Home·The theorem·The proof in one reading·Architecture·Units·Manuscript·Glossary·FAQ·Work items·Data·Method·Provenance·Changes
← III.10 One full nonlinear renormalization step: bounds on irrelevant terms (notes)Part III·Unit outlineIII.12 Bounds uniform in the volume and convergence as the spacing tends to zero (notes) →

III.11. Comparison with the free Gaussian measure: interacting rank and determinant budgets — detailed notes

Contents

1. What this unit proves 1

2. What it uses 6

3. Strategy 8

4. Steps 9

4.1 Haar estimates and the determinant budget 9

4.2 Free covariances 11

4.3 Polynomial calculus and identities 12

4.4 From frequent good scales to trial systems (III.11.4) 13

4.5 The rank comparison (III.11.5) 14

4.6 The stress tensor (III.11.6, III.11.7) 16

5. Where the difficulty lies 17

6. Notation 18

7. What remains unexplained 19

The full account of unit III.11: every proposition with its proof sketch, checked against the Lean in two rounds and edited. The condensed outline and the map of the Lean are on the unit page.

Unit page · All units · Overview

Part: III, Ultraviolet construction (renormalization group). Lean: chapter 24: YM/Next/RANK/, YM/PhysicalNext/AFTERNOON_INTERACTING_RANK_COMPARISON/, YM/PhysicalNext/RANK_COVARIANCE_ATLAS/, YM/MergeProof/ (with West/, North/, East/, South/). Size: 42,556 lines (about 5,300 declarations).

Scope. The unit compares the interacting renormalization-group (RG) history with the free Maxwell field. It also contains the stress-flux argument and the label algebra of the Maxwell equation ideal.

1. What this unit proves

The unit has two outputs. The first is the rank comparison. In the continuum limit, the space \(F_d\) of local fields of degree \(d\) is isomorphic to the corresponding space of the free Maxwell field, compatibly with complex conjugation and with the lattice symmetry group. Here \(F_d\) is the space of labels of degree \(d\) modulo labels whose correlations vanish. The second output is the stress tensor of the limit: a nonzero conserved stress tensor that satisfies the translation and rotation Ward identities and is normalized against the free Maxwell stress tensor along a scaling sequence. Both outputs rest on quantitative estimates proved here: a per-site determinant budget, chessboard bounds for connected parts of corrected observables, and comparisons of covariances with the free curvature kernel. The trial systems test a finite separating family of free one- and two-point correlations.

Notation. Set \(\mu_k = \exp ((\log \log \hat{\beta}_k)^2)\). The centered estimate and the trial systems use this fixed schedule quantity. \(G\) is a compact matrix group, assumed simple only where stated; \(\dim G\) is the dimension of its Lie algebra; \(h\) is normalized Haar measure; \(\text{chord} (u) \geqslant 0\) is the chordal distance from \(u\) to \(1\); and \(\text{shell} (r) = \{u \in G : \text{chord} (u) \leqslant r\}\). The torus of side \(n \geqslant 2\) is \(\Lambda_n = (\mathbb{Z}/ n)^4\), with \(c = n^4\) sites, \(4 c\) bonds and \(6 c\) plaquettes. The prior \(\mu_{\text{prior}}\) is product Haar measure on bonds. An RG representation \(R\) on \(\Lambda_n\) has an inverse coupling \(\beta = \beta (R)\), an exponent \(\text{exponent}_R\) with constant part \(\text{scalar} (R)\), and an unnormalized density \(e^R = e^{\text{exponent}_R} \cdot (\text{polymer gas})\).

Proposition III.11.1 (determinant budget on the torus, and its per-site price). (a) Let \(R\) be a lawful representation on \(\Lambda_n\), \(n \geqslant 2\), with lawful-class parameters \((\sigma, A, \Lambda, \delta, w)\), and assume \(e^R \geqslant 0\) pointwise. Let \(0 < s \leqslant 1 \leqslant b\) be a Haar budget (Lemma 4.1). Let \(g > 0\) with \(g \leqslant \text{threshold} (R)\) and \(\beta g^2 = 1\). Let \(F\) be a \(\mu_{\text{prior}}\)-integrable complex function and \(z \geqslant 0\), \(X \in \mathbb{R}\) such that \(z |F (U) | \leqslant \exp (\text{exponent}_R (U) + cX)\) for every configuration \(U\). Then

\(\displaystyle \frac{z \hspace{0.17em} | \int F \hspace{0.17em} d \mu_{\text{prior}} |}{\int e^R \hspace{0.17em} d \mu_{\text{prior}}} \hspace{0.27em} \leqslant \hspace{0.27em} \exp \left( c \hspace{0.17em} \pi (R, X, g) - 4 \dim G \hspace{0.17em} \log g \right),\)

where the per-site price is

\(\displaystyle \pi = 2 \hspace{0.17em} \text{exCost} (A, \Lambda, \delta) + 6 + \beta \hspace{0.17em} \text{baseBudget} (A, g) + X + 3 \hspace{0.17em} \text{ent} - 4 \log s, \qquad \text{ent} = \log b - \dim G \hspace{0.17em} \log \sigma_R,\)

with \(\sigma_R = \min (1, \ell_R)\), where \(q_R\) and \(\ell_R\) are the quadratic and linear clip coefficients. Here \(\text{exCost} = \delta + \text{loopsPrice}\) bounds the loop and additive activity terms per site, and \(\text{baseBudget} = 3 (q_R + A^2) g^2\) bounds the remaining action on configurations whose bonds all lie in \(\text{shell} (g / 4)\). (b) Let \((\mathcal{A}_m)\) be the levels of an RG schedule for simple \(G\), and let \(O = (s, b)\) be a Haar budget. For all sufficiently large levels \(m\), every period \(p\), every representation \(P\) on the level-\(m\) torus (with \(c_{m, p}\) sites), every \(k \in \mathbb{N}\), every \(X \leqslant 1 + k\), and every \((g, \Lambda, \delta)\) admissible at level \(m\), the following holds:

\(\displaystyle \pi (P, X, g) - \frac{4 \dim G \hspace{0.17em} \log g}{c_{m, p}} \hspace{0.27em} \leqslant \hspace{0.27em} \text{work item} (O, m, k) .\)

Admissible at level \(m\) means: same clip as the stage of \(m\), \(\beta (P) \leqslant 4 \hat{\beta}_m\), \(\beta (P) g^2 = 1\), \(0 < g \leqslant \text{threshold} (P) \leqslant \text{threshold}_m\), \(\delta \leqslant 1\), \(\Lambda \leqslant \text{loopCap}_m\). The work item is

\(\displaystyle \text{work item} = 2 (8 \hspace{0.17em} \text{paths}_m + 2) + 6 + 3 (\text{direct}_m + \text{paths}_m^2) + (1 + k) + 3 \hspace{0.17em} \text{ent} - 4 \log s + 1.\)

In (b), \(\pi\) uses \(A = \text{paths}_m\) and the clip of stage \(m\). The work item depends on that stage, \(O\) and \(k\), uniformly over the admissible couplings and periods. Status: theorems. Lean: determinant, protection, HaarBudget; (b) affordable, Meets, work item.

Proposition III.11.2 (free pair terms). Fix \(G\), \(d, n \in \mathbb{N}\), and two colored plaquette derivative stencils \(L, S\) based at the origin, with scaling degrees \(\Delta_L = 2 + \text{ord} (L) \leqslant d\) and \(\Delta_S = 2 + \text{ord} (S) \leqslant d\). There is \(C \geqslant 0\), depending only on these data, such that the following holds. Let \(x \in (\mathbb{R}^4)^n\) have distinct entries and let \(i \ne j\). Suppose \(D \geqslant 0\) satisfies \(| \text{Re} \hspace{0.17em} \Gamma_{(i, a), (j, b)} | \leqslant D\) for every pair of free-current keys \(a, b\) of scaling degree at most \(d\). Then, for every \(r \geqslant 1\),

\(\displaystyle r \hspace{0.17em} | r^{\Delta_L + \Delta_S} \hspace{0.17em} \langle \text{ev}_L \otimes \text{ev}_S, B_r (i, j) \rangle - \text{light} (L, S) (x_i, x_j) | \leqslant CD.\)

Here

\(\displaystyle B_r (i, j) = \sum_{a, b} \text{Re} \hspace{0.17em} \Gamma_{(i, a), (j, b)} \hspace{0.27em} r^{- \text{gen} (a)} t_a \otimes r^{- \text{gen} (b)} t_b,\)

where \(t_a\) is the polynomial jet sample of key \(a\), \(\text{gen} (a)\) is its scaling degree, and \(\text{ev}_L\) applies the stencil \(L\). The kernel \(\text{light} (L, S)\) is the corresponding derivative of the free Maxwell curvature two-point function. Status: theorem. Lean: Prayer.fence, deck.

Proposition III.11.3 (window estimate for the free covariance, uniform in the regulator). Let \(\alpha, \eta > 0\), \(\Omega \in \mathbb{R}\) and \(n \in \mathbb{N}\). Then there is \(C \geqslant 0\) with the following property. For every regulator \(f\) in the family \(\text{Firm} (\alpha, \Omega, \eta)\) of III.3, every \(G\), and every pair of colored plaquette stencils \(a, b\), the frequency chart \(K = K_f (a, b)\) is measurable, vanishes for \(| \xi | > 2\), is \(C^n\) away from \(0\), and satisfies:

Here \(\partial^i\) denotes the \(i\)-th Fréchet derivative, measured in operator norm, and \(H_{a, b}\) is the degree-zero Maxwell curvature symbol for the plaquettes and colors of \(a, b\), with heat parameter \((2 \pi)^2\). If \(a, b\) have different colors, then \(K = H_{a, b} = 0\). Status: theorem. Lean: window_prices, Window, Cast.

Proposition III.11.4 (frequent good scales produce corrected trial systems). Let \(G\) be simple and \(C\) a Wilson family satisfying the construction record. Let \(d\) be a degree and \(\mathcal{F}\) a stage package for degree \(d\) (fixed band \(b_0 \geqslant \max (6, \hspace{0.17em} s_d + 3)\), with \(s_d\) the label's scaling saving, and Taylor order \(\geqslant d\)). Let \(X\) consist of a real basis of \(F_d^{\text{free}}\), starting with the unit, and finitely many generic one- and two-point tests that separate this space. Choose representatives of that basis in the label space. Let \(O\) be a Haar budget, \(J\) a chart ratio, and \(B\) a binding of the RG history of \(C\) to its price lists. Fix \(w \in \mathbb{R}\). Assume that at infinitely many levels \(k\) the following conditions hold simultaneously:

Then there is a trial system \(Z\) for \(X\) that travels with the station and satisfies the purchase conditions:

Status: theorem, with the frequent-scale conditions carried as hypotheses. In the main chain they are discharged as follows. In Accent.rated, the lantern marks hold eventually in \(k\) by bursts. The four price conditions are supplied to the rank assembly by franchise (this unit) inside the assembly jointOffers. That assembly is closed by uv_log (III.13), whose only hypotheses are that \(G\) is simple and \(b > 0\). Lean: cast_off, wheeled, v_culminates, v_obtain_purchases, Purchases.

Proposition III.11.5 (the field spaces of the limit are those of the free Maxwell field). Let \(G\) be simple. Write \(\mathcal{L}_d\) for abstract labels and \(\mathcal{V}_d\) for their fixed coordinate space. Let \(C\) be a Wilson family satisfying the construction record and the centred estimate: for each \(d\) and \(R, \varepsilon > 0\) there is \(A \geqslant 0\) such that, at every level \(m \geqslant \text{ref} (d)\), eventually in \(N\) and for all \(p\), every label of degree \(d\) and norm \(\leqslant R\) has centred budget \(A \hspace{0.17em} a_m^{- d - \varepsilon}\). Assume:

Then for every extraction order \(f\) there are a refinement \(g\), a smooth limit table \(T\) with a differentiation plan \(E\), and the extraction record of IV.1 along \(f \circ g\) (raw limits, bins, spans, cleaning to \(T\), Euclidean axioms), together with a seed record for \((C, f \circ g, T)\). The seed record consists of: cutoff, ordered and generic point limits, Euclidean axioms, cell hypothesis, and corrected rank widths. Moreover the following holds: there is a real structure on the free Maxwell table and, for every \(d\), a linear isomorphism

\(\displaystyle \Phi_d : \hspace{0.27em} F_d (T) \hspace{0.27em} \xrightarrow{\sim} \hspace{0.27em} F_d^{\text{free}} =\mathcal{L}_d /\mathcal{E}_d\)

that intertwines complex conjugation and the action of the signed permutations of the four axes. More precisely, there is a real, equivariant linear isomorphism \(L_d : \mathcal{L}_d \to \mathcal{V}_d\) carrying \(\mathcal{E}_d\) onto \(\text{null}_d (T)\). Thus \(\dim F_d (T) = \dim F_d^{\text{free}}\); the identification of null spaces uses this corrected map. Status: theorem with hypotheses (i)–(iii). All three are discharged in the main chain: (i) by reception through chaletrail (this unit), (iii) by franchise, and the binding by castle. The one-step package (ii) and the centred estimate are also included in Chateau and consumed by uv_log (III.13). Lean: Accent.rated, Maxwell.Ranks, Spec.Seed, Scotia, Wcmain.

Proposition III.11.6 (the stress tensor of the limit, normalized against free Maxwell). (a) Let \(C\) satisfy the construction record and \(f\) be an extraction order with ordered and generic point limits, giving a table \(S\). Assume:

Then \(W\) satisfies the orientation law, and there are a stress record \(I\) (stress tensor \(\mathcal{J}\), translation parts, spin rows) and a scaling \(X\) of \(S\) such that:

(b) Let \(G\) be simple, let \(C\) carry the shell estimate, and assume the cofinal grid approximation condition and weak completeness of the corrected comparison tests for \(f\), together with the seed record of III.11.5 for \((C, f, S)\). Then for every differentiation plan \(W\) with inclusion and derivative laws there are Maxwell data \(M\) and a stress record \(I\) with \(\mathcal{J} \ne 0\) satisfying the UV interface record \(\text{UV} (C, f, S, W, M, I)\). Status: theorems. In (b), the free model comes from maxwell_supplies (VI.2). The station hypothesis is built from the seed and weak completeness, and the hypothesis DimAbove from the seed's corrected rank widths; this is how III.11.5 feeds III.11.6. The shell budget is proved by laurelE and fugacious (this unit) inside uv_log. Lean: Acme.readyA, Acme.emanate, Aroma.

Proposition III.11.7 (Ward identity: the flux of the stress against a Killing field is a local field). Let \(C\) satisfy the construction record, let \(f\) be an extraction order with the cell hypothesis and generic point limits \(S\), and suppose \(S\) satisfies the Euclidean axioms with real structure \(H\). Let \(W\) satisfy the derivative laws, and assume the shell estimate of III.11.6. Let:

Then there is \(B \in F_D (S)\) with the following property. For every \(n\), all fields \(q_1, \ldots, q_n\), every configuration of distinct points \(x_0, \ldots, x_n\), and every window \(\chi\) adapted to \(x\) (smooth, compactly supported, equal to \(1\) near \(x_0\), with support avoiding \(x_1, \ldots, x_n\)), the flux of \(\mathcal{J}\) against \(U (\cdot - x_0)\) and \(\chi\), inserted with \(v\) at \(x_0\) and \(q_k\) at \(x_k\), equals the correlation of \(B\) at \(x_0\) with the \(q_k\) at \(x_k\). The sign convention is

\(\displaystyle \text{flux} = - \sum_{i, j} \int_{\mathbb{R}^4} \partial_i \chi (y) \hspace{0.17em} U_j (y - x_0) \hspace{0.17em} \langle \mathcal{J}_{ij} (y) \hspace{0.17em} v (x_0) \hspace{0.17em} q_1 (x_1) \cdots q_n (x_n) \rangle \hspace{0.17em} dy.\)

Nonzero homogeneous Killing fields have \(p = 0\) (translations, \(D = d + 1\)) or \(p = 1\) (rotations, \(D = d\)). Status: theorem. Lean: annular_fieldA, tenet, shells_law.

Further inputs.

2. What it uses

Inputs from other units.

Standard mathematics. The chessboard estimate for reflection-positive measures (Fröhlich–Israel–Lieb–Simon, Comm. Math. Phys. 62 (1978); Glimm–Jaffe, Quantum Physics, 2nd ed.), together with reflection positivity of the Wilson action (Osterwalder–Seiler, Ann. Phys. 110 (1978)). Axial (tree) gauge fixing on a torus and invariance of Haar measure (Seiler, Gauge Theories as a Problem of Constructive Quantum Field Theory and Statistical Mechanics, Lecture Notes in Physics 159 (1982)). Haar volume of small balls in a compact Lie group. Wick (Isserlis) expansion. Schur complements for Gaussian conditioning. Symbol estimates in frequency space. Finite-dimensional duality (a vector lies in the range of a linear map if and only if it is annihilated by every functional that annihilates the range). Ward identities for a conserved symmetric stress tensor and Killing fields of \(\mathbb{R}^4\).

Imported constants and assumptions.

3. Strategy

The field-space comparison has two parts. Algebraic equations give an upper bound on the dimension of each interacting field space. Short-distance one- and two-point tests give the reverse bound. A conjugation of projectors then identifies the real representations of the signed permutation group.

At each cutoff, the projector kernel is the image of \(\mathcal{E}_d\) under an invertible corrected label map. Free probes separate the on-shell sector, so \(\dim F_d^{\text{free}} = \dim \mathcal{V}_d - \dim \ker \Pi_{d, N, p}\). The relay estimates carry these equations to the limit, giving the upper bound. Along a sequence of good levels, the corrected test moments approach their free values. Since the free tests separate a basis, the corrected vectors are eventually linearly independent modulo the limit null space. This gives the lower bound.

The two bounds imply \(\text{null}_d (T) = \ker \Pi_{\infty}\) and \(\dim F_d (T) = \dim F_d^{\text{free}}\). An invertible operator close to the identity conjugates a cutoff projector to \(\Pi_{\infty}\). Composing it with the corrected label map identifies the Maxwell ideal with the interacting null space, preserving conjugation and signed permutations.

The quantitative engine is the chessboard estimate. The product of reflected copies of a connected part, integrated against the Wilson measure at cutoff \(N\), is carried by the exact RG history to a coarse level \(m\). There the number of cells equals the number of sites \(c_{m, p}\). The flowed observable is bounded pointwise by \(r^{- |s| q} \exp (\text{exponent} + qX)\) for a set \(s\) of marked modes, and Proposition III.11.1 bounds the resulting ratio of integrals by \(e^{c_{m, p} \pi} g^{- 4 \dim G}\).

The determinant estimate compares two Haar integrals:

The powers of \(g\) cancel up to \(g^{- 4 \dim G} = \beta^{2 \dim G}\), which does not depend on the volume. After taking the \(c_{m, p}\)-th root, its logarithmic contribution is at most \(1\), because \(2 \dim G \log \beta \leqslant 8 \dim G \hspace{0.17em} \hat{\beta}_m \leqslant a_m^{- 1} \leqslant c_{m, p}\) eventually. So the root test costs a per-site bound depending only on the stage parameters. The constants are chosen so that the allowance is \(e^{2 I_k}\) and the error bound is \(e^{- I_k}\), where \(I_k = \log \mu_k / 32 \geqslant \text{work item}\) for the one- and two-point tests; the root of every nonempty connected part is then \(\leqslant e^{I_k} / e^{2 I_k} = e^{- I_k} \to 0\).

All conditions are needed only at infinitely many good levels \(k\). A diagonal choice of depths then produces the trial sequence (III.11.4). The covariance splitting identifies the fluctuation law used in the one-step comparison. The regulator estimates supply the frequency windows, while the uniform train constants control their propagation; the tender condition includes the numerical clearance needed for both.

The stress tensor (III.11.6, III.11.7) is built by comparing the limit with a free Maxwell model along a scaling sequence. The Ward identities follow from a flux argument on shrinking annuli of radii \(r_l \to 0\):

4. Steps

4.1 Haar estimates and the determinant budget

Lemma 4.1 (Haar budget). There are \(0 < s \leqslant 1 \leqslant b\) such that \(h (\text{shell} (r / 4)) \geqslant s \hspace{0.17em} r^{\dim G}\) for \(0 < r \leqslant 1\), and \(\int e^{- \text{chord} (u) / t} \hspace{0.17em} dh (u) \leqslant b \hspace{0.17em} t^{\dim G}\) for \(t > 0\). Sketch: in exponential coordinates near \(1\), the chord is comparable to the Lie-algebra norm and the Haar density is bounded above and below. Compactness extends the upper volume bound \(h (\text{shell} (r)) \leqslant Cr^{\dim G}\) to all \(r > 0\). Majorize the exponential by \(\sum_{k \geqslant 0} e^{- k} 1_{\text{shell} ((k + 1) t)}\) and sum \(Ct^{\dim G} \sum_{k \geqslant 0} (k + 1)^{\dim G} e^{- k}\). Lean: integral_budget.

Lemma 4.2 (plaquette elimination). Let \(n \geqslant 2\) and let \(\varphi : G \to [0, 1]\) be continuous. Then \(\int \prod_p \varphi (U_p) \hspace{0.17em} d \mu_{\text{prior}} (U) \leqslant \left( \int \varphi \hspace{0.17em} dh \right)^{3 c - 3}\), the product running over all \(6 c\) plaquettes of \(\Lambda_n\). Sketch: fix the spanning tree in which each site \(x \ne 0\) is joined to \(x - e_{i (x)}\), where \(i (x)\) is the highest axis with \(x_i \ne 0\). The \(3 c + 1\) non-tree bonds are grouped into tiers by axis. Within each tier, take the wrap bonds first, except one "handle" per tier, then the transverse bonds from higher to lower height. This gives \(3 c - 3\) swept bonds \(e\), each assigned a plaquette \(p (e) \ni e\). Triangularity: if \(e \in p (e')\) with \(e' \ne e\), then \(e'\) has strictly higher priority than \(e\). Integrate the bond of highest priority first. By left and right invariance of Haar measure, \(U_{p (e)}\) is Haar-distributed as a function of \(U_e\), and no other selected plaquette contains \(e\). So \(\int \varphi (U_{p (e)}) \hspace{0.17em} \Psi \hspace{0.17em} d \mu = \left( \int \varphi \hspace{0.17em} dh \right) \int \Psi \hspace{0.17em} d \mu\). By induction, the product over the selected plaquettes integrates to exactly \((\int \varphi)^{3 c - 3}\); the other plaquette factors are bounded by \(1\). Lean: elimination, charged_integral, pull_link, independent. (proof read; the combinatorics of the sweep, i.e. the count and triangularity lemmas, statement only)

Lemma 4.3 (vacuum lower bound). Under the hypotheses of III.11.1(a), \(\int e^R \hspace{0.17em} d \mu_{\text{prior}} \geqslant \exp \left( \text{scalar} (R) - c (\text{exCost} + \beta \hspace{0.17em} \text{baseBudget}) \right) \hspace{0.17em} (s \hspace{0.17em} g^{\dim G})^{3 c + 1}\). Sketch: \(e^R\) is gauge invariant (gauge_eval), so its integral equals the integral over the \(3 c + 1\) free bonds with tree bonds set to \(1\) (free_count). Restrict to the set where every free bond lies in \(\text{shell} (g / 4)\). Every bond then has chord \(\leqslant g / 4\), so every plaquette has chord \(\leqslant g\). On calm configurations the polymer gas factor is \(1\), and the exponent is \(\geqslant \text{scalar} - c (\text{exCost} + \beta \hspace{0.17em} \text{baseBudget})\). Lemma 4.1 with \(r = g \leqslant 1\) bounds the Haar volume of the restriction set from below. Positivity of \(e^R\) handles the complement. Lean: vacuum_bound.

Lemma 4.4 (exponent bound and numerator). Under the hypotheses of III.11.1(a), with \(E = g / \sigma_R\): for every \(U\), \(\text{exponent}_R (U) \leqslant \text{scalar} + c (\text{exCost} + 6) - \sum_p \text{chord} (U_p) / E\). Consequently \(z | \int F| \leqslant \exp (\text{scalar} + c (\text{exCost} + 6 + X)) \hspace{0.17em} (b \hspace{0.17em} E^{\dim G})^{3 c - 3}\). Sketch: for the clipped plaquette action at \(\beta = g^{- 2}\), one has \(\text{chord} / E - 1 \leqslant \beta \cdot (\text{base term})\):

The principal part is an average over signed reflections of the pulled-back configuration. The chord sum is invariant under these, and the loop and activity corrections are bounded by \(c\) times their per-site prices. This gives the first bound, with \(6 =\) the number of plaquette orientations accounting for the \(- 1\) per plaquette. Then \(z |F| \leqslant e^{\text{scalar} + c (\ldots)} \prod_p e^{- \text{chord} (U_p) / E}\), and Lemma 4.2 with \(\varphi = e^{- \text{chord} / E}\) together with Lemma 4.1 conclude. Lean: exponent_retains, numerator.

Lemma 4.5 (the ratio; III.11.1(a)). Divide Lemma 4.4 by Lemma 4.3. With \(u = 3 c - 3\) and \(l = 3 c + 1 = u + 4\), the exponent is

\(\displaystyle c (2 \hspace{0.17em} \text{exCost} + 6 + \beta \hspace{0.17em} \text{baseBudget} + X) + u \hspace{0.17em} \text{ent} - l \log s - 4 \dim G \log g.\)

Since \(\text{ent} \geqslant 0\), \(\log s \leqslant 0\), \(u \leqslant 3 c\) and \(l \leqslant 4 c\), this is at most \(c \pi - 4 \dim G \log g\). Lean: determinant.

Lemma 4.6 (per-site work item; III.11.1(b)). Under the assumptions and uniformly over the parameter ranges of III.11.1(b), for all sufficiently large \(m\):

Sketch: (i) uses \(\text{exCost} \leqslant 1 + (8 \hspace{0.17em} \text{paths} + 1)\), from \(\delta \leqslant 1\), \(\Lambda t^2 \leqslant 1\) and \(\hat{\beta}_m t^3, \hat{\beta}_m t^4 \leqslant 1\) eventually; and \(\beta \hspace{0.17em} \text{baseBudget} = 6 (\text{direct} + \text{paths}^2) / 2\) by \(\beta g^2 = 1\). (ii) Since \(- 4 \log g = 2 \log \beta \leqslant 2 \beta \leqslant 8 \hat{\beta}_m\), and \(\hat{\beta}_m\) grows linearly in \(\log a_m^{- 1}\), one gets \(8 \dim G \hspace{0.17em} \hat{\beta}_m \leqslant a_m^{- 1} \leqslant c_{m, p}\) eventually. Lean: affordable, rolls_law.

Lemma 4.7 (chessboard root). Let \(T\) tile the cutoff torus into \(q\) cells, where \(q\) is the number of sites of the level-\(m\) torus. Suppose the exact RG history has terminal density \(P\). Assume that deleting the additive activities of \(P\) gives a lawful representation with activity parameter \(0\), and that the deleted additive term has absolute value at most \(q \delta\). Let \(V\) be a finite set of modes and let every component \(\Phi_s\) be continuous. Assume \(r > 0\), \(e \geqslant 0\), \(g > 0\), \(g \leqslant \text{threshold} (P)\) and \(\beta (P) g^2 = 1\). Suppose the local activity bound for the flowed stamped polynomial, at radius \(r\) and exponents \((e, l, L)\), gives

\(\displaystyle r^{q \hspace{0.17em} \text{card} \hspace{0.17em} s} |F_s (U) | \leqslant \exp \left( \text{exponent}_P (U) + q (L + \text{card} \hspace{0.17em} V \hspace{0.17em} l) \right)\)

for every terminal configuration \(U\) and every \(s \subseteq V\). Here \(F_s\) is the flow of the unnormalized Wilson density multiplied by the product of the reflected copies of \(\Phi_s\). Then

\(\displaystyle \rho_T (\Phi_s) \leqslant \frac{\exp \left( \pi (P, L + \text{card} \hspace{0.17em} V \hspace{0.17em} l, g) - 4 \dim G \log g / q \right)}{r^{\text{card} \hspace{0.17em} s}},\)

where \(\rho_T\) is the absolute value of the stamped Wilson expectation to the power \(1 / q\).

For the station at level \(k\), assume allocation, per-site affordability at the width level, \(\text{card} \hspace{0.17em} V \leqslant M\), \(l, L \leqslant 1\), and \(r = e^{2 I}\) with \(I \geqslant \text{work item} (O, k, M)\). Then, for every nonempty \(s\), \(\rho_T (\Phi_s) \leqslant e^{- I}\). Sketch: the exact vacuum identity supplies the Wilson partition function as denominator. Apply III.11.1(a) to the representation with its additive activities deleted. Comparing both numerator and denominator costs \(e^{q \delta}\) twice, supplying the \(2 \delta\) already included in \(2 \hspace{0.17em} \text{exCost}\) in \(\pi\). Take the \(q\)-th root. For the last assertion, allocation keeps the stage constant between width and \(k\); \(L + \text{card} \hspace{0.17em} V \hspace{0.17em} l \leqslant 1 + M\), and \(r^{\text{card} \hspace{0.17em} s} \geqslant r\). Lean: s_postage, denture, coupon. (proof read; the flow identity and the cell-stamp bookkeeping, statement only)

4.2 Free covariances

Lemma 4.8 (III.11.2). Sketch: expand the pair term as \(\sum_{a, b} \text{Re} \hspace{0.17em} \Gamma_{ab} \hspace{0.17em} (r^{\Delta_L} \text{ev}_L (r^{- \text{gen} \hspace{0.17em} a} t_a)) (r^{\Delta_S} \text{ev}_S (r^{- \text{gen} \hspace{0.17em} b} t_b))\). Keys of generation below the scaling degree of the stencil evaluate to \(0\). Keys of generation equal to the scaling degree give exactly the free curvature kernel (deck). Every other term carries an extra factor \(r^{- 1}\) or smaller. So \(r\) times the remainder is at most \(D \sum | \text{ev}_L (t_a) || \text{ev}_S (t_b) |\), summed over the non-leading pairs. Lean: fence. Its eventual form is the pair penalty Prayer.penalty.

Lemma 4.9 (measured Gaussian data). Let \(Q\) be the symmetric covariance array and \(m\) the mean vector obtained by applying the normalized stencil and remainder functionals to measured pair and one-point data. Let \(F\) be the free-current covariance on the stencil keys, extended by zero on the remainder indices. Suppose \(B \geqslant 1\), \(T, P, \varepsilon \geqslant 0\), and, entrywise,

\(\displaystyle |m_i | \leqslant PT, \qquad |Q_{ij} - F_{ij} | \leqslant PT^2 .\)

On the stencil keys assume the sharper bounds \(|m_i | \leqslant \varepsilon T\) and \(|Q_{ij} - F_{ij} | \leqslant \varepsilon T^2\). If

\(\displaystyle PT \leqslant B, \qquad |F_{ij} | + PT^2 \leqslant B, \qquad \varepsilon T \leqslant e, \qquad \varepsilon T^2 \leqslant e,\)

then all measured means and covariances are bounded by \(B\), and their restriction to the stencil keys differs from \((0, F)\) by at most \(e\) in both means and covariances. Sketch: use the triangle inequality for \(Q\). In the application, the measured horizons and bridges satisfy the discrete Bianchi identities and scaled finite-difference bounds. The stencil normalization \(T\) bounds the sampling coefficients and the remainder factor \(M^{d - 1}\) for addresses within radius \(M\). These give the displayed broad and sharp bounds. Positivity of \(Q\) is not required. Lean: resolution.

Lemma 4.10 (III.11.3). Sketch: write \(\lambda = 2 \pi\) and let \(T (\zeta)\) be exterior multiplication by \(i \zeta\) from one-forms to two-forms. In the chart \(\zeta = \lambda \xi\), the heat symbol is

\(\displaystyle e^{- | \zeta |^2} | \zeta |^{- 2} (T (\zeta) T (\zeta)^{\ast})_{a, b},\)

with the color Kronecker factor. The regulator estimates in III.3 bound the chart as a symbol of order \(0\) and its difference from this heat symbol as a symbol of order \(1\). The chain rule under \(\xi \mapsto \lambda \xi\) preserves these orders and changes the constants by fixed powers of \(\lambda\). Sum the constants over the finitely many plaquette pairs to obtain a bound independent of \(G\) and the colors. Lean: window_prices, heat_window.

Lemma 4.11 (covariance splitting and propagation). (i) Let an atlas be compatible with the chosen schedule and its retained unmasked quadratic forms, and assume the atlas symmetry, coercivity, inverse and crossing laws. Let \(i < N\), choose a tree forest for the step, and assume the step torus satisfies the size condition for tree inversion. Then

\(\displaystyle \mathcal{C}_{i + 1} - W_i \mathcal{C}_i W_i^{\textsf{T}} = d_f \Gamma_i d_f^{\textsf{T}}, \qquad \Gamma_i \succeq 0,\)

where \(\mathcal{C}_i\) is the complete curvature covariance, \(W_i\) the passage map, \(d_f\) the fine exterior derivative, and \(\Gamma_i\) the uncut tree covariance, extended by zero on tree bonds. (ii) Fix kernel choices and stage indices \(j, t \geqslant 4\). There is \(R \geqslant 1\) such that, for every \(G\), every schedule with these kernel choices, and every \(N, i, p, d\) satisfying \(J_{i + 1} = j\), \(J_i = t\) and \(d - 1 \leqslant j - 2\) (natural-number subtraction), the difference-motion propagation bound holds with constant \(R\). Sketch: (i) is a Schur-complement identity on the curl image; the restricted tree inverse gives positivity. For (ii), use the uniform resolvent estimate for the two stage indices and sum its six plaquette components.

The table prepared chooses, for each kernel, the uniform window, covariance and propagation constants; their required smallness is a separate clearance condition. Lean: one_fold, train_issue.

Lemma 4.12 (Wick expansion). Let \(A\) be a finite index set, let \(\text{loc} : A \to S\) take values in an ordered location set, and let \(Z_a\) be formal Gaussian generators with zero mean and symmetric covariance \(Q\). Assume \(Q (Z_a, Z_b) = 0\) whenever \(\text{loc} (a) = \text{loc} (b)\). For every \(p \subseteq A\), the formal Wick moment of \(\prod_{a \in p} Z_a\) is the sum, over perfect matchings of \(p\) joining different locations, of the products of their covariances. No positivity assumption is imposed on \(Q\). Sketch: remove one index and use the Gaussian moment recursion. The zero mean removes the singleton term, and the covariance hypothesis removes pairs at a common location. The remaining choices are in bijection with the admissible matchings. Lean: headed.

4.3 Polynomial calculus and identities

Lemma 4.13 (polynomial and integral calculus). Write \(\|P\|_1\) for the sum of the absolute values of the coefficients of a polynomial \(P\).

Sketch: Taylor's theorem gives (i), and absolute coefficient summability gives (ii). For (iii), split each factor as \(1 + (f_i - 1)\) and induct on the derivative order; the highest derivative terms add rather than multiply. For (iv), the two-point cumulant is the second moment minus the product of the means; expand the difference and use the three stated bounds. For (v), apply the allocation identity before discarding sectors with the wrong flags, then group the local components. For (vi), the joint local-coordinate marginal is a product measure, so Fubini gives factorization. (proof read; local comparison and allocation inputs statement only)

4.4 From frequent good scales to trial systems (III.11.4)

Lemma 4.14 (multi-step propagation). Fix a level \(k \leqslant N\), a band \(b\), a fitted configuration \(x\) with \(v\) insertions, and an exact RG history. Let \(m_0\) and \(m_1\) be its width and middle levels. Assume allocation at \(k\), admissible representations along \([m_0, m_1]\), a numerical propagation bound for arity \(M\), and

\(\displaystyle \hat{\beta}_u \in [\hat{\beta}_k / 2, \hat{\beta}_k] \quad (m_0 \leqslant u \leqslant m_1), \qquad v \leqslant M.\)

Assume the instance decomposition for the tested labels: at the middle level, the marked sources approximate the negative free-profile terms within the prescribed error, with the required background and source bounds, and the subsequent steps satisfy the exhibited bounds. Finally assume each free-profile term has absolute value at most \(\exp ((\log \log \hat{\beta}_k)^2 / 4)\).

Then the flow of the corrected stamped observable has a local activity decomposition at \(m_0\) with radius \(\text{allowance} (M, k)\) and exponents \((e, 1, 1)\), where \(e\) is the input source exponent. Sketch: use the middle-level decomposition and the free-profile bound to initialize the source estimate. The propagation bound controls its iteration through \([m_0, m_1]\); the coupling window keeps every intermediate estimate in the same numerical range. The correction cancels the leading free-profile terms. This is the local decomposition used in Lemma 4.7. Lean: v_conditioned. (proof read; helper estimates statement only)

Lemma 4.15 (the chain). Under the hypotheses of III.11.4:

  1. At infinitely many \(k\), the simultaneous price and insertion conditions give covariance data, an exact RG history and compatible restarted representations, eventually in \(N\) and for every \(p\). The uniform covariance and test-point estimates supply the local Gaussian comparison. (cast_off via backstage.)

  2. Construct a common correction map and polynomial approximations for all rows. The one-step comparison supplies the coefficient, support and remainder bounds on the same path of representations. (v_paid_factory, proof read; analytic inputs statement only.)

  3. Compare the resulting Gaussian polynomial moments with the free profiles. The eventual finite-difference bounds transfer this comparison from the chosen representatives to the actual corrected columns, giving the instance estimates of Lemma 4.14. (v_culminates, proof read.)

  4. Apply Lemma 4.14 to propagate the corrected estimates to the width level, retaining the terminal density identity and admissibility needed for the root estimate. (v_pay_all, proof read.)

  5. Offers give a trial system. Choose depths \(k_i\) inductively among the good levels, with \(k_{i + 1} > k_i\), strictly increasing widths, and strictly decreasing test scales tending to \(0\). The lifts come from ticketed (bounded, real, with first entry the reference unit). The chessboard bounds come from Lemma 4.7, using the per-site affordability banking, which follows from Lemma 4.6. (v_obtain_purchases, proof read.)

Lean: wheeled composes items 2–5.

Lemma 4.16 (eventual insertion estimates). Assume \(G\) is simple, \(C\) satisfies the construction record, and the one-step cost package uses the prescribed weight sequence and multiplicities at least \(1\). Fix a degree-\(d\) column family \(X\) and stage package whose mark exponent satisfies \(c \geqslant 2 (w (G, d) + 1)\), where \(w (G, d) = w_d\) is the prescribed coefficient exponent. Then, for all sufficiently large \(k\), metering at \(k\) implies the lantern-mark condition for \(X\) at \(k\). Sketch: write \(x = \hat{\beta}_k\) and \(b_a (x) = \exp (a (\log \log x)^2)\). For every fixed \(a\), the inequality \(b_a (x) s \leqslant b_{- 8} (x)\) with \(s \geqslant 0\) gives

\(\displaystyle s \leqslant \exp (- (8 + a) (\log \log x)^2) \leqslant x\)

for sufficiently large \(x\). Hence \(x^{- c / 2} s \leqslant x^{- w (G, d)}\). The one-step estimate and spacing of the rounded test points turn this coefficient bound into the required insertion estimate. The threshold is uniform over the tested coefficients, cutoffs and periods. Lean: bursts, quaking. (proof read; analytic inputs statement only)

Lemma 4.17 (discharge of the frequent-scale conditions). Given compatible pre-\(L\) price data and one-step bounds for a sheet, with the stage thresholds below the chosen block lengths and each degree index eventually present in the stage index sets, a Haar budget, nonnegative mark exponents and simple \(G\), there are panels \(Q_d\) and a waiting window extending any given one with the following property. For every Wilson family \(C\) paid on that window and every binding with these prices, for every \(d\), the four price conditions hold at infinitely many levels. Sketch: the restarted comparison and the scale-selection construction produce the panels and the window; frequency follows because each requested degree index is eventually present in those stage sets. Lean: franchise. (proof structure read; helpers statement only)

4.5 The rank comparison (III.11.5)

Lemma 4.18 (equation count). Under the projector presentation of III.11.5(i), at every cutoff, \(\dim \mathcal{L}_d \leqslant \dim \ker \Pi_{d, N, p} + \dim F_d^{\text{free}}\). Sketch: \(\ker \Pi_{d, N, p}\) is the image of \(\mathcal{E}_d\) under an injective corrected label map, by the projector presentation. The free null space is the image of \(\mathcal{E}_d\) under the label coordinates. This uses projectionExact (a label is null in the free table if and only if its on-shell image lies in the residual) and no_residual (the residual is \(0\): free probes separate the on-shell sector). Then \(\dim F_d^{\text{free}} + \dim \mathcal{E}_d = \dim \mathcal{L}_d\). Lean: savings, EquationCount, projectionExact, no_residual. (proof read; free positivity input statement only)

Lemma 4.19 (short-distance comparison and independence). Assume \(C\) satisfies the construction record, \(f\) gives generic point limits \(S\), and the correlation kernels of \(S\) are continuous on generic configurations and integrable against every separated test function. Let \(Z\) satisfy the purchase conditions, and suppose all its coefficient vectors have iterated limits along \(f\). Then:

Sketch: expand the ordinary moment using the relative connected components and apply the chessboard inequality to every nonempty component. This gives the error \(P_k \gamma_k\) before the cutoff limits, and the bound passes through those limits. Each row functional descends to the quotient by the integrated-test null space. The resulting finite matrix converges to the injective free detector matrix. Injectivity is open in finite dimensions, giving (ii). Lean: measured; the comparison is Purchases.samples/converge in Bought.lean. (proof read; chessboard input statement only)

Lemma 4.20 (seed and null space). Assume \(C\) satisfies the construction record. Under a centred port for \(C\) along \(f\), raw limits \(S\) along that same extraction, relay bins with convergent projectors, the equation count for every \(d\), and trial systems with the purchase conditions for every \(d\), there is a refinement along which the following hold:

Sketch: the relay estimates give \(\ker \Pi_{\infty} \subseteq \text{null}_d (T)\). Lemma 4.19(ii) gives the reverse dimension inequality, using the equation count and eventual stability of the projector kernel dimension. Hence the subspaces coincide, and the field dimensions agree. Lean: arriving_seed, equation_retains. (proof read; analytic inputs statement only)

Lemma 4.21 (equivariant identification). Under the projector presentation of III.11.5(i) and the extraction record for its projectors and cleaned table \(T\), there are the real, signed-permutation-equivariant isomorphisms \(\Phi_d\) of III.11.5. The free table has its canonical real structure and zero residual null space. Sketch: let \(A_i \to \Pi_{\infty}\) be the projectors along the extraction, and set \(Y_i = \Pi_{\infty} A_i + (1 - \Pi_{\infty}) (1 - A_i) \to 1\). For large \(i\), \(Y_i\) is invertible and \(Y_i A_i = \Pi_{\infty} Y_i\). Since \(A_i\) and \(\Pi_{\infty}\) are real and equivariant, so is \(Y_i\). Then \(L = Y_i \circ (\text{corrected label map})\) maps \(\mathcal{E}_d\) onto \(\ker \Pi_{\infty} = \text{null}_d (T)\) (drumroll). The inverse of the induced quotient map, followed by the fixed free coordinate identification, is \(\Phi_d\). Lean: thrush, drumroll.

Lemma 4.22 (III.11.5). Under the hypotheses of III.11.5, its conclusion follows by the following steps:

  1. extract raw point limits (IV.1) and include the label projections;

  2. build, for each \(d\), the free detector columns \(X_d\), which form a basis of \(F_d^{\text{free}}\);

  3. combine assumption (iii) with Lemma 4.16 and apply III.11.4 to get trial systems with the purchase conditions;

  4. apply Lemma 4.20 with Lemma 4.18;

  5. assemble the extraction record (using D05);

  6. conclude with Lemma 4.21.

Lean: rated.

4.6 The stress tensor (III.11.6, III.11.7)

Lemma 4.23 (annihilated flux combinations). Under the hypotheses of III.11.7, let \(I\) be a finite family of generic tester configurations on the schedule grids, each with \(x_0 = 0\), and let \((g_a)\) be coefficients with \(\sum_a g_a \hspace{0.17em} \text{test}_a (z) = 0\) for all \(z \in F_D\). Here \(\text{test}_a\) inserts \(z\) at the root with the remaining fields of tester \(a\); \(\text{flux}_a\) inserts the flux built from \((\mathcal{J}, U, v)\) there. Then \(\sum_a g_a \hspace{0.17em} \text{flux}_a = 0\). Sketch: let \(r_l \to 0\) be the radii of the shell budget. Choose a smooth radial bump \(\chi\), equal to \(1\) on the ball of radius \(1 / 2\) and supported in the unit ball. Conservation makes the flux independent of the window, so for large \(l\) use \(\chi_l (y) = \chi (y / r_l)\). By homogeneity of \(U\), the flux becomes \(- \sum_{ij} \int M_{ij} (y) \hspace{0.17em} r_l^{3 + p} \langle \mathcal{J}_{ij} (r_l y) \cdots \rangle \hspace{0.17em} dy\), where \(M_{ij} = \partial_i \chi \cdot U_j\) is integrable and supported in \(\tfrac{1}{2} \leqslant |y| \leqslant 1\). The annulus bound zone says that, for annihilating combinations, \(\sup_{|y| \in (r_l / 4, 2 r_l)} | \sum_a g_a \hspace{0.17em} r_l^{3 + p} \langle \mathcal{J}(y) \cdots \rangle_a | \to 0\). So the combination is a constant sequence tending to \(0\). Lean: tenet, zone. (proof read; annulus comparison inputs statement only)

Lemma 4.24 (III.11.7). Sketch:

Lean: annular_fieldA, obtainable.

Lemma 4.25 (stress laws). Under the hypotheses of III.11.7, there are linear translation maps \(F_d (S) \to F_{d + 1} (S)\) and rotation maps \(F_d (S) \to F_d (S)\) satisfying the respective flux identities. The rotation maps also depend real-linearly on the skew matrix. Sketch: apply Lemma 4.24 with constant fields \(U = e_i\) (\(p = 0\), \(D = d + 1\)) and with the rotation fields \(U (y) = ky\), \(k\) skew (\(p = 1\), \(D = d\)). Linearity follows from linearity of the flux and separation of fields by correlations. Lean: shells_law.

Lemma 4.26 (III.11.6). Under the respective hypotheses of III.11.6(a) and (b), the stated stress and ultraviolet records exist. Sketch of (a):

  1. D06 gives the orientation law.

  2. The station equivalences identify the field spaces of \(S\) with those of the free model; this uses the hypothesis DimAbove.

  3. produce_current (VI.2) gives \((\mathcal{J}, X)\) with \(\mathcal{J} \ne 0\), the approach, stability, the meter and zero current divergence.

  4. linear_cons converts zero current divergence to conservation. The weak comparison along the scaling sequence supplies the action law.

  5. Lemma 4.25 gives the stress laws.

Sketch of (b): take free Maxwell data for simple \(G\) and replace them by the canonical data (FreeStress.fire). Build the station hypothesis from the seed and weak completeness, and the hypothesis DimAbove from the corrected rank widths. Then apply (a). Lean: readyA, emanate. (proof read; helper estimates statement only)

5. Where the difficulty lies

6. Notation

Written proof Lean (glossary) Meaning
schedule factor \(\mu_k\) monitor \(\exp ((\log \log \hat{\beta}_k)^2)\)
coefficient exponent \(w_d\) winter degree of the input coefficient family plus \(4\)
Haar budget \((s, b)\) HaarBudget (small, big) lower constant for shells, upper constant for \(\int e^{- \text{chord} / t}\)
\(\sigma_R\), \(\text{ent}\) strength, ent \(\min (1, \text{linear clip})\); \(\log b - \dim G \log \sigma\)
per-site price \(\pi\), work item protection, work item III.11.1
admissible at level \(m\) Meets conditions on \((P, g, \Lambda, \delta)\) in III.11.1(b)
chessboard root \(\rho_T\) Cells.Tiling.rootTest \(\lvert E \prod_{\text{cells}} \text{copy} \rvert^{1 / \text{card cells}}\)
free curvature kernel light derivative of the free Maxwell curvature two-point kernel between colored stencils
scale-\(r\) pair term collect, arc, arbor, Beamer.amount \(\sum \Gamma_{ab} \hspace{0.17em} r^{- \text{gen}} t_a \otimes r^{- \text{gen}} t_b\); \(r^{2 + \text{ord}}\)
Fourier chart rate circleRate \(2 \pi\); heat parameter \((2 \pi)^2\)
window estimate Window, Cast symbol bounds of III.11.3
stage package, metering Cohesion, Metering per-degree stage data; charge and launch at \(k\)
price conditions Tender, Metering, ShowHours, Vents, LanternMarks hypotheses of III.11.4 at a level \(k\)
provisions, aggregate, turning, offers s_provisions, s_empire, Turning_, Offers_ successive frequent-scale conditions of Lemma 4.15
trial system, purchase conditions Columns.Trials, Purchases, Travels_ III.11.4
compatible quadratic atlas Binding shape, coercivity, size and path compatibility on the same Wilson family
launch conditions Launch local comparison and propagation conditions at a specified level
projector presentation \(\Pi\) Scotia projectors whose kernels are corrected images of \(\mathcal{E}_d\)
one-step cost package Scrim step costs along gauge doors
Maxwell equation ideal \(\mathcal{E}_d\), \(F_d^{\text{free}}\) Maxwell.equationIdeal, Maxwell.freeTable labels vanishing on the free field; free field spaces
Maxwell ranks Maxwell.Ranks equivariant isomorphisms \(F_d (T) \cong F_d^{\text{free}}\)
numerical upper dimension bound DimAbove \(\dim F_d (S) \leqslant \dim F_d^{\text{free}}\)
equation count EquationCount Lemma 4.18
seed record Spec.Seed cutoff, limits, axioms, cell hypothesis, corrected rank widths
shell budget Aroma, Economics shell estimates, open tabs, radii \(r_l \to 0\)
flux, window flux J χ U x q v, Window (VI.1) flux of the stress against a vector field
arbitrarily fine comparison systems Stations for every order, size and positive tolerances, an admissible station at an arbitrarily small scale
cofinal grid approximation Grill approximation systems at cofinally many grid levels
weak completeness CompleteWeak comparison systems for every fixed arity, degree and positive size bound
free lighthouse Lighthouse, MaxwellData free Maxwell comparison model

7. What remains unexplained

  1. Analytic inputs to the annulus estimate. The reduction in zone and the flux argument are checked. The discrete stencil bound, interpolation estimate and shell estimates invoked there have not been independently checked through their analytic dependencies.

  2. Detailed one-step interfaces. The principal-data construction and the proof assembling the aggregate comparison were checked at their application points. The underlying nonlinear integration and remainder estimates were not checked in full. The relation between the older conditional interfaces in LedgerGate.lean and every restarted replacement remains to be documented; their shared names do not identify their hypotheses.

  3. Record definitions. The stage, launch, hours, vents, relay, station and shell hypotheses in the numbered statements refer to the linked records. Their complete translation into ordinary mathematical estimates remains to be written. Lemmas 4.11, 4.13(iv)–(vi), 4.14 and 4.17 retain these structural hypotheses; the displayed inequalities alone are not their full assumptions.

  4. Auxiliary results and upstream estimates. The additional estimates listed after III.11.7 were checked at statement level, including their parameter order. The sweep triangularity in Lemma 4.2 was checked at statement level. The proofs of the lower-level producers of the centred estimate and one-step package, and of the schedule-existence theorem in III.10, were not checked in full. The source links identify these dependencies; the present outline does not independently establish them.

← III.10 One full nonlinear renormalization step: bounds on irrelevant terms (notes)Part III·Unit outlineIII.12 Bounds uniform in the volume and convergence as the spacing tends to zero (notes) →