stokes streamfunction potential and then taking the curl keeps it (the θ-independent swirl needs no potential [p. 58]), and m_{n,1} = 0 makes that potential compactly supported [p. 52, p. 56] (M5.12). Derivatives of χ(aq) are bounded independently of a, since aq lies between 1/2 and 1 on the transition region of χ, and each scale is chosen so that half of that order's power of q beats its constant [p. 58]. Comparison with one long partial sum, where all its cutoffs equal one, shows the residual is flat [p. 59]. Lemma 5.4 is stated abstractly so that Section 9 can reuse it [p. 14, p. 56, p. 57]. Proposition 5.5 delivers the identity (5.41) with flat EB, closeness to the leading field (5.42), weighted closeness of the stress to T (5.43), an untouched exterior, a mean patch left exactly leading (5.44), and limits at t = 1 away from the origin, with every asymptotic coefficient unchanged [p. 7, p. 60, p. 61, p. 62] (M5.14).

What breaks without each move

• M5.1 (q^{2h} expansion): the omitted terms stay at size q^{-3/2+h}, with nothing organizing them into solvable pieces.
• M5.2 (linear order-n system): each order becomes a nonlinear problem, and the next pressure source Ω_n/X may be singular at the axis.
• M5.3 (first-order system with K_n): a nonlocal radial average and 1/X coefficients remain, with no explicit solution operator from the axis.
• M5.4 (at most ⌈k/2⌉ η-derivatives): the solvable radius depends on C_n and shrinks along the induction, leaving high-order stress in the core.
• M5.5 (parity): coefficients are smooth in r but not known smooth in r², hence not Cartesian smooth at the axis.
• M5.6 (five moments): pressure, streamfunction and stress keep exterior constants and R^{-2}, R^{-1} tails where no wave acts.
• M5.7 (bumps on Ipos): the moment conditions no longer reduce to fixed invertible matrices, and adjusting elsewhere disturbs the inner solution or the exterior.
• M5.8 (closure past X₊): the pressure fails to close where the order-zero radial remainder is nonzero, and positive orders reach the heat exterior.
• M5.9 (conservative forms): zero moments are not known to kill the total residual integrals, so the stress may keep an outer tail.
• M5.10 (edge bound on T1): the order-one stress may exceed the weight ζ near Xb, beyond what wave amplitudes carrying √ζ [p. 18] can absorb.
• M5.11 (N-independent loss): decay gained in value is lost in derivatives, and the summation hypothesis (5.33) fails.
• M5.12 (streamfunction potentials): multiplying velocities by χ(c_n q) destroys exact incompressibility.
• M5.13 (shrinking cutoffs): a possibly divergent formal series defines no actual smooth field with a flat residual.
• M5.14 (realized base field): nothing certifies that summation kept the exterior, the mean patch, closeness to T, and the limits at t = 1.

Three questions the section leaves open

1. How fast do the constants C_{n,m} of (5.17) grow in n [p. 52]? Under a factorial bound, the series in q^{2h} might be summed with an exponentially small remainder at an explicit rate, instead of a flat remainder fixed by arbitrary cutoff scales. That would make EB quantitative and the background less dependent on choices, which matters for any numerical check.

2. Is holomorphy in η needed? Lemma 5.1 works on an η-neighborhood that may depend on the order [p. 47]. Our reading of the count behind (5.8): since k steps carry at most ⌈k/2⌉ η-derivatives, a Gevrey class in η of order below 2 might still give convergence on every finite interval. The answer would show which property of the analytic axis profiles [p. 10] the higher orders use, and whether the fixed interval survives other leading profiles.

3. Is the edge loss in the order-one stress intrinsic? T1 reaches Xb because axial viscosity still acts on the leading swirl out to Xb; near the edge that swirl is the heat field times a multiplier depending on z through X [p. 54], and the bound on T1 carries inverse powers of δ [p. 50]. Section 7 keeps the positive representation for T0 alone and treats the rest by signed corrections [p. 61]. If a different terminal multiplier or a sixth moment confined T1 inside the annulus, as for n ≥ 2, the downstream corrections would face no δ losses from the base.

Page tags

p. 3, p. 4, p. 7, p. 8, p. 9, p. 10, p. 11, p. 12, p. 14, p. 16, p. 18, p. 45, p. 46, p. 47, p. 48, p. 49, p. 50, p. 51, p. 52, p. 53, p. 54, p. 55, p. 56, p. 57, p. 58, p. 59, p. 60, p. 61, p. 62.

Extraction note: pages 45 to 62 all flatten sub- and superscripts and fractions (q2nh means q^{2nh}), worst at (5.8) on p. 48 and on p. 54, where the terminal multiplier prints as "fo". No page was unreadable.