Home·The theorem·The proof in one reading·Architecture·Units·Manuscript·Glossary·FAQ·Work items·Data·Method·Provenance·Changes
← III.13 Alternative construction: one Wilson family for all labels (notes)Part III·Unit outlineIII.15 Cost accounting for observables and the remainder comparison (notes) →

III.14. The construction run: flat representations and the ultraviolet rate — detailed notes

Contents

1. What this unit proves 2

2. What it uses 6

3. Strategy 8

4. Steps 8

4.1 Flat representations and their propagation 8

4.2 Choice of widths and passage to position space 10

4.3 Uniqueness of the Gaussian part; the three recipe components 11

4.4 Realizing the recipe on a run 12

4.5 The run 13

4.6 The packages behind the numerical floors 13

5. Where the difficulty lies 14

6. Notation 15

7. What remains unexplained 16

The full account of unit III.14: 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:

Size: 14,290 lines in the main-theorem closure (2,509 declarations).

The unit selects the construction run and proves a quantitative limit theorem for its Gaussian quadratic part.

Use in the main theorem. VIII.1 (continuum_with_position_tests) uses the run, its penalty, the pillars and the funded property. It discards the attached-recipe conjunct.

The selection theorem of III.2 accepts any nonempty family of demands. Here the demand is a recipe, with lower bounds on the block factors and numerical requests on the running coupling. The Gaussian limit is therefore an additional property of the selected run; VIII.1 does not invoke it separately. Replacing this selection by one with fewer demands is a separate question (§7).

1. What this unit proves

Conventions. We use the level systems and sufficiency condition of III.1. Throughout, \(G\) is a compact matrix gauge group; “simple” means that its real Lie algebra is simple and nonabelian.

Proposition III.14.1 (the construction run). Let \(G\) be simple and \(b > 0\). There exist a run \(R\), a campaign \(T\) for \(R\) and pillars \(W\) for \((R, T)\) such that:

  1. the penalty of the block-spin kernel of \(R\) equals \(b\);

  2. \(R\) satisfies the full funded condition of III.15 relative to the local-estimate package \(\Delta_{10}\). Its thirteen clauses (§6) include the prescribed local estimates, \(Q (i) \to \infty\), support exponent \(100\) at every stage, and the extraction, OPE and thermal conditions. The selection uses \(Q (i) = i + 2\);

  3. the native component recipe is attached to \(R\). There are a sheet \(s\), a price map \(q\), a recipe \(C\) for \(s\) (Proposition III.14.2) and floors \(n \geqslant \max (n_s, E_C)\), where \(n_s\) is the sheet's floor and \(E_C\) the combined recipe floor of Lemma 4.14, such that:

Status: theorem. A run is fixed by choice as fundedRun.

Lean: YM.Independent.constructionRun; R2I.Funded; garment10; Campaign.Pillars; ActualRunNativeComponentRecipe.Attached.

Proposition III.14.2 (existence of the native component recipe). Let \(G\) be simple and let \(s\) be a sheet (the kernel choice and stage profile, fixed before any block factor). There exist a price map \(X\) (the secondary price map) and floors \(E^{\flat}, E^{\text{sw}}, E^{\text{du}} : \mathbb{N} \to \mathbb{N}\) with the following properties, for every level system \(\mathcal{L}\) with the kernel and stage profile of \(s\).

Status: theorem.

Lean: NativeComponentRecipe.Choice, available; the three components follow from Flat.entrance_gate, Cast.assured (with Bid.Sworn) and Aura.dual_support (with Aura.licensed).

Proposition III.14.3 (ultraviolet rate for the Gaussian quadratic part). Let \(G\) be simple and \(s\) a sheet. There exist a price map \(X\) and a floor \(E : \mathbb{N} \to \mathbb{N}\) with the following property. Let \(\mathcal{L}\) be a level system with the kernel and stage profile of \(s\) and with \(L_i \geqslant E (J_{i + 1})\) for all \(i \geqslant 0\). Let \(e \in \mathbb{N}\) satisfy \(\Omega^X_{J_i} \leqslant \log \hat{\beta}_i\) for \(i \geqslant e\). Then there are kernels \(F_i\) (\(i \geqslant e\)) on the links of \(\mathbb{Z}^4\) and constants \(A, B \geqslant 0\), \(c > 0\), \(d, l \in \mathbb{N}\) such that the following holds for every horizon \(H\) of \(\mathcal{L}\) sufficient for some price map, every reporting scale \(k \in \mathbb{N}\) and every \(i \geqslant e\). Write \(w_j = \min (1, \alpha^X_j, \alpha^X_{j + 1})\).

One may take \(c = 1 / 4\), \(d = l = 1\), \(A = 34 \sigma_0 |\mathcal{F}| (1 + s_G) / (1 - 3^{- 1 / 8})\) and \(B = \sigma_0 |\mathcal{F}| (1 + s_G)\), where \(\sigma_0 = |\mathcal{F}|^2 \lambda_d^2\) and \(\lambda_d\) is a width-one strip bound for the lattice exterior derivative. These constants depend only on \(G\) and the lattice, independently of the sheet, level system, entry stage and horizon.

Status: theorem. This is an order-zero estimate (see Mesh.lean).

Lean: Flat.Quorum (the three clauses), Flat.entrance_gate (existence), Flat.supplies, Flat.meshed, Flat.clock, Flat.curve, Flat.Delay.

Proposition III.14.4 (flat representations at all stages, abstract form). Let \(P\) be a price map, \(\theta > 0\), and \(w : \mathbb{N} \to (0, \infty)\). Let \((\mathcal{L}_x, H_x, r_x, N_x, n_x, k_x, Z_x)_{x \in X}\) be a family of level systems with horizons, integers \(r_x \leqslant N_x \leqslant n_x\), reporting scales \(k_x \in \mathbb{N}\) and offsets \(Z_x \geqslant 0\). Put \(\Lambda_x (i) = 2 + Z_x + |s_i - s_{k_x} |\), with the scale data understood to belong to \(\mathcal{L}_x\). Fix \(A_0 \geqslant 0\) and \(d_0 \in \mathbb{N}\), independent of \(x\). Assume:

Then there are \(A \geqslant 0\), \(c > 0\), \(d \in \mathbb{N}\) and kernels \(K_{x, i}\) such that, for all \(x\) and \(r_x \leqslant i \leqslant N_x\), \(K_{x, i}\) is a flat representation (Definition 4.1) of \(\mathcal{K}_{n_x, i} -\mathcal{K}_{N_x, i}\) at stage \(i\), of width \(w_{J_i}\) and cost \(A \hspace{0.17em} \Lambda_x (i)^d (a_{N_x} / a_i)^c\). One may take \(c = \min (1, \theta / 2)\) and \(d = d_0\), with \(A\) depending only on \(A_0, d_0, \theta\).

Status: theorem.

Lean: Flat.World.overlap, with World, World.Funding, World.Provides.

Proposition III.14.5 (Schur identity for two cutoffs, and the flat step).

Identity. Fix one RG step: a fine torus \(\mathbb{T}_f\) with \(L\) times the side of the coarse torus \(\mathbb{T}_c\), the block average \(Q\) and penalty \(b > 0\). Say that \((W ; M, M')\) satisfies the background-field relations when

\(\displaystyle \mathcal{K} (M) \hspace{0.17em} W = Q^{\top} \mathcal{K} (M'), \qquad QW = 1 - b^{- 1} \mathcal{K} (M') .\)

If \(M\) is symmetric and positive semidefinite, these relations make \(WA'\) a minimizer of \(\langle A, \mathcal{K}(M) A \rangle + b |A' - QA|^2\), with minimum \(\langle A', \mathcal{K}(M') A' \rangle\). For the algebraic identity below, positivity is unnecessary: if \((W_1 ; M_1, M_1')\) and \((W_2 ; M_2, M_2')\) satisfy the relations and \(M_2, M_2'\) are symmetric, then

\(\displaystyle \mathcal{K} (M_2') -\mathcal{K} (M_1') = W_2^{\top} (\mathcal{K}(M_2) -\mathcal{K}(M_1)) W_1 .\)

Flat step. Let \(\tau > 0\), \(r \geqslant 0\), and let \(B\) be a neutral, translation-invariant plaquette kernel with \(\lVert B \rVert^{\text{pl}}_{\tau} \leqslant r\), whose periodization represents \(\mathcal{K} (M_2) -\mathcal{K} (M_1)\) on \(\mathbb{T}_f\). Suppose \(T, U\) are order-two strip-bounded curvature-transport kernels with constants \(q, p \geqslant 0\) in width \(\omega\). Assume \(T (x + Lv, y + v) = T (x, y)\) and the same identity for \(U\), that their periodizations \(C_2, C_1\) satisfy \(dW_2 = C_2 d\), \(dW_1 = C_1 d\), and that \(\sum_z U ((x, \varphi), (z, \psi))\) is independent of \(x\) for each pair of orientations. If \(0 \leqslant \sigma \leqslant \omega\) and \(2 \sigma \leqslant L \tau\), then

\(\displaystyle B' (y, z) = \sum_{x, w} T (x, y) B (x, w) U (w, z), \qquad \lVert B' \rVert^{\text{pl}}_{\sigma} \leqslant \frac{2 |\mathcal{F}|^2 qp}{\tau L} r.\)

The kernel \(B'\) is neutral and translation invariant, and its periodization represents \(\mathcal{K} (M_2') -\mathcal{K} (M_1')\) on \(\mathbb{T}_c\).

Sketch: transpose the second pair of background-field relations and substitute the first pair; the mixed \(b^{- 1} \mathcal{K} (M_2') \mathcal{K} (M_1')\) terms cancel. The intertwining identities then give the asserted curvature representative. Neutrality permits subtraction of the second transport kernel at the first base point, yielding the norm estimate in Lemma 4.3.

Status: theorem. No inverse of the curvature form is used. Preservation of neutrality requires the separate constant-row-sum condition.

Lean: Flat.poised (identity), Flat.toward (step), Packets.push_null, neutralCoin.

Proposition III.14.6 (rate lemma for fine-to-coarse recursions). For each \(x \in X\), fix a scale system with \(L_i \geqslant 3\), stages \(r_x \leqslant i \leqslant t_x\) and age \(\Lambda_x (i) = 2 + Z_x + |s_i - s_{k_x} |\), where \(Z_x \geqslant 0\). Fix \(\theta, c_0 > 0\), \(A_0 \geqslant 0\) and \(d_0 \in \mathbb{N}\), uniformly in \(x\). Suppose \(f_x (i), g_x (i) \geqslant 0\) on these stage ranges and

Then \(f_x (i) \leqslant A \Lambda_x (i)^{d_0} (a_{t_x} / a_i)^{\mu}\) throughout each range, where \(\mu = \min (c_0, \theta / 2) > 0\). One may take \(A = A_0 (K + 1)^{d_0} / (1 - 3^{- \theta' / 2})\), with \(\theta' = \theta - \mu > 0\) and \(K = 2 (d_0 + 1) / \theta'\).

Status: theorem.

Lean: Rate.Stair.fade_solve, poly_solve, slope, Fade.cap.

Further content. The auxiliary estimates in §4.6 supply numerical requirements for the recipe. They include the one-step transit estimate harbor, curvature-coefficient lifts arc and warehouse, and formal Gaussian identities bottle, still and absorbed.

2. What it uses

From other units.

Standard mathematics.

Imported constants and assumptions.

3. Strategy

The endpoint applies the selection theorem of III.2 to the demand for a recipe. First choose the sheet, primary price map, recipe and combined floor \(E_C\). Next choose the block factors \(\ell_j\) above that floor. Finally raise the coupling thresholds to pay the bill \(B_C (\ell, j)\) and the countable numerical requests. The selection retains the resulting inequalities on the run itself, so Lemma 4.15 realizes the recipe on that run's levels and horizon.

For the Gaussian estimate, compare two cutoffs \(N \leqslant n\) at a common stage \(i\). At stage \(N\), their metric difference \(M^{\infty}_{n, N} - M^{\infty}_{N, N}\) represents the quadratic-form difference and has zero row sums in each orientation, since both Fourier symbols equal the identity at zero momentum. The mixed Schur identity transports this representative through the two background-field maps. A transport kernel has row scale \(L^{- 2}\), column scale \(L^2\) and first-difference scale \(L^{- 3}\). Neutrality permits subtraction in one leg, gaining a factor \(L^{- 1}\); the distance introduced by subtraction is absorbed using half of the input exponential weight. The block-factor floor makes the one-step factor at most \(L_i^{- 1 / 2}\). A discrete Grönwall estimate absorbs polynomial growth in the age and gives the rate \((a_N / a_i)^{1 / 4}\). Passing from curvature representatives to link kernels and then taking a Cauchy limit gives Proposition III.14.3.

The argument carries representatives modulo kernels annihilated by \(d^{\top} (\cdot) d\), without inverting the degenerate curvature form. The plaquette metrics do have inverses, used in the transport kernels. The Gaussian quadratic forms are independent of the sufficient horizon, because the terminal form and the Schur recursion determine them. This permits the estimate to be proved on a shadow level system with altered clip parameters, where III.3 supplies canonical free transfer data, and then transferred to the original horizon.

4. Steps

4.1 Flat representations and their propagation

Definition 4.1 (flat representation). Let \(H\) be a horizon of \(\mathcal{L}\), \(N \leqslant n\), \(i \leqslant N\), and \(\sigma, p \geqslant 0\). A kernel \(K\) on plaquettes \(\mathbb{Z}^4 \times \mathcal{F}\) is a flat representation of \(\mathcal{K}_{n, i} -\mathcal{K}_{N, i}\) at stage \(i\), of width \(\sigma\) and cost \(p\), if:

Lean: Flat.FlatRep, Sigil, Folded, neutral, town.

Lemma 4.2 (the finest stage). Let \(H\) be sufficient for \(P\), \(N \leqslant n\), and \(0 < \sigma \leqslant \alpha_{J_N}\). Then \(M^{\infty}_{n, N} - M^{\infty}_{N, N}\) is a flat representation at stage \(N\) of width \(\sigma\) and cost \(2 |\mathcal{F}| \Omega_{J_N}\).

Sketch: each metric has plaquette norm \(\leqslant |\mathcal{F}| \Omega_{J_N}\) by its strip bound. The difference is neutral because \(\hat{M} (0) = 1\) for both metrics. The torus identity is \(\mathcal{K} (M_{n, N}) -\mathcal{K} (M_{N, N}) = d^{\top} (M_{n, N} - M_{N, N}) d\).

Lean: finest-stage lemma (private, World.lean), via turn_class; cost topTab.

Lemma 4.3 (one step to a coarser stage). Let \(H\) be sufficient for \(P\), with \(i < N \leqslant n\). Assume the plaquette-transfer kernel of \(H\) at \((N, i)\) is the canonical free kernel. Let \(K\) be a flat representation at stage \(i + 1\) of width \(u > 0\) and cost \(p \geqslant 0\), and let \(0 < \sigma \leqslant \min (1, \alpha_{J_i}, \alpha_{J_{i + 1}})\) with \(2 \sigma \leqslant L_i u\). Then

\(\displaystyle K' (y, z) = \sum_{x, w} T_{n, i} (x, y) \hspace{0.17em} K (x, w) \hspace{0.17em} T_{N, i} (w, z)\)

is a flat representation at stage \(i\) of width \(\sigma\) and cost \(q_i (u) \hspace{0.17em} p\), where:

Sketch: the four properties come from four inputs.

Lean: one-step lemma (private, World.lean); toward, poised, push_null, quote.

Lemma 4.4 (induction over stages). Under the hypotheses of Proposition III.14.4 there are flat representations \(K_{x, i}\) (\(r_x \leqslant i \leqslant N_x\)) of width \(w_{J_i}\) and cost \(c_x (i) = 2 |\mathcal{F}| \Omega_{J_{N_x}} (a_{N_x} / a_i)^{\theta}\).

Sketch: the costs are defined explicitly, so that \(c_x (i) = L_i^{- \theta} c_x (i + 1)\). The proof is downward induction on \(N_x - i\):

The representatives for all \((x, i)\) are then fixed by choice.

Lean: downward-induction lemma (private, World.lean).

Lemma 4.5 (the rate; proof of Proposition III.14.4). Write \(g_x (i) = 2 |\mathcal{F}| \Omega_{J_{N_x}} \boldsymbol{1}_{i = N_x}\). The costs of Lemma 4.4 satisfy the recursion of Proposition III.14.6 with forcing \(g\). By the last hypothesis of III.14.4, \(g\) has the required form with \(c_0 = 1\) (it vanishes below the top). Proposition III.14.6 then gives cost \(\leqslant A \Lambda_x (i)^{d_0} (a_{N_x} / a_i)^{\min (1, \theta / 2)}\).

Proof of III.14.6.

  1. Multiply by \((a_i / a_t)^{\mu}\); the recursion keeps its form with \(\theta' = \theta - \mu \geqslant \theta / 2\).

  2. Prove \(\tilde{f}_x (i) \leqslant \tilde{B} \hspace{0.17em} (K + \Lambda_x (i))^{d_0}\) by downward induction, with \(\tilde{B} = A_0 / (1 - 3^{- \theta' / 2})\).

  3. The step uses the polynomial estimate: since \(\Lambda_x (i + 1) \leqslant \Lambda_x (i) + \log L_i\), one has \((K + \Lambda_x (i + 1))^d L_i^{- \theta'} \leqslant (K + \Lambda_x (i))^d L_i^{- \theta' / 2}\) when \(d \leqslant K \theta' / 2\). Then \(L_i^{- \theta' / 2} \leqslant 3^{- \theta' / 2}\), and \(\tilde{B} 3^{- \theta' / 2} + A_0 = \tilde{B}\).

Finally, \(\Lambda_x (i) \geqslant 2\) gives \((K + \Lambda_x (i))^{d_0} \leqslant (K + 1)^{d_0} \Lambda_x (i)^{d_0}\). Undoing the tilt gives the stated value of \(A\).

Lean: World.overlap, fade_solve, poly_solve, slope, Fade.of_top, drop_length.

4.2 Choice of widths and passage to position space

Lemma 4.6 (choice of widths). Let \(G\) be simple, \(X\) a price map and \(w_j = \min (1, \alpha^X_j, \alpha^X_{j + 1})\). Put

\(\displaystyle \beta_{j, j'} = 2 |\mathcal{F}|^2 h_{j, j'}^2 / w_j, \qquad h_{j, j'} = \exp (4) |\mathcal{F}|^2 (\Omega^X_j)^2 \Omega^X_{j'}, \qquad E^{\flat} (j) = \lceil 2 / w_j + \beta_{j, j}^2 + \beta_{j, j - 1}^2 \rceil .\)

Here \(j - 1\) is truncated at zero. Suppose \(L_i \geqslant E^{\flat} (J_{i + 1})\) and \(\Omega^X_{J_i} \leqslant \log \hat{\beta}_i\) for every \(i \geqslant e\). With \(q_i\) computed from \(X\), the following hold for \(i \geqslant e\):

Sketch: since \(J_i \in \{J_{i + 1}, J_{i + 1} - 1\}\), the constant \(h_i\) is one of \(h_{j, j}, h_{j, j - 1}\) with \(j = J_{i + 1}\). Then \(q_i (w_j) = \beta / L_i \leqslant L_i^{- 1 / 2}\) because \(\beta^2 \leqslant L_i\). Next, \(2 w_{J_i} \leqslant 2 \leqslant L_i w_j\). For the age cap: \(\hat{\beta}\) is affine in \(s\), so \(\log \hat{\beta}_j \leqslant \log \hat{\beta}_k + s_G |s_j - s_k |\).

Lean: spacing_choice, universeClock, work items, entrance, before, Spacing.

Lemma 4.7 (the family estimate). Assume the conclusions of Lemma 4.6 and let \(H\) be sufficient for \(X\). Index the family by \(x = (k, N, n)\) with \(k \in \mathbb{N}\) and \(e \leqslant N \leqslant n\), and put \(r_x = e\), \(\theta = 1 / 2\) and \(Z_x = 3 + \log \hat{\beta}_k\). Then the numerical hypotheses of Proposition III.14.4 hold with \(A_0 = 2 |\mathcal{F}| (1 + s_G)\) and \(d_0 = 1\). If the plaquette-transfer data at every \((N, i)\) with \(e \leqslant i < N\) are the canonical free data, all its hypotheses hold.

Sketch: the first three width estimates are exactly the required one-step inequalities. The age bound gives \(2 |\mathcal{F}| \Omega_{J_N} \leqslant A_0 \Lambda_{k, N}\) uniformly in \(k, N, n\).

Lean: berth, Spacing.funded.

Lemma 4.8 (from a representative to the position-space kernel). Let \(0 \leqslant \sigma, \varepsilon \leqslant 1\) and \(p, C \geqslant 0\). Suppose both metric kernels are strip-bounded in width \(\sigma\) with constant \(p\), and that \(K\) represents \(\mathcal{K}_{n, i} -\mathcal{K}_{N, i}\) on tori whose sides tend to infinity, with \(\lVert K \rVert^{\text{pl}}_{\varepsilon} \leqslant C\). Then

\(\displaystyle \mathcal{O}_{n, i} -\mathcal{O}_{N, i} = d^{\top} K \hspace{0.17em} d \quad \text{on the links of } \mathbb{Z}^4, \qquad \lVert \mathcal{O}_{n, i} -\mathcal{O}_{N, i} \rVert_{\varepsilon} \leqslant \sigma_0 C, \qquad \sigma_0 = |\mathcal{F}|^2 \lambda_d^2,\)

where \(\lambda_d\) is a width-one strip bound for \(d\).

Sketch: \(\mathcal{O}_{n, i} -\mathcal{O}_{N, i}\) and \(d^{\top} Kd\) have finite row norms and the same periodizations on the given tori. Dominated convergence as the torus side tends to infinity identifies the kernels. Composition with the two exterior derivatives gives the factor \(\sigma_0\).

Lean: resolve, output_cost, sumToken.

Lemma 4.9 (rate for pairs of cutoffs). Under the conclusions of Lemma 4.6, with \(H\) sufficient for \(X\) and with canonical free plaquette-transfer data at all \((N, i)\), \(e \leqslant i < N\), the constant \(A\) of Proposition III.14.3 satisfies

\(\displaystyle \lVert \mathcal{O}_{n, i} -\mathcal{O}_{N, i} \rVert_{w_{J_i}} \leqslant A \hspace{0.17em} \Lambda_{k, i} \hspace{0.17em} (a_N / a_i)^{1 / 4} \qquad \text{for all } k \text{ and } e \leqslant i \leqslant N \leqslant n.\)

Sketch: apply Proposition III.14.4 to the family of Lemma 4.7, then Lemma 4.8.

Lean: meshed.

Lemma 4.10 (the limit kernel). Under the same hypotheses there are kernels \(F_i\) satisfying (a)–(c) of Proposition III.14.3, in the same width \(w_{J_i}\).

Sketch: for fixed \(i\), Lemma 4.9 gives a Cauchy modulus \(A \Lambda_{k, i} (a_n / a_i)^{1 / 4} \to 0\) in the weighted row norm. Completeness of the scalar entries gives a pointwise limit, and finite-sum bounds followed by monotone convergence preserve the row estimates (solvent). This gives norm convergence and (c). For \(n \geqslant i\), \(\lVert \mathcal{O}_{n, i} \rVert_{w_{J_i}} \leqslant \sigma_0 |\mathcal{F}| \Omega_{J_i} \leqslant \sigma_0 |\mathcal{F}| (1 + s_G) \Lambda_{k, i}\), which passes to the limit and gives (a). Uniqueness of pointwise limits makes \(F_i\) independent of \(k\).

Lean: supplies.

4.3 Uniqueness of the Gaussian part; the three recipe components

Lemma 4.11 (the Gaussian quadratic forms do not depend on the horizon). Let \(H, H'\) be horizons, each sufficient for some price map, on level systems \(\mathcal{L}, \mathcal{L}'\) that differ at most in the clip parameters of their stages; for example \(\mathcal{L}'\) may be the shadow of \(\mathcal{L}\), with clip profile replaced by \(\gamma\). Then \(\mathcal{K}^{\mathbb{T}}_{N, i} (H) =\mathcal{K}^{\mathbb{T}}_{N, i} (H')\) for all \(i \leqslant N\) and all tori, and \(\mathcal{O}^H_{N, i} =\mathcal{O}^{H'}_{N, i}\).

Sketch: downward induction from \(i = N\), where both forms equal \(d^{\top} d\). If the forms agree at stage \(i + 1\), the Schur identity gives \(\mathcal{K}_{N, i} (H) -\mathcal{K}_{N, i} (H') = W^{\top} \cdot 0 \cdot W' = 0\); the clip parameters do not enter the step data. This proves equality on the tori of the level system. Lemma 4.8 then identifies the position-space kernels; periodizing that equality gives equality on every torus.

Lean: same_path, same_output, same_terminal, shadow, same_shadow.

Lemma 4.12 (the flat component; proof of Proposition III.14.3). For a simple \(G\) and a sheet \(s\), take the clip profile \(\gamma\) and floor \(E_{\text{free}}\) from III.3, and obtain \(X\) by applying its sufficiency construction to the sheet with clip profile \(\gamma\). Set \(E = \max (E^{\flat}, E_{\text{free}})\). Given \(\mathcal{L}\) and \(e\) as in Proposition III.14.3, the shadow \(\mathcal{L}'\) meets the assumptions of III.3, so it carries a horizon \(Y\) sufficient for \(X\) with canonical free transfer data. Apply Lemmas 4.6–4.10 to \((\mathcal{L}', Y)\); the delay condition is unchanged, since \(\hat{\beta}\) and the scale data are unchanged. Transfer (b) and (c) to every sufficient horizon \(H\) of \(\mathcal{L}\) by Lemma 4.11. The floor \(E\) is required at every stage because III.3 requires its floor everywhere.

Lean: entrance_gate, rearrange.

Lemma 4.13 (the other two components of the recipe). For each sheet, the point-group and background-field conclusions of Proposition III.14.2 hold above suitable floors. Both use the shadow construction of Lemma 4.12. Their floors may be chosen to be \(E_{\text{free}}\); the flat component also requires \(E^{\flat}\).

Lean: Cast.assured, Aura.dual_support, available.

4.4 Realizing the recipe on a run

Lemma 4.14 (realization). Let \(s\) be a sheet, \(q\) a price map, \(C\) a recipe for \(s\) with secondary price map \(X\), and \(\mathcal{L}\) a level system with the kernel and profile of \(s\). Let \(H\) be sufficient for the primary price map \(P_{s, q}\). Let \(E_C\) be the maximum of \(E^{\flat}, E^{\text{sw}}, E^{\text{du}}\) and the two numerical floors in §4.6, and let \(B_C (\ell, j)\) be the maximum of the two post-length bills there. Assume

Then on \((\mathcal{L}, H)\):

Sketch: the floor bounds apply to \(L_i = \ell_{J_{i + 1}}\) and to all smoothing-index requirements. For the delay, the second bill is the exponential of a finite bound \(S_i\) for the actual patch, with \(\Omega^X_q \leqslant S_i\) for every \(q \leqslant J_{i + 1}\). Its payment gives \(S_i \leqslant \log \hat{\beta}_i\) (Yamen.hatch). Since \(J_i \leqslant J_{i + 1}\), the delay holds with \(e = 0\). Apply the three components of \(C\); the fixed chart of III.3 supplies the remaining local-chart condition.

Lean: realize, Realized, floor, post, numeric.

Lemma 4.15 (binding to the run). Let \(s, q, C\) be as in Lemma 4.14 and \(n : \mathbb{N} \to \mathbb{N}\) with \(n \geqslant E_C\). Let \(R\) be a run whose sheet is \(s\) with floors raised to \(n\), whose price map is \(P_{s, q}\), and whose levels pay \(B_C\). Then the recipe is realized on \(R\)'s own levels and horizon.

Sketch: the sheet and price-map identities identify the kernel, profile and sufficient horizon required in Lemma 4.14. The run's block-factor inequalities give \(\ell_j \geqslant n (j) \geqslant E_C (j)\). Apply that lemma. The retained selection data of III.2 contain exactly these identities and inequalities.

Lean: on_run, of_preselection.

4.5 The run

Lemma 4.16 (selection with a preselected demand; III.2). Fix the comparison and custody data used in III.2. Let \(G\) be simple, \(b > 0\), \(\Lambda\) a property of level systems that can be enforced by raising coupling-entry thresholds, and \(\Delta\) a local-estimate package. For every sheet \(s\) and price map \(q\), suppose a nonempty demand set \(\mathcal{D} (s, q)\) is given. For each \(C \in \mathcal{D} (s, q)\), prescribe an integer floor \(E_C (j)\), real requests \(U_C (r, j)\), and real post-length requests \(V_C (\ell, r, j)\) depending on the block-factor sequence \(\ell\).

Then there are a run \(R\), a campaign \(T\), a sheet \(s\), a price map \(q\), a choice \(C \in \mathcal{D} (s, q)\) and an integer floor \(n \geqslant \max (n_s, E_C)\) such that:

The comparison and custody data have canonical choices depending only on \(G\), used in Lemma 4.17. The additional conclusions about orders, periods and extraction in III.2 are not needed here.

Lean: selectRunWithTreasuryPeriod_preselected_quote_orders_postnumeric; the plain version is selectRun. (statement only; the construction behind it belongs to III.2)

Lemma 4.17 (assembly of the construction run). Proposition III.14.1 holds.

Sketch: use the canonical comparison and custody data and apply Lemma 4.16 with \(\Lambda\) identically true, \(\Delta = \Delta_{10}\), and \(\mathcal{D} (s, q)\) the set of recipes for \(s\). Proposition III.14.2 makes this set nonempty. Use floor \(E_C\), reciprocal-width requests \(U_C (r, j) = \rho_C (r, j)^{- 1}\), where \(\rho_C (r, j) > 0\) is the recipe width for demand \(r\) at smoothing index \(j\), and \(V_C (\ell, r, j) = B_C (\ell, j)\), independent of \(r\). Lemma 4.15 realizes the selected recipe on the run.

Lean: constructionRun, postnumeric.

4.6 The packages behind the numerical floors

The combined recipe floor includes rail, and its post-length bill uses receipt/after. The second numerical component uses Yamen.need and Yamen.stamp.

The supporting estimates below supply finite widths, numerical floors and post-length bounds. The widths also enter the reciprocal-width requests of Lemma 4.17. These are summaries under the hypotheses of the linked results; the complete auxiliary statements are not reproduced here.

5. Where the difficulty lies

6. Notation

Written proof Lean Meaning
run \(R\), campaign \(T\), pillars \(W\) Run, Campaign, Pillars Wilson family with its scale and extraction data; campaign comparison data; selected histories from stage \(0\) for all \((N, p)\), with local step data and accounting identities
funded R2I.Funded 13 clauses: localBudget, attentive, logCopy, locrian, wild, silk, poly, fixed, tic, halo, paced, stoves, hasSky
\(\Delta_{10}\) garment10, Decal local one-step estimate package with its eventual-validity lemma
sheet \(s\) Sheet kernel choice, stage profile, floors; fixed before \(L\)
price map \(P = (\alpha_j, \Omega_j, \eta_j)\) MapPrices strip width, upper bound and coercivity bound for each smoothing index
horizon \(H\), sufficient for \(P\) Horizon, Suffices metrics \(M_{N, i}\) and transfer data, with strip bounds and Schur relations
\(\mathcal{K}^{\mathbb{T}}_{N, i} = d^{\top} M^{\mathbb{T}}_{N, i} d\) Precision.K, quadratic Gaussian quadratic form on a torus
\(\mathcal{O}_{N, i}\) output, writ position-space kernel \(d^{\top} M^{\infty} d\) on links
background-field relations Inputs \(\mathcal{K}W = Q^{\top} \mathcal{K}'\), \(QW = 1 - b^{- 1} \mathcal{K}'\)
curvature transport \(T_{m, i}\) fine_share, response \((M^{\infty}_{m, i + 1})^{- 1} \mathcal{T}_{m, i} M^{\infty}_{m, i}\)
\(h_i\), \(q_i (u)\) header, quote, neutralCoin \(\exp (4) \lvert \mathcal{F} \rvert^2 \Omega_{J_{i + 1}}^2 \Omega_{J_i}\); \(2 \lvert \mathcal{F} \rvert^2 h_i^2 / (uL_i)\)
flat representation FlatRep, Sigil Definition 4.1
weighted norms Folded, RowCost, Atlas plaquette norm \(\lVert K \rVert^{\text{pl}}_{\sigma}\); link row norm \(\lVert F \rVert_{\sigma}\); row, column and first-difference bounds of order \(m\)
neutral, translation invariant neutral, town zero symbol at zero momentum; \(K (x + u, y + u) = K (x, y)\)
\(w_j\), \(\Lambda_{k, i}\) curve, clock, Lambda' \(\min (1, \alpha_j, \alpha_{j + 1})\); \(5 + \log \hat{\beta}_k + \lvert s_i - s_k \rvert\)
delay condition Delay \(\Omega_{J_i} \leqslant \log \hat{\beta}_i\) for \(i \geqslant e\)
flat rate (III.14.3) Quorum clauses (a)–(c)
family and numerical assumptions World, Funding, Spacing hypotheses of III.14.4; their form on a common level system
stairs, drop, weight Stair, drop, Fade index family; \(L_i^{- \theta}\); age polynomial times \((a_t / a_i)^c\)
shadow shadow same level system with clip profile replaced
recipe \(C\), floor \(E_C\), bill \(B_C\) Choice, floor, post three components with pre-length floors; combined floor \(E_C\) and post-length bill \(B_C\)
point-group invariance ("sworn") Bid.Sworn colour-summed bilinear form invariant under signed axis permutations
background-field identities ("licensed") Aura.licensed free background map satisfies the background-field identities for the actual forms
"prices", "bill", "floor", "numeric", "post" — numerical bounds on block factors and the reference coupling

7. What remains unexplained

  1. Selection in III.2. Lemma 4.16 is used as an input; the underlying construction of the scale and extraction data is not audited here. Its two auxiliary inputs have canonical choices. Removing the recipe demand would give a separate existence statement, not an identification of the resulting run with the fixed run of Proposition III.14.1.

  2. The funded condition. Its thirteen clauses are retained in full, rather than replaced by their summary in Proposition III.14.1. The route through fromFunded uses the clauses for extraction, comparison, OPE and asymptotic control; their detailed mathematical meanings and proofs belong to III.15 and later units. In particular, that route does not project the local-estimate clause involving \(\Delta_{10}\).

  3. Auxiliary estimates in §4.6. Only their statements and numerical uses are summarized. Their full admissibility conditions and proofs, including the curvature-lift, marked-kernel and formal Ward estimates, remain outside this verification. The summaries are not substitutes for those hypotheses.

  4. Inputs from III.3 and III.8. The free-kernel existence theorem and the underlying lattice kernel estimates are used at their stated interfaces. The proofs of those upstream results are not audited here.

← III.13 Alternative construction: one Wilson family for all labels (notes)Part III·Unit outlineIII.15 Cost accounting for observables and the remainder comparison (notes) →