1. What this unit proves 3
2. What it uses 9
3. Strategy 11
4. Steps 11
4.1 Rates and sequences 11
4.2 Prescription and Reader 12
4.3 Norms in the ground algebra 13
4.4 The lattice cluster estimate 14
4.5 The OPE at grid configurations 16
4.6 Arbitrary configurations and exponents 17
4.7 The Reader and the separated OPE 18
4.8 Integrated and nested forms 19
4.9 The slab bound of V.1 20
4.10 Contact extensions 21
5. Where the difficulty lies 24
6. Notation 25
7. What remains unexplained 26
The full account of unit VII.4: 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: VII, identification as Yang–Mills. Lean: YM/OPE. Its folders, by role:
prescription and Reader: Splitting, Orbits;
bounds in the ground algebra: Eye, Cards, Exchange;
lattice cluster estimate: Sweep, Replicas, Grouping, Columns, Prices;
passage to arbitrary configurations: Schemes, Interpolate, Height, Rate;
trees: Nesting, Trees, Single, Subpower.
Also YM/Next/PHASE, YM/Contacts (Normal, Score, Layers, Harvest, Solstice, Transport), YM/MultipleContact and YM/SharpGram. The public statements are publicOPE and contactDistribution.
Size: about 10,200 lines in 2,035 declarations assigned to this unit:
OPE about 4,600, mostly in Grouping, Replicas, Columns, Trees, Sweep;
Next/PHASE 840;
Contacts about 3,400;
MultipleContact about 290;
SharpGram about 250.
Conventions.
\(G\) is the compact gauge group, with simple Lie algebra, and \(\mathcal{W}\) is the Wilson family of the construction. Its lattice levels \(k = 0, 1, 2, \ldots\) have spacings \(a_k \downarrow 0\). Larger \(k\) means finer. The sequence is gentle: \(\log (a_k / a_{k + 1}) / \log (1 / a_k) \to 0\).
\(I_{N, p} (v, x)\) is the renormalized lattice observable of a lattice label (token) \(v\) at the site \(x\). \(\|F\|_{\mathcal{T}_k}\) is the chessboard root of a lattice function \(F\) in the centred tiling of scale \(a_k\). Both follow IV.1.
\(S\) is the limit table along the refined extraction order. It has field spaces \(F_d\) and the inclusions \(\iota_{m, n} : F_m \to F_n\) (\(m \leqslant n\)) of the field algebra. A label list is \(a = (a_i)_{i \in I}\) with \(a_i \in F_{d_i}\), and its total degree is \(w (a) = \sum_i d_i\).
\(\mathcal{A}\) is the ground algebra of V.1: the commutative C*-algebra generated by the compressed slab operators \(M (u)\) of separated words \(u\). Its vacuum is \(\Omega\), its Gelfand spectrum is the phase space \(X\) with measure \(\nu\) of full support, and its characters are \(\chi_{\xi}\).
\(S^{\xi} (u) = \chi_{\xi} (M (u))\) and \(S^{\text{mix}} (u) = \langle \Omega, M (u) \Omega \rangle\).
Hence, for \(A \in \mathcal{A}\) and \(c \geqslant 0\): \(\|A\| \leqslant c\) if and only if \(|s (A) | \leqslant c\) for every state \(s \in X \sqcup \{ \text{mix} \}\) (glance_contract, Cards.pressure).
Clusters and clearance follow Definition 1.13 of the statement.
\((x, o)\) is an \((r, M, \eta)\)-cluster if \(|x_i - o| \leqslant Mr\) and \(|x_i - x_{i'} | \geqslant \eta r\) for \(i \ne i'\).
\(y \in (\mathbb{R}^4)^J\) is \(c\)-clear of \(o\) if \(|o - y_j | \geqslant c\) and \(|y_j - y_{j'} | \geqslant c\) for \(j \ne j'\).
\(S^s (a, x ; b, y)\) and \(M (a, x ; b, y)\) denote the concatenated word.
"For all sufficiently small \(r\)" means: there is \(r_0 > 0\), depending only on the data quantified before the phrase, such that the assertion holds for all \(r \in (0, r_0)\) and all configurations and states quantified after it.
Proposition VII.4.1 (separated OPE; properties (O1)–(O2) of the main theorem).
Setting.
\((\mathcal{W}, S, F)\) is a family, limit table and final record satisfying the RG hypotheses of Remark VII.4.7, denoted (RG).
\(\mathcal{P}\) is any coherent prescription (VII.4.6).
\(\mathcal{R}\) is the Reader built from \(\mathcal{P}\) and the rotation and reality structures of \(F\) (Lemma VII.4.c). It provides strata \(E_j \subset F_j\), bases fixed by conjugation \((O_{j, \kappa})_{\kappa \in \text{Ch}_j}\) of \(E_j\), and coefficients \(c_{j, \kappa} (a ; x, o)\).
Conclusions.
(O1) Fix a finite \(I\), labels \(a\), \(M \geqslant 0\), \(\eta > 0\), \(\varepsilon > 0\), \(j\) and \(\kappa \in \text{Ch}_j\). For all sufficiently small \(r\) and every \((r, M, \eta)\)-cluster \((x, o)\): \(|c_{j, \kappa} (a ; x, o) | \leqslant r^{\hspace{0.17em} j - w (a) - \varepsilon}\).
(O2) Fix finite \(I, J\), labels \(a, b\), \(M \geqslant 0\), \(\eta, c, \varepsilon > 0\) and \(D \in \mathbb{N}\). For all sufficiently small \(r\), every \((r, M, \eta)\)-cluster \((x, o)\) and every \(y\) that is \(c\)-clear of \(o\):
the points of \((x, y)\) are pairwise distinct, and so are those of \((o, y)\);
the remainder satisfies
hence \(| S^s (a, x ; b, y) - \sum_{j \leqslant D} \sum_{\kappa} c_{j, \kappa} (a ; x, o) \hspace{0.17em} S^s ((j, O_{j, \kappa}), o ; b, y) | \leqslant r^{D + 1 - w (a) - \varepsilon}\) for every \(s \in X \sqcup \{ \text{mix} \}\).
The coefficients do not depend on \(b\), \(y\), \(s\) or \(D\). The thresholds do not depend on \(x\), \(o\), \(y\) or \(s\).
Status: theorem for every Reader on the given final record under (RG). The main construction supplies the coherent prescription and (RG) in VIII.1.
Lean:
Orbits.Foundation: clause coefficients is (O1); clause discrepancy gives injectivity, the bound in \(\mathcal{A}\) and the bound in every state.
Orbits.produce, coefficient_separated, quantitative_separated.
The error element is algebraError.
Headline: SeparatedOPE, clauses coefficient and remainder, the latter via Independent.Core.point.
Public form: publicOPE into Hierarchy.OperatorProductExpansion.
Proposition VII.4.2 (integrated remainder; item (D8)). Fix the data of (O2), a measurable space \(K\) and \(B \geqslant 0\). For all sufficiently small \(r\) the following holds.
Hypotheses. Let \(\mu\) be a positive measure on \(K\), \(w \in L^1 (\mu ; \mathbb{C})\) with \(\int |w| \hspace{0.17em} d \mu \leqslant B\), and \(k \mapsto (x (k), o (k), y (k))\) an a.e.-measurable configuration map. Assume that \(\mu\)-a.e. on \(\{w \ne 0\}\), \((x, o)\) is an \((r, M, \eta)\)-cluster and \(y\) is \(c\)-clear of \(o\).
Conclusion. Take for \(s (k)\) either the mixed state, or \(\xi (k)\) for an arbitrary a.e.-measurable phase map \(\xi : K \to X\). Then:
\(w \hspace{0.17em} S^s (a, x ; b, y)\), \(w\) times the truncated sum, and \(w\) times their difference are integrable;
the integrals satisfy
In particular, for every probability measure \(\nu'\) on \(X\), the \(\nu'\)-mixture of the phases satisfies the bound of (O2) (probability_mixtures). The same pointwise threshold works for every \(K\) and \(B\).
Status: theorem.
Lean: estimates_weighted, JointEstimate, FitsOn; clause profiles of SeparatedOPE, through Core.joint.
Proposition VII.4.3 (nested OPE, i.e. composition of coefficients over trees; item (D10)).
Trees and plots.
\(t\) is a finite rooted tree whose root is an inner node, and \(v\) labels its leaves.
\(w_p\) is the total degree of the leaves below the inner node \(p\).
A plot \(z\) assigns points to the leaves and an anchor \(z_p\) to each inner node.
Radii \(s_p > 0\) are attached to the inner nodes, with \(\rho_p = s_p / s_{\text{parent} (p)}\) for non-root \(p\).
Admissibility. \((s, z)\) is \((r, M, \eta)\)-admissible if:
\(s_{\text{root}} = r\) and \(s_p \leqslant s_{\text{parent} (p)}\);
at every inner node \(p\), all positions of the subtree lie within \(Ms_p\) of \(z_p\);
positions in different children of \(p\) are at distance \(\geqslant \eta s_p\).
Costs. Put
Conclusions. Fix \(m\), \(\kappa \in \text{Ch}_m\), truncation degrees \(D_p\) at the inner nodes, \(M \geqslant 0\) and \(\eta, \varepsilon > 0\). For all sufficiently small \(r\) and every admissible \((s, z)\):
(a) \(|c_{m, \kappa} (v ; z_{\text{leaves}}, z_{\text{root}}) | \leqslant \text{toll}_m (s)\).
(b) \(|c_{m, \kappa} (v ; \cdot) - c^{(D)}_{m, \kappa} (v ; z) | \leqslant \text{toll}_m (s) \hspace{0.17em} \text{nest}_D (s)\). Define \(c^{(D)}\) recursively: each inner child \(q\) contributes a channel \((j_q, \kappa_q)\) with \(j_q \leqslant D_q\), its composed coefficient, and the output field \(O_{j_q, \kappa_q}\) at \(z_q\). Each leaf child contributes its original label and the factor \(1\). Multiply these factors by the root coefficient for the resulting child labels, and sum over the inner children's channels. The root output is fixed at \((m, \kappa)\); \(D_{\text{root}}\) is unused.
(c) Fix a finite spectator list \(b\) and \(\delta > 0\) before taking \(r\) small. If \(y\) is at distance \(\geqslant \delta\) from every point of the plot, with mutual distances \(\geqslant \delta\), then the degree-\(m\) remainder of VII.4.1 for the leaves at \(z_{\text{root}}\) has \(\mathcal{A}\)-norm, and absolute value in every state, at most \(\text{toll}_{m + 1} (s)\).
(d) Fix a measurable space \(K\). For each small \(r\), keep the radii \(s\) fixed throughout the integration. Let \(\mu\) be a positive measure on \(K\) and \(q \in L^1 (\mu ; \mathbb{C})\) with \(\int |q| \hspace{0.17em} d \mu \leqslant B\). Let the plot, and in (c) the spectators, be a.e.-measurable maps satisfying the stated geometry a.e. where \(q \ne 0\). The direct and composed coefficients are integrable after multiplication by \(q\); their integral and the difference of their integrals satisfy (a)–(b) with an additional factor \(B\). The tested remainder has the integrated conclusion of VII.4.2 with bound \(B \hspace{0.17em} \text{toll}_{m + 1} (s)\), for the mixed state or any a.e.-measurable phase map. The thresholds are uniform in \(s, \mu, q, B\) and the configuration maps.
Status: theorem for every Reader on the final record under (RG).
Lean:
Composite.token, Composite.nShot distinguish unchanged leaf inputs from inner-node expansions.
NestedOPE (clauses direct, tested, joint, jointLaw).
Accept, Tree.toll, Tree.nestPrice, Composite.stacked, CoefIntegral.
Proposition VII.4.4 (contact extensions; items (D2), (D11)–(D13)).
Setting.
\(\mathcal{M}\) is a system of contact devices on the final record. Its prescription is the Reader's (\(\mathcal{M}. \text{basis} =\mathcal{P}\); this is (D11)) and its rotation action is that of \(F\).
(RG) holds.
For labels \(a_1, \ldots, a_k\) of degrees \(d_i\) and a state \(s\):
let \(\text{cl} (n) > 0\) be the clearance of the devices up to degree \(n\);
put \(B_0 = \min (1, \text{cl} (w (a)))\) and fix a subtraction length \(s_0 \in (0, B_0 / 2]\); the public form uses \(s_0 = B_0 / 4\);
let \(\mathcal{D}^s_a (f) = \int \langle \text{full} \rangle_s (f ; x) \hspace{0.17em} dx\), the kernel being given by the forest formula of Lemma VII.4.r.
Conclusions.
(a) \(\mathcal{D}^s_a\) is a distribution of order at most \((2 + w (a)) \hspace{0.17em} k\). For every compact \(K\) there is \(C_K\) with \(\int | \langle \text{full} \rangle_s (f ; x) | \hspace{0.17em} dx \leqslant C_K \max_{| \alpha | \leqslant (2 + w (a)) k} \| \partial^{\alpha} f\|_{\infty}\) for all \(f\) supported in \(K\).
(b) (D2) \(\mathcal{D}^s_a (f) = \int f (x) \hspace{0.17em} S^s (a ; x) \hspace{0.17em} dx\) for every \(f \in C_c^{\infty}\) supported in \(\text{Conf}_k\).
(c) The same construction gives distributions on affine slices \(o + P (E)\), where \(P\) is a linear isometry whose range contains the normal space of the diagonal of a group \(g \subset \{1, \ldots, k\}\), and \(o\) lies on that diagonal.
(d) (D12) For the admissible stage families described below, the pairing equals the sum over subtraction choices and dyadic shells. Every shell kernel is integrable, the sum of their \(L^1\) norms is finite, and the iterated forest sums give the same pairing.
(e) (D12) For the single-block partition \(Q = \{g\}\) with \(g \ne \emptyset\), use pure dilation in its normal space \(K_g\), with \(R \in (0, 1]\). Put \(q_g = 4 (|g| - 1)\) and \(f_R (x) = R^{- q_g} f (R^{- 1} x_{K_g} + x_{K_g^{\perp}})\). If \(\text{supp} f\) lies in the rescaled stage and \(\|f\|_{C^N} \leqslant H\), where \(N = (2 + \sum_{i = 1}^k d_i) (|g| - 1)\), then for every \(\varepsilon > 0\) the integral of the absolute kernel is at most \(CHR^{\hspace{0.17em} u - w (a_g) - \varepsilon}\). Here \(u = 0\) for the full resolution, \(u = n\) for the weight-\(n\) coefficient and \(u = n + 1\) for the remainder after degree \(n\). For the last two bounds the stage must also lie within clearance \(\text{cl} (n)\) of the \(g\)-diagonal. For \(|g| > 1\), tests supported away from that diagonal give the same pairing as the punctured expression, which already includes proper-subcluster subtractions.
(f) (D13) The device strata have bases fixed by conjugation, and the basis in degree 0 contains the unit field.
(g) (D13) For every positive sequence \((f_n)\) there are thresholds \(0 < t_n \leqslant f_n\) with \(4 t_{n + 1} < t_n\), \(t_n \to 0\) and \(2 (n + 1) t_n \leqslant \text{cl} (n)\).
In (d)–(e), an admissible stage family consists of a partition \(Q\) of a group \(g\), affine placements \(o + P (E)\), positive subtraction lengths \(s\), and block scales \(R_i \in (0, 1]\). Each \(o\) lies on the block diagonals, and the range of \(P\) contains their normal spaces. A common ceiling \(B \leqslant \min (1, \text{cl} (\sum_i d_i))\) satisfies \(2 s \leqslant B\). At each placement a common linear frame acts by \(R_i\) on every block normal space and preserves each orthogonal complement. The closed stages lie in a fixed ball in frame coordinates and within normal distance \(B\) of each block diagonal; for nonsingleton blocks \(R_i \leqslant C_1 s\) with fixed \(C_1\). Original and spectator labels have uniform norm bounds. Spectator positions and labels are continuous on each stage and constant in each block's normal directions. All centroids of nonempty subgroups of different blocks, and all spectators, are uniformly separated by some \(\delta > 0\). These conditions are the geometric and field hypotheses for the summation and scaling estimates; their constants may depend on the fixed family.
Uniqueness, reflection positivity on overlapping supports, contact Ward identities for arbitrary subtractions and real-time products are not asserted.
Status: theorem (provided). In the main theorem the devices are assembled in contacts_on (VIII.1), with detectors from detector_exists. (D2) appears in the detailed theorem constructed, not in the public joint existence statement.
Lean:
Stage hypotheses: Route, Route.Authority, Backdrop.Eligible; output clearance: sees; derivative order: stairs.
MultipleLight, with clauses slope = Sunshine, moment = Resolution, and distribution = Charge of order totalTestOrder.
Public form: contactDistribution for Hierarchy.Contacts.
Proposition VII.4.5 (slab-operator bound; the bench V.1.7).
Hypotheses. \((\mathcal{W}, f, S)\) is a seed:
\(\mathcal{W}\) has the construction properties;
the extraction is ordered and has generic point limits;
\(S\) satisfies the Euclidean hierarchy;
the corrected field estimates of IV.1 hold;
the cell hypothesis holds: reflection positivity and the chessboard inequality at every cutoff, and centred bounds \(\|I_{N, p} (v, x)\|_{\mathcal{T}_m} \leqslant A \hspace{0.17em} a_m^{- d - \varepsilon} + \eta\) late in the extraction.
Conclusion. For every \(D \in \mathbb{N}\) and \(R > 0\) there are \(C_0 > 0\) and \(P \in \mathbb{N}\) with the following property. Let \(d > 0\), \(r \in (0, 1]\), and let \(w \neq \emptyset\) be a word with these properties:
its tokens have degree \(\leqslant D\) and coefficient norm \(\leqslant R\);
it lies in the time slab \((0, d)\) and is \(r\)-buffered there.
Then there are a level \(b\) and a width \(0 < s \leqslant d\) with \(s \in a_b \mathbb{N}\) and \(w\) in the slab \((0, s)\), such that every word \(v\) supported at positive times has a constant \(B_v\) with
Here \(v_n\) is \(v\) preceded by \(2 n\) consecutive slabs of width \(s\) that hold alternately \(w\) and its reflection. In V.1 an iterated Schwarz inequality turns this into \(\|T_s (w)\| \leqslant \rho\). One may take \(P = 4 (D + 1)\).
Status: theorem: Next.Phase.seats proves Bid.Seats.
It enters the phase construction of V.1 through Next.Phase.vacuum.
The OPE proofs of this unit do not use it directly; they use the algebra \(\mathcal{A}\) it helps to build.
Lean: Bid.Bench, OS.QRead, Spread.radius; steps in Lemmas VII.4.o–p.
Proposition VII.4.6 (coherent prescription).
Setting.
A probe is a finite separated configuration of tokens; its first point is the anchor, and the other points are taken relative to it.
\(\text{sp}_n : F_n \to \mathbb{C}^{\{ \text{probes} \}}\) sends \(\phi\) to the correlations of \(\phi\) at the anchor with the probe tokens. This map is injective, by the definition of \(F_n\) as a quotient by null fields, and compatible with the inclusions.
A mode is a pair \(i = (n, \alpha)\), where \(\alpha\) indexes a basis of \(F_n / \iota (F_{n - 1})\), with \(F_{- 1} = 0\).
Conclusion. There are, for every mode, a vector \(u_i \in F_n\) (needle) and a finitely supported family \(\lambda_i\) of complex weights on probes (sheet), with the following property. Write \(\lambda_i^{(m)} = \sum_p \lambda_i (p) \hspace{0.17em} \text{sp}_m (\cdot) (p)\) for the induced functional (test) on \(F_m\). Then for every \(n\):
\(\lambda_i^{(n)} (\iota u_j) = \delta_{ij}\) for all modes \(i, j\) of degree \(\leqslant n\);
\(\phi = \sum_{\deg i \leqslant n} \lambda_i^{(n)} (\phi) \hspace{0.17em} \iota u_i\) for all \(\phi \in F_n\).
Tests of degree \(> n\) vanish on \(F_n\). One prescription serves every truncation degree.
Status: theorem.
Lean: Prescription, SignedOff, Coherent, Prescription.chosen, highVan.
Remark VII.4.7 (hypotheses and order of choices).
The order of choices. In the main theorem (WilsonExistence), the family is chosen first, for \(G\) with simple Lie algebra and any penalty \(b > 0\). Then, for every extraction order, a refinement, the table, the final record, the Reader and the devices are chosen together.
The hypotheses (RG) (Services) bundle four items:
the grid-approximant property;
simplicity of the Lie algebra of \(G\);
the per-step and cluster RG estimates;
the continuum extraction record.
For the table of the construction these hypotheses are proved by Run.extractible17 (VIII.1, from Part III outputs; §2 input 6).
Separate existence theorems. through_final and the raw tree theorem allow any penalty \(b > 0\). Orbits.foundation, the single-cluster existence theorem, the covariant tree existence theorem and celestial choose \(b = 1\). The main theorem instead applies the OPE estimates to its already chosen family.
Further results. The same construction also gives:
the OPE in the raw (complex, non-covariant) needle basis (SingleSubpower);
tree versions for every coherent prescription (AllShapes, via Services.everything).
What is not claimed. The coefficients depend on the prescription and the Reader. Independence of the extraction subsequence is not claimed (Assessment §7.5).
Inputs from other units
The final record and its seed (IV.1, V.1–V.3, VI.1; assembled in VIII.1). This comprises:
the cutoff family with the construction properties (II.2, III.13);
the hierarchy (Euclidean axioms) of \(S\), and generic point limits on grid configurations;
the field algebra: inclusions \(\iota\), derivative identities, rotation action and its laws, reality structure.
Lean: Seed, Locals, Final.
Reflection positivity and centred bounds (IV.1, interface item 2; I.1). Two parts:
the chessboard inequality at every cutoff;
for every degree \(d\) and \(\varepsilon, R > 0\), there are \(m_0\) and \(A \geqslant 0\) such that, for every \(m \geqslant m_0\) and \(\eta > 0\), eventually along the extraction every \(I_{N, p} (v, x)\) with \(\|v\| \leqslant R\) is interior to its centred cell at scale \(a_m\), with root \(\leqslant A \hspace{0.17em} a_m^{- d - \varepsilon} + \eta\), uniformly in \(v\) and the site \(x\).
Lean: CellHyp.
The ground algebra and phases (V.1). This comprises:
\(\mathcal{A}\) commutative, with \(X\) its Gelfand spectrum and \(\nu\) of full support;
B17.answer: \(\langle \Omega, (u^{\ast} u)^n \Omega \rangle\) is the limit, as the spacing \(\to \infty\), of the continuum correlation of \(2 n\) time-translates of \(u\), alternately conjugated;
Retain.bit: if \(c \geqslant 0\) and \(\langle \Omega, (u^{\ast} u)^n \Omega \rangle \leqslant c^{2 n}\) for all \(n\), then \(| \chi_{\xi} (u) | \leqslant c\) for all \(\xi\);
the selection, the every-phase tables, and their smoothness.
The derivative identity (VI.1, IV.1): the derivative in one point of a correlation is the correlation with the derivative label, of degree one higher. It is used in Lemma VII.4.i.
Haar measure on \(O (4)\) and the rotation laws (VI.1). They are used for the covariant Reader.
The RG hypotheses (RG). The estimates come from Part III and are collected in VIII.1:
Org.Grill: for every degree bound \(d\) and level \(m\) there is \(k \geqslant m\) such that conjugation-fixed token tuples of degree \(\leqslant d\) on generic level-\(k\) grid configurations admit coverings that work along the extraction. It is supplied in III.11 and used in the a priori bounds (Lemma VII.4.e).
Package = TableTickets, together with cluster product cuts for every arity. It supplies per-step RG bounds, coefficient transport and column estimates, halo summability, polymer representations, and per-cell scale choices. These are the Part III inputs to Lemmas VII.4.f–g, collected in VIII.1.
The continuum extraction record (IV.1, Next/EXTRACTION/Base.lean): the extraction to the smooth continuum table. It is supplied with the extraction in III.11 and used in the grid estimates (Lemma VII.4.h).
Standard mathematics
Commutative C*-algebras: the Gelfand isometry and norm \(= \sup\) over characters (Bratteli–Robinson I, §2.3.5).
Reflection positivity and chessboard estimates (Fröhlich–Israel–Lieb–Simon 1978; Fröhlich–Simon 1977; Glimm–Jaffe, Ch. 6 and 18).
The iterated Schwarz inequality for transfer operators on an invariant pre-Hilbert domain: exponential bounds on the iterates of an adjoint pair give \(\|Av\| \leqslant D \|v\|\). This is used in V.1 (Glimm–Jaffe, Ch. 6).
Block-spin renormalization and polymer representations (Bałaban; Brydges, Lectures on the renormalisation group; Gawędzki–Kupiainen).
The Wilson–Zimmermann OPE (Wilson 1969; Zimmermann 1973) and its axiomatic form (Hollands–Wald 2008). The conclusions here are finite-order asymptotic bounds with \(\varepsilon\)-loss.
Extension of distributions across diagonals with Taylor subtraction of order given by the scaling degree (Epstein–Glaser 1973; Brunetti–Fredenhagen 2000; Hörmander I, Thm 3.2.3–3.2.4). Forest formulas (Zimmermann 1969).
Normalized Haar measure on \(O (4)\), finite-dimensional duality, Taylor's formula with remainder, and the mean value theorem.
Imported constants and assumptions
\(\varepsilon > 0\) is an arbitrary exponent loss. The estimates do not determine logarithmic corrections.
Constants of the cluster estimate (Lemma VII.4.f):
the gain margin \(1 / 16\), from the RG transport factor \((a_k / a_m)^{1 / 8}\);
the loss \((a_m / a_j)^{4 |I|}\), where the exponent \(4\) is bookkeeping slack (\(L_i^4 \geqslant 6 L_i\));
subpower factors in the per-cell, halo and insertion bounds, absorbed by \(\varepsilon\).
The constant \(P = 4 (D + 1)\) of VII.4.5 comes from degree-\(d\) bounds of the form \(a_k^{- (d + 1)}\) and \(a_k \geqslant c \hspace{0.17em} (r / |w|)^4\).
A coherent system of probes reads every output degree from the mixed correlation table. For mode \(i\), the raw coefficient is \(\lambda_i (a ; x, o) = \sum_p \lambda_i (p) S (a, x - o ; \text{probe } p)\), with each reading set to zero when the combined configuration is not separated (Prescription.ope). Biorthogonality makes these readings exact on single fields (sumAll). Averaging over \(O (4)\), symmetrizing the input and output conjugations, and removing lower-degree components gives the Reader's real covariant strata. These choices are fixed before the spectator labels, state and truncation degree.
The RG estimate replaces a lattice cluster by one label \(Q\) of prescribed degree \(t\). Its remainder gains \((a_m / a_k)^{t + 1 / 16}\), while the label and remainder both pay the factor \(a_j^{- w} (a_m / a_j)^{4 |I|}\). Along the scale induction, nearby groups fuse and separated groups pass through the next block-spin step. After extraction, exact reconstruction of \(Q\) by the probes yields the grid OPE. Reflection positivity bounds decks of widely translated copies by products of chessboard roots, treating a cluster remainder as one factor. The deck limit gives all vacuum moments of \(u^{\ast} u\) in the ground algebra; full support of the phase measure and the Gelfand isometry then give \(\|u\|_{\mathcal{A}}\). Each spectator costs only its own one-field bound. The resulting norm estimate controls every phase and the mixed state with the same coefficients.
Exponential rates organize the choices of lattice scales. A downward induction in total degree extends the grid bounds to arbitrary configurations: each derivative raises the degree by one, while the distance to the fine grid recovers one power. At sufficiently high degree the crude bound already suffices to start the induction. Bringing the auxiliary scale heights together gives an arbitrary \(\varepsilon\)-loss; refinement arguments turn these rates into uniform small-radius estimates. Tree expansions repeat this replacement recursively, leaving leaf labels unchanged. Contacts use a forest recursion: each retained weight receives a normal Taylor subtraction of its own order, and the remainder gains enough radial power for integration. Partial collisions are handled by the same recursion on finer partitions. Separately, the slab-operator bound follows by applying the two-scale chessboard estimate to a reflection-positive train and using iterated Schwarz to remove the fixed exterior cost.
Lemma VII.4.a (from sequences and rates to uniform small-\(r\) bounds).
(i) Let \(P (r, z)\) and \(Q (r, z)\) be predicates. Suppose that for every choice \(r \mapsto z_r\) and every non-trivial refinement of \(r \to 0^+\) along which \(P (r, z_r)\) holds eventually, a further refinement makes \(Q (r, z_r)\) hold eventually. Then, for all sufficiently small \(r\), \(P (r, z)\) implies \(Q (r, z)\) for every \(z\).
(ii) Rates. A quantity has upper rate \(\kappa\) along \(T \to \infty\) if \(|f| \leqslant e^{\kappa' T}\) eventually, for every \(\kappa' > \kappa\). With \(T = - \log r\) this means \(|f| \leqslant r^{- \kappa - \varepsilon}\) eventually, for every \(\varepsilon > 0\). Below, “rate” means upper rate unless an exact rate is specified.
(iii) Clusters and spectators. Under \(T = - \log r\), an \((r, M, \eta)\)-cluster has extent of upper rate \(- 1\) and reciprocal separation of upper rate \(1\). For \(c\)-clear spectators, both the reciprocal mutual separation and the reciprocal distance to the centre have upper rate \(0\).
(iv) Heights. A sequence of levels \(m \to \infty\) has height \(c\) if \(\log (1 / a_m) / T \to c\). Every height \(c \geqslant 0\) is attained for a gentle spacing sequence.
Proof sketch. For (i), choose a bad configuration at each bad radius and refine to a non-trivial filter supported on those radii. The assumed further refinement is a contradiction. Parts (ii)–(iii) follow from \(r = e^{- T}\). For (iv), choose the first level whose logarithmic spacing exceeds \(cT\) when \(c > 0\), and \(\sqrt{T}\) when \(c = 0\); gentleness makes the relative overshoot tend to zero.
Lean: refine_geometry, RateLE, loglife, room_regime, free_clear, Height, found.
Lemma VII.4.b (extension one level at a time). A prescription that is biorthogonal and reconstructing on \(F_n\) extends to one with the same property on \(F_{n + 1}\), without changing the needles and sheets of degree \(\leqslant n\).
Proof sketch.
Let \(B = \sum_{\deg i \leqslant n} \lambda^{(n + 1)}_i (\cdot) \hspace{0.17em} \iota_{\deg i, n} u_i : F_{n + 1} \to F_n\), a left inverse of \(\iota_{n, n + 1}\).
Take any linear section of \(F_{n + 1} \to F_{n + 1} / \iota F_n\), and subtract from it its projection \(\iota \circ B\). This gives a lift \(\Phi\) whose image lies in the kernel of all old tests.
The new needles are \(\Phi\) applied to a basis of the quotient.
The new sheets realize the dual coordinates composed with the quotient map. Such sheets exist because the probe functionals span the dual of \(F_{n + 1}\) (reachReads, by faithfulness).
A recursive choice, collected along \(n\), gives a coherent prescription.
Lean: startRows, nextRows, Prescription.chosen.
Lemma VII.4.c (the Reader). Fix a coherent prescription.
Construction.
The raw vector of degree \(n\) is \(\sigma_n (a ; x, o) = \sum_{\deg i \leqslant n} \lambda_i (a ; x, o) \hspace{0.17em} \iota u_i \in F_n\).
Define the covariant average
where \(\rho (g^{- 1})\) acts on every input label. Set \(A_n = 0\) outside the domain where all rotated probes through degree \(n\) stay clear. The conjugation-symmetric reading is
Define recursively \(\pi_n = \text{id} - \sum_{i < n} \iota \circ \pi_i \circ \ell_{i, n}\) on \(F_n\), where \(\ell_{i, n} (\phi) = \bar{\sigma}_i (\phi ; 0, 0)\).
The cut tests are \(\tau_{n, q} = \pi_n \circ \ell_{n, q}\), and the stratum is \(E_n = \bigcap_{i < n} \ker \tau_{i, n}\).
The fresh vector is \(\pi_n \bar{\sigma}_n (a ; x, o) \in E_n\).
\((O_{n, \kappa})\) is a basis fixed by conjugation of \(E_n\), and \(c_{n, \kappa} (a ; x, o)\) are the coordinates of the fresh vector in it.
Properties.
\(\tau_{n, q} = 0\) for \(q < n\).
\(\sum_{i \leqslant n} \iota \tau_{i, q} = \iota_{q, n}\) for \(q \leqslant n\).
The fresh vectors and cut tests intertwine rotations and conjugation; simultaneous translations and permutations of the inputs preserve the fresh vectors. Coordinates transform in the corresponding channel representations.
The coefficients are smooth where the probes are clear.
Proof sketch.
On a single field of degree \(j \leqslant n\) at the origin, the degree-\(n\) averaged reading is \(\iota_{j, n}\) (checkReal_fixed); this uses coherence and the rotation laws.
The identities for \(\tau\) then follow by strong induction on \(n\).
A complex finite-dimensional space with an antilinear involution has a basis of fixed vectors.
Lean: checkValue, Reader, averageV, checkReal, knife, cutTest, reconstruct, kills_lower, stratum, fresh, Reader.basis, coef.
Lemma VII.4.d (norm from decks). Let \(u \in \mathcal{A}\) come from a finite combination of separated grid words. It has lattice versions \(U_{N, p}\) at the cutoffs.
Hypothesis. Let \(c \geqslant 0\). The lattice labels converge to those defining \(u\) along a non-trivial refinement of the extraction. For every \(n\) and all sufficiently large translation spacings \(t\), eventually along that refinement the lattice moment of the deck \(\prod_{l < 2 n} U^{(\ast)}_{N, p} \circ \tau_{2 tl}\) (time translates, even ones conjugated) has modulus \(\leqslant c^{2 n}\).
Conclusion. \(\|u\|_{\mathcal{A}} \leqslant c\).
Proof sketch.
The deck moments converge to the continuum deck correlation.
As \(t \to \infty\), that correlation tends to \(\langle \Omega, (u^{\ast} u)^n \Omega \rangle = \int | \chi_{\xi} (u) |^{2 n} d \nu\) (B17, V.1).
Since \(\nu\) has full support, the bound for all \(n\) gives \(| \chi_{\xi} (u) | \leqslant c\) for every \(\xi\).
By Gelfand, \(\|u\| = \sup_{\xi} | \chi_{\xi} (u) |\).
Lean: Cards.carrying, Cards.pressure. (proof read; the geometric inputs to B17.answer are taken from V.1)
Lemma VII.4.e (a priori and replacement bounds in \(\mathcal{A}\)). Assume (RG).
Setting. Main points are at mutual distance \(> 2 a_b\). Exterior points are at mutual distance \(> 2 a_l\), and at distance \(> 2 a_b\) from the main points. The levels satisfy \(b \leqslant f \leqslant l \leqslant n\) and \(f \leqslant q\), with
Main tokens have centred bounds \(B_i\) at level \(q\), and exterior tokens have bounds \(D_j\) at level \(n\).
Conclusions.
(i) \(\|M (\text{main} \cup \text{exterior})\|_{\mathcal{A}} \leqslant \prod_i B_i \prod_j D_j\).
(ii) For a cluster on a fixed grid with distinct coordinates on every axis, suppose lattice labels \(Q_{N, p}\) of degree \(t\) converge to \(Q\) along a non-trivial refinement of the extraction. Suppose \(R_{N, p} = \prod_i I_{N, p} (v_i, x_i) - I_{N, p} (Q_{N, p}, 0)\) is interior to its centred cell at level \(q\) with root \(\leqslant \alpha\) along that refinement. Require \(b\) no finer than the grid, the origin to be \(2 a_b\)-clear of the exterior, and both the cluster–exterior and origin–exterior configurations to be separated. Use the cell-fitting inequalities above with one main factor. Then
(iii) In rate form, let \(0 \leqslant \alpha \leqslant \beta\). If reciprocal main–main and main–exterior distances have upper rate \(\alpha\), and reciprocal exterior–exterior distances have upper rate \(\beta\), then \(M\) has upper rate \(w (\text{main}) \alpha + w (\text{ext}) \beta\), uniformly in states.
Proof sketch.
(i)–(ii): the two-scale chessboard inequality is applied to decks of \(2 n\) translates. Main factors occupy distinct cells of the coarse tiling. Exterior factors, possibly several per coarse cell, are grouped and bounded in a finer aligned tiling. This gives the deck bound \((\prod B \prod D)^{2 n}\) or \((\alpha \prod D)^{2 n}\).
Lemma VII.4.d then gives the norm bound at grid configurations. Smoothness of correlations extends it to the limit configuration.
(iii): take bounds \(a_k^{- d - \delta}\) at heights slightly above the separation rates.
Lean: Cards.plant, Cards.mixed, Cards.Hall, civic, viewing_admits, mirrorMatch, admitted, phase_budget, viewer_cost. (proof read; TileDoc.global statement only)
Lemma VII.4.f (lattice cluster replacement). Fix the RG representation, product-cut and transport data of §2 for output degree \(t\). Let \(I\) be finite and nonempty. Levels \(J \leqslant k \leqslant m < m + 1 \leqslant j < N\) are ordered from coarse to fine. Require \(\text{ref} (t) \leqslant k\) and \(\text{ref} (d_i) \leqslant j\) for every input degree \(d_i\).
Hypotheses.
Geometry. Lattice observables \(f_i = I_{N, p} (s_i, o_i)\) of degrees \(d_i\) sit at level-\(j\) sites whose level-\((m + 1)\) ancestor is the centre cell. They are pairwise at least \(B_j^2\) level-\(j\) units apart, where \(B_j\) is the polymer scale. Require
Here \(\mathfrak{c}_k\) is the supplied cover count, \(\ell_t\) is the output stencil radius plus \(4\), and \(\tau_m = 30 a_m (\jmath_m + 5)\) is the cone-tail bound with supplied aperture parameter \(\jmath_m\). All \(f_i\) and all \(I_{N, p} (z, 0)\) are interior to the centre chamber.
Weights. Choose \(\alpha_i > 0\) with \(\alpha_i \hspace{0.17em} \text{paid}_j (s_i) \leqslant \zeta (d_i, j)\). Here \(\text{paid}_j\) is the insertion-size bound and \(\zeta\) the allowed RG weight. Choose \(0 < \alpha \leqslant 1\) below every halo allowance from \(m\) through \(j - 1\). With \(H = |I|\), require
Output weights. Let \(\delta_m, x_m, \varphi_m, \zeta (t, k)\) and \(c_m\) be the supplied RG constants, with \(c_m\) the column count. Write \(\mathfrak{m}_{k, m} (b)\) for the weight obtained by transporting \(b > 0\) from \(k\) to \(m\), and \(A_{k, t}\) for the output transport map. Require \(e > 0\) and
Conclusion. There is a lattice label \(z\) of degree \(t\) with these properties:
its coefficient size at level \(m\) is at most \(2 e\);
\(R = \prod_i f_i - I_{N, p} (z, 0)\) is interior to the centre chamber;
\(\|R\|_{\mathcal{T}_J} \leqslant 2 \hspace{0.17em} \text{stakes} (J, k) / b\).
Here \(\text{stakes} (J, k) = \exp ((a_J / a_k)^4 \text{treasury}_k + 5)\) is the supplied per-cell chessboard factor. In Lemma VII.4.g the two coarse levels are chosen so that this factor has upper rate \(0\).
Proof sketch.
Transport. From \(N\) down to \(j\), each insertion is transported separately, since the points are separated.
Induction from \(j\) down to \(m + 1\). The carried state is a polymer representation of the RG-pushed generating function, with the points grouped and the representatives spaced. At each step one of two things happens:
if the representatives stay spaced, one block-spin step is taken, and the weights lose at most a factor \(L_i^4\) per point;
otherwise a maximal spaced set is kept and the remaining representatives are fused into it. Enlarging polymer supports to reach a representative costs a bounded factor in halo summability. Each fusion pays a subpower toll.
The weights. Their surviving product contains \((a_j / a_{m + 1})^{4 |I|} (\alpha / 8)^{|I| (|I| - s)}\) when \(s\) groups remain. Its reciprocal contributes to the final error bound.
Final step. At level \(m + 1\), fuse the remaining groups into one variable per cell copy. The RG step to level \(m\) splits the result into a coefficient \(Q_0\) of degree \(\leqslant t\), with its coefficient and remainder bounds, and an irrelevant remainder. Transport to the coarse level gives a factor \((a_m / a_k)^{t + 1}\) in the remainder bound.
The constant \(2\) comes from two halves, each \(\leqslant \text{stakes} / b\), joined by the triangle inequality for roots.
Lean: joins, inductionLive, Ballet, ranch, hitB. (statement read; proof read in outline, at the level of the calls in Ballet, inductionLive, Live.gathers, bequeathed)
Lemma VII.4.g (replacement rates and the limit label). Assume (RG), fix a nonempty finite token list of total degree \(w\), and take heights \(0 \leqslant b < a < c\) for the coarse, output and insertion levels. Fix a target degree \(t\), and put \(Z' = w \hspace{0.17em} c + 4 |I| (c - a)\).
Conclusion. There are rate functions \(g\) and \(o\) with these properties:
\(g\) has exact rate \(Z' - ta\);
\(o\) has upper rate \(Z' - (t + \tfrac{1}{16}) (a - b)\);
eventually, the replacement of Lemma VII.4.f has coefficient size \(\leqslant g\) and remainder root \(\leqslant o\), whenever its cluster and interior conditions hold.
Up to subpower factors, the coefficient size is bounded by \(a_j^{- w} (a_m / a_j)^{4 |I|} a_m^t\), and the remainder root by \(a_j^{- w} (a_m / a_j)^{4 |I|} (a_m / a_k)^{t + 1 / 16}\). For each fixed configuration satisfying the cluster and interior conditions, bounded coefficient size gives a convergent refinement of the extraction in the finite-dimensional label space. Its limit \(Q\) retains the remainder bound.
Proof sketch.
Apply Lemma VII.4.f with all parameters exponential in \(T\), and check its size conditions as inequalities between exponents.
The loss \(4 |I| (c - a)\) is the rate of \((a_{m + 1} / a_j)^{4 |I|}\), the reciprocal of the surviving weight factor.
The binding condition is the transport condition, which needs exponent \(< (t + \tfrac{1}{8}) (a - b) - Z'\). The value \(\tfrac{1}{16}\) leaves strict slack.
Subpower factors have rate \(0\).
Lean: funding, PriceAt, price_plot, Survival, for_plot.
Lemma VII.4.h (grid estimates).
Setting. Assume (RG) and fix a coherent prescription and a nonempty finite token list \(a\). Heights satisfy \(0 < \beta < a_0 < x < d\), for levels \(k, m, j\) at heights \(\beta, a_0, d\). The cluster consists of level-\(j\) grid points with distinct coordinates on every axis, extent of upper rate \(- x\) about the origin and reciprocal separation of upper rate \(x\). Spectator distances from the origin have reciprocal upper rate \(0 \leqslant c < \beta\), and their mutual distances have reciprocal upper rate \(u \geqslant c\). Put \(\text{tax} = w (a) d + 4 |I| (d - a_0)\).
Conclusions.
each fixed raw coefficient \(\lambda_i (a ; \text{cluster}, 0)\) has upper rate \(\text{tax} - (\deg i) a_0\);
for every \(N \in \mathbb{N}\), the correlation with the spectators minus \(\sum_{\deg i \leqslant N} \lambda_i \hspace{0.17em} S^s (u_i, 0 ; \text{spect.})\) has upper rate \(\text{tax} - (N + 1) (a_0 - c) + w (b) u\), uniformly in the state \(s\).
Proof sketch. For the remainder choose \(q \geqslant N\) with \((N + 1) a_0 \leqslant (q + \tfrac{1}{16}) (a_0 - \beta)\); for a coefficient choose \(q \geqslant \deg i\) with the analogous inequality. Let \(Q\) be the degree-\(q\) label of Lemma VII.4.g. Then:
(1) by Lemma VII.4.e(ii), replacing the cluster by \(Q\) at \(0\) costs rate \(\text{tax} - (q + \tfrac{1}{16}) (a_0 - \beta) + w (b) u\) in \(\mathcal{A}\);
(2) the same with the probes in the role of spectators: \(| \lambda_i (\text{cluster}) - \lambda_i (Q) |\) has the rate of the remainder root;
(3) for \(\deg i \leqslant q\), the coefficient of \(Q\) is bounded by its coefficient size times \(a_m^{- (q - \deg i)}\), up to subpower factors, and hence has rate \(\text{tax} - (\deg i) a_0\);
(4) exact reconstruction: \(S (Q, 0 ; \cdot) = \sum_{\deg i \leqslant q} \lambda_i (Q) \hspace{0.17em} S (u_i, 0 ; \cdot)\);
(5) the modes of degree in \((N, q]\) contribute at most rate \(\text{tax} - (N + 1) (a_0 - c) + w (b) u\), using the upper rate \((\deg i) c + w (b) u\) for \(S^s (u_i, 0 ; \text{spect.})\) from Lemma VII.4.e(iii).
Lean: grid_coeff, grid_remainder, see_cost, det_cost, at_grade, sumAll.
Lemma VII.4.i (derivative bootstrap). Fix a finite index set \(I\) and levels \(j\) of height \(d > x_0\). A configuration path \(x (T)\) lies in the height-\(x_0\) corridor if its extent about the origin has upper rate \(- x_0\) and its reciprocal separation has upper rate \(x_0\). Let \(h_T (v ; x)\) be functions for every token list \(v\) on \(I\). Suppose the following hold for every such list and every configuration path in this corridor, with constants \(c_0, d_0\) independent of the list:
(rough) \(h_T (v ; x (T))\) has upper rate \(x_0 w (v) + c_0\);
(grid) on level-\(j\) grid paths with distinct coordinates on every axis, \(h_T (v ; x (T))\) has upper rate \(dw (v) + d_0\);
(derivative) eventually \(h_T (v ; \cdot)\) is differentiable at \(x (T)\), and its derivative in each point and coordinate equals \(h_T\) with that token replaced by its derivative token, of degree one higher.
Then \(h\) has rate \(d \hspace{0.17em} w (v) + d_0\) on the whole corridor.
Proof sketch.
Use induction downward on \(w (v)\).
For \(w (v)\) large, \((d - x_0) w (v) \geqslant c_0 - d_0\), so the rough bound suffices.
Otherwise choose a grid path whose distance from \(x (T)\) has upper rate \(- d\). All intervening points stay in the same corridor. The mean value theorem and the induction bound for the derivative give rate \(- d + d (w (v) + 1) + d_0 = dw (v) + d_0\) for the change.
Lean: interpolated, cloud_jump, leap_cost. (proof read for interpolated)
Lemma VII.4.j (raw coefficient and remainder rates).
Setting. Assume (RG) and fix a coherent prescription and finite token lists \(a, b\). Along \(T \to \infty\), the cluster has extent of upper rate \(- x\) around \(o\) and reciprocal separation of upper rate \(x > 0\). Spectator distances from \(o\) have reciprocal upper rate \(0 \leqslant y < x\), and their mutual distances have reciprocal upper rate \(u \geqslant y\).
Conclusions.
For each fixed mode \(i\), \(\lambda_i (a ; \cdot, o)\) has upper rate \((w (a) - \deg i) x\).
For every \(N\), the remainder of the raw truncation has upper rate \(w (a) x - (N + 1) (x - y) + w (b) u\), uniformly in states and in \(\mathcal{A}\).
In powers, with \(r = e^{- xT}\), spectator distance \(\rho\) and spectator separation \(\rho'\): the remainder is \(\lesssim r^{- w (a)} (r / \rho)^{N + 1} {\rho'}^{- w (b)}\) up to \(\varepsilon\)-losses.
Proof sketch.
Apply Lemma VII.4.i to Lemma VII.4.h, with grid height \(d\) and auxiliary height \(a_0\). The rough bound is Lemma VII.4.e(iii).
This gives the bounds of Lemma VII.4.h with \(\text{tax} = wd + 4 |I| (d - a_0)\). Let \(a_0 \uparrow x\) and \(d \downarrow x\); continuity of the exponents and availability of all intermediate heights give the claimed rates. Empty clusters are handled by the exact unit identities.
Translation invariance moves the centre to \(o\).
The bound in \(\mathcal{A}\) follows from the bounds in every state by the norm characterization in the conventions.
Lean: coefficient_rough, common_rate, early_rate, group_coeff, group_remainder, ground_cost, groundError, arrangement_cost.
Lemma VII.4.k (transfer to the Reader). In the regime of Lemma VII.4.j:
the fresh vector of degree \(n\) has rate \((w (a) - n) x\), and hence so does each \(c_{n, \kappa}\);
the error element of VII.4.1 has rate \(w (a) x - (D + 1) (x - y) + w (b) u\) in \(\mathcal{A}\).
Proof sketch.
Compare the cluster with its raw truncation at a horizon \(H\), placed at \(o\). Inside every rotated probe correlation, the difference has rate \((w - H - 1) x\) by Lemma VII.4.j.
The raw estimate holds uniformly over all rotations, including the conjugate input reading. Integrating over normalized Haar measure and applying the fixed linear projections preserve its upper rate. Thus the fresh vector differs from \(\tau_{n, H}\) of the raw vector by upper rate \((w - H - 1) x\).
In \(\tau_{n, n}\) of the raw vector, only modes of degree \(n\) survive, with rate \((w - n) x\).
For the remainder, write the error as the raw error plus \(\sum_{j \leqslant D}\) (the correlation of \(\tau_{j, D} (\text{raw}) - \text{fresh}_j\)), and bound each term.
Lean: fresh_bridge, fresh_price, balance_rate, coefficient_rate, algebra_rate. (proof read; fit_old, discrepancy_rate_old statement only)
Lemma VII.4.l (proof of VII.4.1). For (O1) and (O2), apply Lemma VII.4.a with \(T = - \log r\).
An \((r, M, \eta)\)-cluster sits at height \(x = 1\).
Fixed clearance \(c\) gives \(y = u = 0\).
So Lemma VII.4.k gives rates \(w - n\) and \(w - (D + 1)\). These are the powers \(r^{n - w - \varepsilon}\) and \(r^{D + 1 - w - \varepsilon}\), uniformly in configurations and states.
Lean: coefficient_separated, quantitative_separated, produce; assembly in R2IGoals.fromFunded.
Lemma VII.4.m (integrated forms; VII.4.2).
Proof sketch.
For small \(r\), every admissible configuration lies in an open set on which the correlations and the coefficients are continuous. There, the raw correlation is bounded by \(r^{- w - \varepsilon}\) and the remainder by \(r^{D + 1 - w - \varepsilon}\), in every state.
So the integrands are measurable and dominated by \(|w|\) times a constant, and the integrals are bounded by \(B\) times the pointwise bound.
For any probability measure on \(X\), continuity of each fixed separated correlation in \(\xi\) permits termwise integration of the finite truncation; the same pointwise bound then gives the mixture estimate.
Lean: estimates_weighted, probability_mixtures.
Lemma VII.4.n (nested OPE; VII.4.3).
Proof sketch.
Along a sequence, put \(T = \sum_p \log (1 / s_p)\), summing over the inner nodes. The normalized logarithms lie in a compact simplex, so a refinement gives heights \(a_p \geqslant 0\) with \(\sum_p a_p = 1\). This also covers scales shrinking faster than every power of the root radius.
Then induct over the tree.
At a fork, children whose rate equals the parent's are merged into the parent cluster.
Each other child is replaced by its own expansion at an auxiliary horizon high enough that its remainder is below any prescribed rate.
The replacements are telescoped in a fixed triangular order, so that each switch sees only outputs already fixed.
Summing the node costs gives (a). The composition error gains the smallest value of \((D_p + 1) (a_p - a_{\text{parent} (p)})\) over non-root inner nodes. The extra \(\varepsilon\)-loss converts this rate estimate into (b). If there are no such nodes, the composed coefficient equals the direct coefficient exactly.
(c) and (d) follow as in Lemmas VII.4.k–m.
Lean: Fusion.tariff, Quantified.lean (Limits.coe_power, Limits.tested_power), Tree.Tapered, Tree.Relative. (proof read for coe_power, tested_power, tariff; supplies proof read; sequential and high statement only)
Lemma VII.4.o (point and train bounds). Take levels \(m \leqslant k \leqslant l \leqslant b\).
(a) Point bound. Assume:
all points are separated; main–main and main–exterior sup-distances exceed \(a_m\), and exterior–exterior sup-distances exceed \(a_l\);
main points are congruent mod \(a_m\) to one of \(P\) fixed points;
\(P \cdot 10 a_k < a_m\) and \(\sharp (\text{ext}) \cdot 6 a_b < a_l\);
\(A, B \geqslant 0\); main tokens have one-field bounds \(A\) at level \(k\) for all signed and reflected variants, and exterior tokens have bounds \(B\) at level \(b\), eventually along the extraction;
the cutoff family has the construction properties, \(S\) is smooth on separated configurations, and generic point limits hold along this extraction.
Then \(|S (x ; v) | \leqslant A^{\sharp \text{main}} B^{\sharp \text{ext}}\).
(b) Train bound. Assume the Euclidean hierarchy and the same cutoff and point-limit hypotheses. Let \(w\) be \(r\)-buffered in a slab of width \(d \in a_m \mathbb{N}\), with \(a_m < r\). Let \(v\) be a word supported at positive times with buffer larger than \(a_l\). Require \(2 |w| \cdot 10 a_k < a_m\) and \(2 |v| \cdot 6 a_b < a_l\), and the one-field bounds in (a). Then for every \(q \in \mathbb{N}\) the Gram pairing satisfies
Proof sketch.
(a) The two-scale chessboard estimate at the cutoff: main factors are in separated coarse cells, and exterior groups in a finer tiling, with no restriction on how many exterior factors share a coarse cell. A translation of less than one exterior cell fits all the margins. Point limits at generic configurations and continuity give the continuum bound.
(b) The Gram pairing is the correlation of \(4 q |w|\) main and \(2 |v|\) exterior insertions at separated points. Apply (a).
Lean: cutoff_bound, point_bound, Gapped, travel_price.
Lemma VII.4.p (uniform constants; proof of VII.4.5).
Proof sketch.
One-field bounds. For degrees \(\leqslant D\) and norms \(\leqslant R\) there are \(F\) and \(J > 0\) such that, for \(k \geqslant F\), every such token, together with all its signed and reflected variants, has one-field bound \(J \hspace{0.17em} a_k^{- (D + 1)}\).
Choice of level. For \(r \in (0, 1]\), take the first \(m \geqslant F\) with \(a_m < r / 32\). Gentleness gives a uniform lower bound \(a_m \geqslant c_1 r^2\). A second level choice gives \(k \geqslant m\) with \(2 |w| \cdot 10 a_k < a_m\) and \(a_k \geqslant c (r / |w|)^4\). Hence \(A^{|w|} \leqslant \rho = C_0^{|w|} (|w| / r)^{4 (D + 1) |w|}\) with \(C_0 = Jc^{- (D + 1)}\).
Exterior. For each exterior word \(v\), choose its own levels and bound \(X_v\). Lemma VII.4.o(b) then gives the bound of VII.4.5 with \(B_v = X_v^{2 |v|}\).
Width. A sub-width \(s \leqslant d\) covered at level \(m\) is chosen before the exterior.
Lean: rates, replicas, seats. In V.1: Drive.against, producer.
Lemma VII.4.q (Taylor deviation on shells). Let \(E\) be a finite-dimensional Euclidean space, \(K \subset E\) a subspace, \(m, n \in \mathbb{N}\), \(B \geqslant 1\), and \(U\) a compact subset of \(\{ \tfrac{1}{2} \leqslant |x_K | \leqslant 2\}\). Write \(D_a x = ax_K + x_{K^{\perp}}\). There is \(X \geqslant 1\), depending only on \(E, K, U, m, n, B\), with the following property. Let \(a, r, s > 0\) and let \(A : E \to E\) be a linear isomorphism satisfying \((Ax)_K = rx_K\) for every \(x\) and \(Ax = rx\) for \(x \in K\). Let \(f\) be smooth and assume \(ra / s \leqslant B\), \(r \leqslant Bs\), and
Then the cut-off normal Taylor operator satisfies
Proof sketch. Taylor's formula with remainder, together with jet bounds for the smooth cut-off.
Lean: compact_deviation.
Lemma VII.4.r (the subtraction formula).
Notation.
\(\varphi_i\) is the field \(a_i\) at \(x_i\).
For a group \(h \subset \{1, \ldots, k\}\), \(|h| \geqslant 2\): \(x^{\perp}_h\) is the component of \(x\) normal to the \(h\)-diagonal, \(\bar{x}_h\) the centroid, and \(L_h = \max \left( 0, \sum_{i \in h} d_i + 1 - 4 (|h| - 1) \right)\).
\(T^h_m f (x) = \chi (|x_h^{\perp} |^2 / s_0^2) \sum_{l \leqslant m} \frac{1}{l!} D^l f (x - x^{\perp}_h) [x^{\perp}_h, \ldots, x^{\perp}_h]\), with \(\chi = 1\) near \(0\) and \(\chi (|x^{\perp}_h |^2 / s_0^2) = 0\) for \(|x^{\perp}_h | \geqslant 2 s_0\).
\(R^h_w\) replaces a field polynomial by the weight-\(w\) component of its OPE projection, placed at \(\bar{x}_h\). The device detectors use finite probe sets, a linear decoder, an \(O (4)\)-average and simultaneous input and output conjugation, followed by the device's graded splitting.
Definition. Set \(C_{\{i\}} = \varphi_i \otimes 1\) and, for \(|h| \geqslant 2\),
where \(\pi\) runs over partitions of \(h\) into at least two blocks. For a singleton the full expression is \(\varphi_i \otimes 1\); for the empty group it is the unit. The kernel of \(\mathcal{D}^s_a\) at \(x\) is the \(s\)-expectation of \(\text{full}_{\{1.. k\}}\) applied to \(f\). Expanding the recursion gives a Zimmermann-type forest sum. The child expressions are constructed recursively first. On the test function, the parent Taylor operator acts first, followed by the operators already present in the child expression; nested operators need not commute.
Per-group form. \(\left[ \Pi_h - \sum_{w < L_h} R^h_w \Pi_h \right] f + \sum_{w < L_h} R^h_w (\Pi_h) \hspace{0.17em} (1 - T^h_{L_h - 1 - w}) f\).
Put \(d_h = \sum_{i \in h} d_i\) and \(q_h = 4 (|h| - 1)\). After retaining weights \(w < L_h\), the remainder has radial order \(O (|x^{\perp}_h |^{L_h - d_h - \varepsilon})\), with \(L_h - d_h \geqslant 1 - q_h\).
A retained weight-\(w\) term has coefficient order \(O (|x^{\perp}_h |^{w - d_h - \varepsilon})\) and test deviation \(O (|x^{\perp}_h |^{L_h - w})\).
For \(0 < \varepsilon < 1\) these radial powers are integrable in \(q_h\) normal dimensions. Angular sectors containing partial collisions require the subcluster induction of Lemma VII.4.s.
Each weight-\(w\) coefficient has scaling degree at most \(d_h - w\), and receives its own Taylor subtraction.
If \(\sum_{i \in h} d_i + 1 \leqslant 4 (|h| - 1)\), nothing is subtracted.
Lean: takeRight, Score/Formula.lean, Normal.taylor, majorKernel, Devices, Devices.clearance. (definitions read)
Lemma VII.4.s (absolute bound; summation and scaling). Fix the devices, labels, state and subtraction length as in VII.4.4. For every compact \(U \subset (\mathbb{R}^4)^k\) there is \(C > 0\) such that
The summation formulas and scaling bounds hold under the stage-family hypotheses of VII.4.4(d)–(e).
Proof sketch.
Patching. Cover \(U\) by finitely many patches. On each patch the relevant partition is the set of coincidence classes of its centre. A smooth partition of unity and the Leibniz rule combine the patch bounds.
Induction. On a patch, use strong induction on \(|h|\) minus the number of blocks.
Base case: the separated correlation times \(f\), bounded pointwise.
Inductive step: split into the remainder arm and the weight-\(w\) arms. Cut into dyadic normal shells \(|x^{\perp}_h | \sim 2^{- n}\) and angular fans. Rescale each sector to a finer partition, and apply the induction hypothesis.
OPE input. The raw coefficient rate and the remainder rate in \(\mathcal{A}\) from Lemma VII.4.j.
Shell exponent. Including normal volume, the effective exponent is \(4 (|h| - 1) + L_h - d_h \geqslant 1\):
the remainder gains \(L_h\);
a weight-\(w\) arm gains \(w\) from the coefficient and \(L_h - w\) from Lemma VII.4.q.
Summing. Choose \(\gamma > 0\) with \((3^{|g| - |Q|} + 1) \gamma \leqslant 1\), where \(|Q|\) is the number of blocks. The accumulated exponent losses leave a factor \(2^{- n \gamma}\) in the shell sum. The corresponding geometric series converges.
Lean: glues_total, global_charge, commissions, Claim, Annulus. (proof read in outline)
Lemma VII.4.t (distribution and restriction).
A kernel with the bound of Lemma VII.4.s defines a continuous linear functional on \(C_c^{\infty}\), of order \(\leqslant (2 + w (a)) k\).
With the full group, \(P = \text{id}\), \(o = 0\), \(B_0 = \min (1, \text{cl} (w (a)))\) and \(s_0 = B_0 / 4\), the hypotheses of Lemma VII.4.s hold on all of \((\mathbb{R}^4)^k\).
If \(\text{supp} f \subset \text{Conf}_k\), the Taylor terms evaluate jets of \(f\) on diagonals, where \(f\) vanishes with all its derivatives. So the kernel equals \(f (x) S^s (a ; x)\).
Lean: Charge, Transport.harmonic, contactDistribution.
Lemma VII.4.u (real bases and thresholds).
The device strata have bases fixed by the conjugation, and the basis in degree 0 contains the unit. Proof: real and imaginary parts span, and a real independent set extends to a basis.
Given \(f_n > 0\), a valid choice of thresholds is
The thresholds define smooth scalar cutoffs. At a length \(0 < \ell \leqslant t_n\), the cutoff-weighted projection agrees with the original projection through degree \(n\). These cutoffs do not enter \(\mathcal{D}^s_a\).
Lean: real_basis, installs.
Routine but long.
The rate bookkeeping (Lemmas VII.4.a, j).
The \(O (4)\)-covariant Reader (Lemmas VII.4.c, k).
The tree combinatorics (Lemma VII.4.n).
The geometry of normal shells, angular sectors and compact patches.
Main estimates.
The OPE comes from the RG. The basic estimate (Lemmas VII.4.f–g) is a lattice statement: a cluster of renormalized observables equals one observable of degree \(t\) plus a remainder that gains \((a_m / a_k)^{t + 1 / 16}\), at a cost \((a_m / a_j)^{4 |I|}\). The margin \(\tfrac{1}{16}\) is set by the transport gain \((a_k / a_m)^{1 / 8}\) of Part III's RG, not by the contraction \((a_m / a_k)^{t + 1}\) of the irrelevant part. This is where Part III enters.
Uniformity in states and spectators comes from the C*-norm. The norm of \(\mathcal{A}\) is computed by chessboard bounds on decks of translated copies, together with Gelfand and the full support of \(\nu\) (Lemmas VII.4.d–e). The spectators then cost only their own one-field bounds, at their own scale, and the bound holds in every phase at once.
The derivative bootstrap (Lemma VII.4.i). Each derivative costs one power of the fine spacing, so a first-order comparison with the grid gains nothing. The induction therefore starts from high degree, where the crude bound is already strong enough.
One coefficient system for every truncation degree. It comes from a coherent prescription (VII.4.6) and the strata of the Reader.
Unusual, weaker or stronger than expected.
Exponents. The estimates have an arbitrary \(\varepsilon\)-loss and do not determine logarithmic corrections. The degree \(w\) is the sum of the label degrees; the statement does not identify it with a scaling dimension.
Coefficients. They depend on the prescription and the Reader, and nothing relates them to perturbation theory (VII.5, VII.6).
Spectator distance. The remainder bound holds with power-separated spectators, \((r / \rho)^{N + 1}\). The main theorem uses only a fixed clearance.
Contacts. The contact extension is an existence statement. Its order \((2 + w (a)) k\) is crude. Its OPE projection uses the device detectors, with the same prescription as the Reader but separate probe sets.
Interface to Part III. The cluster replacement requires the numerical and geometric hypotheses in Lemma VII.4.f.
| object | code name | proposed notation |
| upper and exact exponential rates | RateLE, RateIs | \(f \preceq e^{\kappa T}\); \(\log f / T \to \kappa\) with \(f > 0\) eventually |
| height of a level sequence | Height | \(m \to \infty\), \(\log (1 / a_m) / T \to c\) |
| cluster, clearance | WithinRoom, ClearOf | \((r, M, \eta)\)-cluster; \(c\)-clear |
| total degree | weight, sigma | \(w (a)\) |
| needles, sheets, tests | Prescription | \(u_i\), \(\lambda_i\), \(\lambda_i^{(n)}\) |
| raw coefficient | Prescription.ope | \(\lambda_i (a ; x, o)\) |
| Reader | Orbits.Reader | \(\mathcal{R}\) |
| lower-degree removal, cut test, stratum | knife, cutTest, stratum | \(\pi_n\), \(\tau_{n, q}\), \(E_n\) |
| channels, actors, coefficients | channels, actor, coef | \(\text{Ch}_j\), \(O_{j, \kappa}\), \(c_{j, \kappa} (a ; x, o)\) |
| ground algebra, states | unselected.algebra, glance, Viewers | \(\mathcal{A}\); \(s \in X \sqcup \{ \text{mix} \}\) |
| error element | algebraError, groundError | \(\mathcal{E}_D (a, x ; b, y)\) |
| chessboard root with interior condition | Cards.PaidAt | \(|F|_{\mathcal{T}_k} \leqslant A\) |
| deck of translates | deck, Episode | \(\prod_{l < 2 n} U^{(\ast)} \circ \tau_{2 tl}\) |
| limit label of a cluster | Survival | \(Q\) |
| coefficient size of a label | quota | \(|z|_m\) |
| RG insertion, step and column bounds | paid, ζ, δ, φ, x, columns, alto, march, gainChart | \(\text{paid}_j\), \(\zeta\), \(\delta\), \(\varphi\), \(x\), \(c_m\), \(A_{k, t}\), \(\mathfrak{m}_{k, m}\), \((a_k / a_m)^{1 / 8}\) |
| cell and stencil fitting data | coverSize, lensScope, GridApprox.tail, clock.J | \(\mathfrak{c}_k\), \(\ell_t\), \(\tau_m\), \(\jmath_m\) |
| RG package | Services | (RG) |
| tree plot, admissibility | Accept, Plot | \((s, z)\) admissible |
| tree costs | Tree.toll, nestPrice | \(\text{toll}_m\), \(\text{nest}_D\) |
| one-field bound, separation | Next.Phase.Single, Gapped | \(|I (v, \cdot) |_{\mathcal{T}_k} \leqslant A\) |
| train, quantum read | ride, QRead | \(v_n\); bench bound \(\rho\) |
| contact devices, clearance | Contacts.Devices, clearance | \(\mathcal{M}\), \(\text{cl} (n)\) |
| cumulant, punctured, full | cumulant, punctured, full | \(C_h\), \(\Pi_h\), \(\text{full}_h\) |
| normal Taylor operator | Normal.taylor | \(T^h_m\) |
| distribution of finite order | Charge, totalTestOrder | \(\mathcal{D}^s_a\), order \((2 + w) k\) |
| contact thresholds | Thresholds | \(t_n\) |
RG inputs below the cluster replacement. Lemmas VII.4.f–g identify the scalar inequalities and the scale induction. The polymer norm estimates inside the one-step representation, fusion and final remainder transport have not been checked line by line. The dependence of their constants, and the complete Part III proofs of these inputs, remain to be reconciled with the stated RG package.
Proofs used through their statements. The full geometry behind the vacuum deck limit and the two-scale continuum transfer belongs to V.1. The rotated comparison estimates used in Lemma VII.4.k, and the triangular replacement and high-horizon lemmas used in VII.4.n, have not been checked internally. Their uses and exponent arithmetic agree with their stated conclusions.
Contact sector induction. The global patching argument, subtraction recursion, shell Taylor bound and scaling normalization have been checked. The finer angular decompositions and the full induction through partially coincident strata remain checked only at the level of their theorem calls. Lemma VII.4.r's radial estimate alone does not settle these partial collisions.
Contact projection versus Reader coefficients. The devices share the Reader's prescription but use separate detectors and a separate graded splitting. An identity between their weight-\(w\) rows and \(\sum_{\kappa} c_{w, \kappa} O_{w, \kappa}\) has not been located. No such identity is used in the contact statement above.
Limits of the conclusions. The results do not assert uniqueness of coefficients or contact extensions, independence of the extraction subsequence, convergence as \(D \to \infty\), reflection positivity or Ward identities for the extensions, or agreement with perturbative coefficients.