1. What this unit proves 2
2. What it uses 11
3. Strategy 13
4. Steps 13
4.1 Numerical existence lemmas 13
4.2 The level schedule 14
4.3 Shooting 15
4.4 Persistent purchases and their composition 17
4.5 The analytic lemmas 20
5. Where the difficulty lies 20
6. Notation 21
7. What remains unexplained 22
The full account of unit III.2: 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: YM/Price/ (the constants and the level schedule, 406 lines), YM/Shoot/ (shooting, 228 lines), YM/Integration/RG05Preconstruction/, YM/Integration/NativeSelection/ and other files of YM/Integration/, YM/RG08/Calendar/, and parts of YM/RG08/, YM/RG12/, YM/FullVolume/ (in particular Roll/) and YM/TraceExtension/. Size: 18,156 lines in the main-theorem closure:
| part | lines |
| proof of the plentiful construction (III.2.5) | about 6,250 |
| proof of the joint same-factor purchase (III.2.4) | about 4,700 |
| proof of the final run (III.2.7) | about 4,000 |
| proof of the sheet and radar purchase | about 2,400 |
| the two directories that carry the mathematics of the schedule (YM/Price/, YM/Shoot/) | 634 |
Conventions. These are those of III.9.
Levels. Levels \(m = 0, 1, \ldots, N\) run from the coarsest (\(m = 0\), unit spacing) to the finest (\(m = N\), the Wilson lattice).
Steps and scales. The step \(m + 1 \to m\) blocks by an odd factor \(L_m \geqslant 3\). The scale is \(s_m = \sum_{i < m} \log L_i\). The level-\(m\) torus at volume index \(p\) has side \(n_p e^{s_m}\), where \(n_p\) is the side of the level-0 torus. III.9 writes \(L_p\) for \(n_p\); the letter \(L\) is kept here for block factors.
Slope. \(\kappa = 11 C_G / (24 \pi^2)\) (III.9.1), and \(\kappa > 0\) for simple \(G\).
Reference quantities (III.9.2). \(\hat{\beta}_m = \bar{\beta} + \kappa s_m\), \(M_m = e^{(\log \log \hat{\beta}_m)^2}\), the threshold \(t_m = \hat{\beta}_m^{- 1 / 2} M_m\), the cover size \(B_m = \lceil (\log \hat{\beta}_m)^2 \rceil\) and the band \(\mathfrak{B}_m = [\hat{\beta}_m / 2, \hspace{0.17em} 2 \hat{\beta}_m]\).
Patch radius. \(R_m = (J_{m + 1} + 3) L_m\) (Patch.radius).
The main results concern numerical choices and a conditional shooting construction:
A long list of requirements can be met simultaneously by choosing the constants in one fixed order. Each requirement has the form "this inequality holds once the block factor, the coupling or the lattice is large enough, given the constants chosen before".
There is a level schedule meeting every entrance condition.
Given the one-step construction as a hypothesis, there is a trajectory of effective actions with \(\beta_{N, p, 0} = \bar{\beta}\) (shooting).
The terminology of sheets, prices and purchases records numerical bounds and their order of choice (glossary, family 1). The unit also contains a group of analytic lemmas of the Gaussian step, filed here by the dependency assignment (III.2.9).
Proposition III.2.1 (the constants, in the order in which they are chosen). Let \(G\) be simple. The construction fixes the following data, in this order.
Sheet (data chosen before the block factors). It consists of the following.
A kernel choice \(K\) with branches on every patch.
For every stage \(j \in \mathbb{N}\), a stage record \(X_j\) with:
a clip with relaxation \(r_j\);
derivative order \(k_j \geqslant 6\) and spatial power \(q_j \geqslant 2\);
support exponent \(a_j > 0\), width \(w_j > 0\) and path count \(\varpi_j > 0\);
drift allowance \(\omega_j \geqslant 0\) and column bound \(\gamma_j \geqslant 1\);
the constraint \(r_j w_j \geqslant 12\).
A first stage \(j_0 \geqslant 4\).
Floors \(f_j \in \mathbb{N}\).
The sequences \(k_j\), \(q_j\), \(a_j\) are nondecreasing.
Stage block factors \(\ell_j\). They are odd, satisfy \(\ell_j \geqslant \max (3, f_j)\), are nondecreasing in \(j\), and for \(j \geqslant j_0\)
Smallness exponents \(\theta_j > 0\). They are nonincreasing, with \(\theta_j \leqslant 1 / j\) for \(j \geqslant j_0\).
Delays. These are the following.
Entry couplings \(\beta^{\text{ent}}_j \in \mathbb{R}\).
Entry scales \(\sigma^{\text{ent}}_j (\bar{\beta})\). These are functions of the stage and of the barrier still to be chosen.
Volume shells \(S_j \in \mathbb{N}\) and growth exponents \(g_j \in \mathbb{N}\).
Delays are ordered componentwise.
Barrier \(\bar{\beta} > 1\). It satisfies \(\beta^{\text{ent}}_{j_0} \leqslant \bar{\beta}\) and \(x^{- 1 / 2} e^{(\log \log x)^2} \leqslant 1\) for all \(x \geqslant \bar{\beta}\).
Wait. This consists of:
readiness scales \(\sigma^{\text{rdy}}_j\);
windows \(h_j : \mathbb{R} \to \mathbb{N}\) with \(h_j (s) / s \to 0\) as \(s \to \infty\) (for each fixed \(j\), not uniformly);
a minimal period \(n_{\min}\).
Waits are ordered componentwise by readiness scales, windows and minimal period.
Items 2, 3 and 5 always exist:
item 2 exists for every sheet;
in item 3, \(\theta\) can be chosen below any prescribed positive sequence;
in item 5, \(\bar{\beta}\) can be chosen above any prescribed real number.
Item 4 has no further conditions. In item 6, only the stated sublinearity is required; the readiness scales are real and \(n_{\min} \in \mathbb{N}\).
Status: items 1–6 are definitions (specifications). There are three existence theorems. The existence of item 2 uses \(\kappa > 0\), hence the simplicity of \(G\). The later existence theorems III.2.4–III.2.7 meet specified requirements in this order. Lean: Sheet, Stage, Large, Rolls, Rolls.exists_rolls, Protection, Protection.exists_margin, Delays, DelayLe, Bar, Bar.exists_above, Wait.
Proposition III.2.2 (a level schedule meeting all entrance conditions). Let \(G\) be simple and let data 1–6 of III.2.1 be given. Then there is a level system with the following data:
a stage index \(J_m\);
step block factors \(L_m = \ell_{J_{m + 1}}\);
level-0 sides \(n_p = 4 (M + p + 1)\) for some \(M \in \mathbb{N}\);
reference couplings \(\hat{\beta}_m = \bar{\beta} + \kappa s_m\).
It satisfies:
(a) identification: the kernel, stage records, \(\ell\), \(\theta\) and \(\bar{\beta}\) are those of III.2.1, and \(J_0 = J_1 = j_0\);
(b) entry: \(\beta^{\text{ent}}_{J_{m + 1}} \leqslant \hat{\beta}_m\) for all \(m\), and \(\sigma^{\text{ent}}_{J_{m + 1}} (\bar{\beta}) \leqslant s_m\) whenever \(J_{m + 1} > j_0\);
(c) volume: \(n_p \geqslant n_{\min}\), and \(S_j (1 + B_m + B_{m + 1})^{g_j} \leqslant n_p e^{s_m}\) for all \(m, p\), with \(j = J_{m + 1}\);
(d) windows: for every \(j \geqslant j_0\) there is a level \(k\) with \(\sigma^{\text{rdy}}_j \leqslant s_k\), \(h_j (s_k) \leqslant k\), and \(J_i = j\) for \(|i - k| \leqslant h_j (s_k)\);
(e) the scale conditions of III.9.2:
\(J\) is nondecreasing, \(J_{m + 1} \leqslant J_m + 1\) and \(J_m \to \infty\);
\(k_{J_m}\), \(q_{J_m}\), \(a_{J_m}\) are nondecreasing in \(m\);
\(\hat{\beta}_m > 1\) and \(t_m \leqslant 1\);
\(100 (B_m + 1) \leqslant n_p e^{s_m}\) and \(100 (R_m + 1) \leqslant n_p e^{s_{m + 1}}\);
\(0 < \theta_{J_m} \leqslant 1 / J_m\), and \(\theta\) is nonincreasing;
\(\omega_{J_{m + 1}} / \log L_m \leqslant J_{m + 1}^{- 2}\) and \(\log \mathcal{C}_{J_{m + 1}} / \log L_m \leqslant J_{m + 1}^{- 2}\);
\(\log L_m / s_m \to 0\);
(f) block control: at every step the kernel is well posed, symmetric under the lattice symmetries and gauge covariant.
The same construction also ensures that every positive integer divides \(n_p\) for infinitely many \(p\).
Status: theorem. The divisibility variant is filed in III.10 and is the one used in the final assembly (III.2.7). Note that (d) asserts the existence of one level \(k\) per stage. Taking \(\sigma^{\text{rdy}}_j\) large forces \(k\) to be large. Lean: Wait.ready, Paid, Calendar.Fits, Levels, Schedule.Gentle, Calendar.blocks_go; variant Wait.engine, AsDivisors.
Proposition III.2.3 (shooting: a selected history from a recipe). Let \(G\) be simple. Fix a level system with
This holds for every level system of III.2.2, by III.2.1(2).
Recipe. Let a recipe be given: path data \(P^{\mathcal{R}}_{N, m}\) and three predicates (a start test, a retention test at each level, a transition test for consecutive levels).
Admissible representations. A level-\(m\) representation \(\rho\) (III.9: path data, clip, activities, budget point, \(\beta\), \(v\)) is admissible for \((N, p)\) if:
its path data are \(P^{\mathcal{R}}_{N, m}\);
its budget point, clip and polymer bounds are those of level \(m\): width \(w_{J_m}\), path count \(\varpi_{J_m}\), loop cap \(2 M_m^{\theta_{J_m}}\), allowances \(\hat{\beta}_m^{- 11 / 10}\) and \(e^{- M_m^{1 / 5}}\);
it passes the retention test.
Continuous families. A family \(x \mapsto \rho_x\) on an interval is continuous if:
\(\beta\) and \(v\) are continuous in \(x\);
the address sets of the regular and forced activities do not depend on \(x\);
differences of regular activities tend to zero in the sum of budget masses of their derivative bounds, up to the budget order on the healthy domain; differences of forced activities tend to zero in the sum of budget masses of their uniform bounds over all configurations.
Hypotheses. Assume that for every \(N\) and \(p\):
(start) There is a continuous family \(\beta \mapsto \rho_{\beta}\), \(\beta \in \mathfrak{B}_N\), of admissible level-\(N\) representations. Each \(\rho_{\beta}\) has coupling \(\beta\), \(v = 0\), density equal to the Wilson weight at \(\beta\), and passes the start test.
(advance) Let \(m < N\) and \(l \leqslant u\), and let \(x \mapsto \rho'_x\), \(x \in [l, u]\), be a continuous family of admissible level-\((m + 1)\) representations with \(\beta (\rho'_x) \in \mathfrak{B}_{m + 1}\) for all \(x\). Then there is a continuous family \(x \mapsto \rho_x\) of admissible level-\(m\) representations such that, for every \(x\):
the density of \(\rho_x\) is the block-spin push-forward of the density of \(\rho'_x\) at coupling \(\beta (\rho'_x)\);
\(| \beta (\rho'_x) - \beta (\rho_x) - \kappa \log L_m | \leqslant \omega_{J_{m + 1}}\);
the transition test holds.
Conclusion. There are representations \(\rho_{N, p, m}\), \(0 \leqslant m \leqslant N\), for all \(N, p\), such that:
every \(\rho_{N, p, m}\) is admissible with \(\beta_{N, p, m} \in \mathfrak{B}_m\);
\(\rho_{N, p, N}\) is the Wilson weight with \(v = 0\);
consecutive levels are related as in (advance);
\(\beta_{N, p, 0} = \bar{\beta}\);
the path data at \((N, m)\) do not depend on \(p\).
The resulting history is selected in the sense of III.9.3. In particular it obeys the running law: for every \(\varepsilon > 0\) there is \(m_0\) such that \(| \beta_{N, p, m} - \hat{\beta}_m | \leqslant \varepsilon s_m\) for all \(m \geqslant m_0\), \(N \geqslant m\) and all \(p\).
Status: theorem, conditional on (start) and (advance).
In the main chain both are verified in III.10 (Graces.winnow):
(start) from the Wilson start-up curves of III.9 (hops);
(advance) from the one-step construction with coupling reset (Mallow.faces).
The drift hypothesis is derived from III.2.2 by Drifting.pay.
A second route, through the step program of III.1/III.9 (Directed.Program), is not used by the main theorem.
Lean: Recipe.races, Recipe, Lives, Curve, Recipe.Relays, Recipe.Starter, Recipe.Issues, Drifting, History.Selected.
Proposition III.2.4 (the joint purchase on one emitted factor).
Hypotheses. Assume:
\(G\) is simple;
\(K\) is a kernel choice with branches at every patch;
order data \(v\) are given:
a first stage \(\geqslant 4\);
derivative orders \(k_j \geqslant \max (6, j)\) and powers \(q_j \geqslant \max (2, j)\), both nondecreasing;
support exponents \(a_j\), nondecreasing and \(\geqslant 100\);
\(Q : \mathbb{N} \to \mathbb{N}\) is nondecreasing with \(Q \geqslant 1\);
the records \(\Omega\) (Contemple) and \(\Delta\) (Decal) are given, together with parameters \((\xi, \omega_v, \omega_t)\) and a degree-\(300\) interpolation tableau \(\theta_1\) at parameter \(\xi\) (Tableau), and the contraction condition sphere at these parameters; in particular \(\xi \geqslant 2\), \(0 < \omega_v \leqslant 1\) and \(\omega_t > 0\);
\(\pi\) is a predicate on emitted factors (a prediction) that is prepaid at every sheet and price map.
Conclusion. There exist:
a sheet \(s\) with kernel \(K\) and the first stage, orders, powers and support exponents prescribed by \(v\);
a price map \(T\);
a real sequence \(Y\) and a nondecreasing real sequence \(K' \geqslant 1\);
a floor \(\tau \geqslant f^s\).
These are followed by the alternation below:
for every floor sequence \(k \geqslant \tau\) (block factors exist for the sheet \(s_k\), which is \(s\) with floors \(k\)), and every choice of stage block factors \(\ell\) for \(s_k\), there is a wage threshold \(U\) such that for every wage \(w \geqslant U\) and every prescribed postage rate and exchange datum, there are post-\(L\) cost constants \(F\) such that for every positive sequence \(\theta^0\) there is a smallness exponent \(\theta \leqslant \theta^0\) such that for every prescribed menu, auxiliary record and delay \(T_0\) there is a delay \(V \geqslant T_0\) such that for every delay \(V' \geqslant V\), every barrier \(\bar{\beta}\) for \(V'\) and every wait \(W\), there is a wait \(Z \geqslant W\) such that for every wait \(W' \geqslant Z\) and every level system \(a\) that meets the entrance conditions of \(W'\) (III.2.2(a)–(d), with (e)–(f)) and whose periods have the divisibility property, there exist a directed datum \(B\) on \(a\) (kinetic data and a recipe on the level system \(a\)) and an emitted factor \(X\) such that:
\(X\) records all these choices;
\(\pi (X)\) holds;
all requirements listed below hold.
The requirements (with the unit that defines each predicate) are:
| requirement | owner |
| wages, exchanges, menus and auxiliary records equal the prescribed data; postage rates do not exceed the prescribed rates | III.10 (Factor) |
| Degree, Savvy, Roofing, PaysW, Spire, Wild, tower bound Tower.Sav with Canyon, Cadre | III.12 |
| AccordV (with the sequence \(Q\)) | III.11 |
| kinetic data at the top level (AtTop) | III.9 |
| amplification (Amplify) | III.1 |
| support exponents \(\geqslant 100\) at every level | this unit |
| \(\exp ((a_j + 2) (2 \hspace{0.17em} \text{span}_{ij} + 13)^4) \leqslant \ell_i\) for all \(i, j \in \mathbb{N}\) with \(j \leqslant i \leqslant j + 1\) | this unit |
| a bandage \(b\) (Bandage, III.10) with the extras \((\Omega, \Delta, b)\) (Extras) | III.15 |
Here \(\text{span}_{ij} \in \mathbb{N}\) is the factor's transition support parameter; the exponent is given by base and toll.
The bandage also supplies the conditional family package of Unlatched with a vacuous additional conclusion; the family hypotheses are those stated in III.2.5.
Status: theorem. The estimates hold on the same emitted factor, with the displayed order of choice. The hypotheses \(\Omega\), \(\Delta\), \(\theta_1\), \((\xi, \omega_v, \omega_t)\) and \(v\) are constructed in the final assembly III.2.7, or supplied there by III.14. The version without a prediction is richFares. Lean: richFaresWithPrediction, built on joinRadarsShipWithPrediction and joinExpensesShipWithPrediction.
Proposition III.2.5 (the plentiful construction). Assume the hypotheses of III.2.4 and \(Q_i \geqslant 2\). Then its conclusion holds with the following additional clause. Consider any trajectory compatible with the emitted factor, any reference levels \(t (d)\) with \(d + 3 \leqslant J_{t (d)}\), any section-transfer data and any gauge-door data between consecutive levels. Assume that the resulting family satisfies the family-consistency conditions, that its gauge doors satisfy the step compatibility conditions, and that its section transfers satisfy the prescribed bounds from level \(t (d)\) whenever \(t (d) \leqslant N\). Then, for every extraction order \(f\), the family satisfies the complete weak-continuum requirement (CompleteWeak, III.11) and the trace gallery (Gallery, VII.2). These are the hypotheses and additional conclusion of Unlatched. The floors, delays and waits may increase, and the prediction is retained. Status: theorem. This is where the requirements of the trace test (Part VII) are bought on the same constants as the ultraviolet estimates. Lean: plentifulWithPrediction, plentiful (vacuous prediction).
Proposition III.2.6 (a prediction on the same factor). Let \(G\) be simple, let \(s\) carry the estimate package of Lemma 4.16 for fixed tariff and scope, and let \(\tau > 0\). Let \(\pi\) be a prediction that is prepaid at the price map of \(s\), in the following sense.
For every choice of targets with the canonical chart, expenses with prescribed deficit \(- 7 / 16\), programs, initial wages and baton, there is a length floor \(u\).
For every floor \(l \geqslant u\), every choice of block factors, wages, duration bounds, register, postage, scouts, protection, step records and post-\(L\) costs, and every delay \(V\), there is a delay \(W \geqslant V\). The step records must have \(\varepsilon \leqslant 1 / 4\) on transitions whose output stage is at least the first stage.
For every delay \(W' \geqslant W\), every barrier, wait, pricing record and directed datum, and every choice of the remaining factor data, \(\pi\) holds on the assembled factor if its shelter and step conditions hold and its periods have the divisibility property.
Then the complete native purchase of Lemma 4.17 can be carried out so that, on level systems with divisible periods, \(\pi\) holds on the factor it emits. The original floor is replaced by its maximum with \(u\). The delay \(W\) is chosen before the original booth and final delay. Status: theorem. In the main-theorem closure it is used only with the vacuous prediction (Always): the only caller of III.2.7 passes it. Lean: company, Prepaid, Prediction, AllPrices.
Proposition III.2.7 (the complete run).
Inputs. Let \(G\) be simple and \(b > 0\) (the kernel penalty). Let the following records be given: \(\Omega\), \(\Delta\) and a custody request \(\mathcal{H}\). Let \(\Lambda\) be a property of level systems that is eventually paid (III.2.8(a)). Let a demand be given:
for every sheet \(s\) and price map \(q\), a nonempty set of choices \(c\);
for each choice \(c\), a floor \(e_c : \mathbb{N} \to \mathbb{N}\);
a real array \(F_c (n, i)\);
a real array \(F'_c [\ell] (n, i)\), which may depend on the block-factor sequence \(\ell\).
Conclusion. There is a run \(\mathcal{R}\) (the record of the whole construction, III.12) with a campaign such that:
(a) the kernel penalty is \(b\);
(b) preselection: there are \(s, q, c\) and a floor \(n \geqslant \max (f^s, e_c)\) such that:
\(\mathcal{R}\) uses the sheet \(s\) with floors \(n\) and the price map derived from \(q\);
both arrays are paid in the sense of III.2.8(b) on the level system of \(\mathcal{R}\), with \(F'\) evaluated at the actual block factors \(\ell^{\mathcal{R}}\);
\(F'_c [\ell^{\mathcal{R}}] (0, J_{m + 1}) \leqslant \hat{\beta}_m\) for all \(m\);
(c) \(k_j \geqslant j\) and \(q_j \geqslant j\) at every stage;
(d) \(\Lambda\) holds;
(e) the treasury of III.2.8(d) is funded;
(f) the common-period purchase is funded;
(g) the campaign has pillars, the funded, news and read-out records hold, and the custody and sky clauses hold.
There is also a version in which any prediction prepaid at every sheet and price map holds on the factor of the run; the intermediate factor theorem below supplies it. Status: theorem. Its only caller in the main closure is III.14 (constructionRun), which passes:
\(\Lambda \equiv\) true;
the demand of its native component recipe: a floor and the pre-\(L\) and post-\(L\) arrays;
concrete \(\Omega, \mathcal{H}, \Delta\);
the vacuous prediction.
Lean: selectRunWithTreasuryPeriod_preselected_quote_orders_postnumeric, PostnumericPreselection, allFactorWithTreasuryPeriod_preselected_quote_orders_postnumeric_of_prediction.
Proposition III.2.8 (purchases that persist under all later choices).
(a) Eventually paid properties. A property \(U\) of level systems is eventually paid if the following holds. For every sheet, block factors, smallness exponent and delay \(V\) there is a delay \(X \geqslant V\) such that every level system meeting the entrance conditions of any delay \(W \geqslant X\) (with any barrier and wait) satisfies \(U\). A conjunction of two eventually paid properties is eventually paid.
(b) Diagonal payment of constants. For every real array \(F (n, i)\), the following property is eventually paid. At every level \(m\), \(1 \leqslant 1 + \log \log \hat{\beta}_m\). For every \(n\) there is \(m_n\) such that
Likewise, for every \(F : \mathbb{N} \to \mathbb{R}\), the property "\(F (J_{m + 1}) \leqslant \hat{\beta}_m\) for all \(m\)" is eventually paid.
(c) Countably many step requests. Fix a countable index set with an encoding in \(\mathbb{N}\), and let \((w_{\alpha})\) be a family of requests indexed by it. Each request consists of a floor \(\text{pre}_{\alpha} : \mathbb{N} \to \mathbb{N}\) and a cost function of a stage transition \((j, i)\), its block geometry and a stage record; at step \(m\) the record is \(X_{J_m}\). Then there is a single floor \(\text{pre}\), depending only on the family, with the following property. If \(\ell_j \geqslant \text{pre} (j)\) for all \(j\), then every delay can be raised so that every later level system satisfying the entrance conditions for the enlarged delay, with any barrier and wait, satisfies, for all \(\alpha\) and \(n\) with \(\text{code} (\alpha) \leqslant J_n\) and all \(m \geqslant n\):
\(\text{pre}_{\alpha} (J_{m + 1}) \leqslant L_m\);
the cost of step \(m\) is at most \(1 + \log (2 + \log \hat{\beta}_m)\).
(d) The treasury. For simple \(G\) and every kernel choice there is a plan. A plan fixes a naming of charges and, for every request, a dictionary of finite words and spectra. If the block factors exceed the plan's floor, then every delay can be raised so that every later level system satisfying its entrance conditions is treasury funded: it meets the step requests of the plan (as in (c)) and the inverse charge from level 0 on.
Status: theorems. Dictionaries are chosen once, before the block factors, and never reselected. Lean: EventuallyPaid, postnumeric_quote_diagonal_pair_requests, eventuallyPaid, PaidRequests, StagePaid, minDelay, Need, Affords, Plan, plan_exists, buy, TreasuryFunded.
Proposition III.2.9 (analytic estimates used by the purchases). The row and cross norms are componentwise exponentially weighted \(\ell^1\) bounds on displacement kernels. Multilinear kernels use the corresponding sum over displacements and colour indices. The colour spaces are fixed and finite; constants may depend on them. A cross maps coarse fields to fine fields at block size \(L\). All vanishing orders refer to the Fourier symbol at zero momentum.
Pullback gain. For \(\sigma, \tau > 0\), \(C \in \mathbb{R}\) and \(d \in \mathbb{N}\), there are \(0 < \eta \leqslant \sigma\) and \(K \geqslant 0\) such that, for every positive odd integer \(L\) with \(\tau \leqslant L \eta\), the following holds. If \(\|A\|_{\tau} \leqslant C / L\), \(\| \nabla^s B\|_{\tau} \leqslant C / L^{1 + |s|}\) for every list of coordinate differences with \(|s| \leqslant d\), and \(R\) has \(\|R\|_{\sigma} \leqslant M\) with \(M \geqslant 0\) and vanishes to order \(d\), then \(\|A^{\textsf{T}} RB\|_{\tau} \leqslant KML^{2 - d}\). pullback_gain
Contraction of Hessian differences. For \(\sigma, \tau > 0\) and \(C \in \mathbb{R}\), there is \(N \in \mathbb{N}\) such that, for every positive odd \(L \geqslant N\), crosses \(A, B\) obeying the preceding bounds through order \(3\), and a row \(R\) on one-form components with \(M \geqslant 0\), \(\|R\|_{\sigma} \leqslant M\) and vanishing order \(3\), one has \(\|A^{\textsf{T}} RB\|_{\tau} \leqslant ML^{- 1 / 2}\). large_contraction
Soft transfer of multilinear kernels. For \(r, d \in \mathbb{N}\), \(\sigma, w > 0\) and \(C \geqslant 0\), there are \(0 < \eta \leqslant \sigma\) and \(K \geqslant 0\), fixed before \(L\), with the following property. Let \(F\) be a real \((r + 1)\)-linear coefficient kernel with width \(\sigma\) and bound \(B\), supported on tuples whose first site is the origin, and vanishing to order \(d\) in the independent momenta. Let \(L\) be positive and odd with \((r + 2) w \leqslant L \eta\), and let \(Q\) satisfy \(\| \nabla^i Q\|_{(r + 2) w} \leqslant C / L^{1 + |i|}\) for every list \(|i| \leqslant d\). The transfer through \(r + 1\) copies of \(Q\), followed by symmetrization and recentering, has width \(w\) and bound \(KBL^4 / L^{r + 1 + d}\). flow_soft
Division by differences. Fix a finite coordinate set \(I\), \(n \in \mathbb{N}\) and \(\sigma > 0\). There are \(0 < \eta \leqslant \sigma\) and \(K \geqslant 0\) such that, for every \(A \geqslant 0\) and real kernel \(f : \mathbb{Z}^I \to \mathbb{R}\) with \(\sum_x e^{\sigma |x|_1} |f (x) | \leqslant A\) whose symbol vanishes to order \(n\), there are real kernels \(g_b\), indexed by \(b : \{1, \ldots, n\} \to I\), with \(\sum_x e^{\eta |x|_1} |g_b (x) | \leqslant KA\) and
for every real momentum \(p\). real_difference_division
Odd-part cancellation. Fix a sequence of step scenes whose input coupling tends to infinity, leaves \(M_i\), a cutoff parameter \(\chi\), chart radii satisfying the logarithm-chart conditions, and fluctuation data \(z_i\). Let \(H_i = \frac{1}{2} \log \beta_i^{\text{out}}\) and \(\eta_{\ast, i} = e^{- H_i}\). Assume that the fluctuation bounds, the leaf weights and the third-derivative bounds \(D_i\) have upper exponential rate \(0\) relative to \(H_i\); that eventually each leaf is a logarithmic term given by a masked regular activity whose expression in logarithmic coordinates, multiplied by the square of its threshold, has zero first derivative at \(0\) and derivatives through order \(3\) bounded by \(D_i\) on the unit ball; that both \(z_i\) and \(- z_i\) lie in the healthy domain; and that both cutoffs equal \(1\) near the origin in the two deformation parameters. If \(P_i (z_i ; y)\) denotes the paved expansion, then, for every \(\varepsilon > 0\), eventually
Here an upper rate \(a\) means an eventual bound \(e^{rH_i}\) for every \(r > a\). mirror_price
Uniqueness of local read-out. Fix a representation \(R\), a degree-\(d\) frame and a section readable at the origin for the budget of \(R\). Let \(P, Q\) be coefficient presentations in the prescribed format for the same \(R\): their regular terms have supports containing the origin, vanish at the identity and have zero local degree-\(d\) projection on nonwrapping supports; their nonzero forced terms have supports containing the origin and satisfy the eligibility, support, gauge-invariance and measurability conditions. Suppose they have equal densities for this section and remainder bounds \(a, b\) with the same positive forced-term parameter. For every \(m \in \mathbb{N}\) with \(10 (m (4 B + 2) + 2 B) < n\), their coefficient vectors satisfy
where \(n\) is the torus side, \(B\) and \(a_{\ast}\) are the cover size and support exponent of \(R\), and \(K_{F, H}\) depends on the frame and holonomy tests. unique_part
Further linked estimates concern a post-block Hessian bound (design), extraction of cubic and quartic local coefficients (tanker), and asymptotic bounds for the trace insertion (cloche). Status: theorems, with the hypotheses above. The detailed scene, coefficient-format and norm interfaces are listed among the open points in section 7.
From other units.
I.1: \(\kappa > 0\) for simple \(G\) (slopeConstant_pos).
III.1 (interfaces):
III.9: the reference couplings, threshold and band (III.9.2); the meaning of a selected history (III.9.3); the Wilson start-up.
III.10:
the verification of the shooting hypotheses (Graces.winnow);
the divisible-period schedule (Wait.engine);
the emitted factor and bandage records.
III.9, III.11, III.12, III.15, VII.2: the predicates bought in III.2.4–III.2.5 (table in III.2.4) and their suppliers. This unit proves only that they hold on one common choice of constants. Their analytic content belongs to those units.
III.12: the run record (Run). III.14: the demand passed to III.2.7.
III.3, III.6 (probable): the free-kernel and stationary-branch estimates from which the sheet is built. The two entry points, reapsY and verdant, are themselves filed in this unit.
III.16: strip and Laurent norms, Taylor jets (for III.2.9).
Standard mathematics.
The intermediate value theorem, in a refined form (Lemma 4.11).
Elementary asymptotics: \((\log x)^2 = o (x)\); polynomials in \(\log s\) are \(o (e^s)\).
Diagonal enumeration of countably many requests.
For background on the order of choice and bare-coupling tuning in constructive RG:
Choosing constants in a fixed order: first the constants of the step, then the block factor \(L\) large depending on them, then the coupling large depending on \(L\). See Bałaban's papers on lattice gauge theories (Comm. Math. Phys. 1983–1989) and Bauerschmidt–Brydges–Slade, Introduction to a renormalisation group method (LNM 2242, 2019).
Tuning the bare coupling as a function of the cutoff so that the effective coupling at unit scale is fixed. Compare Gawędzki–Kupiainen on the Gross–Neveu and \(\phi^4_4\) models (Comm. Math. Phys. 99, 102, 1985).
Here the tuning is a nested-interval argument on continuous families (§4.3).
Imported constants and hypotheses.
\(\kappa\) enters only through \(\kappa > 0\) and through the form \(\hat{\beta}_m = \bar{\beta} + \kappa s_m\). The shift of \(\beta\) by \(\kappa \log L\) up to error \(\omega\) is part of the (advance) hypothesis of III.2.3, supplied in III.10.
The hypotheses (start) and (advance) of III.2.3.
In III.2.7, the records \(\Omega\), \(\mathcal{H}\), \(\Delta\) and the demand come from III.14. Inside the assembly, the kernel with penalty \(b\) comes from III.1 (Axial.viable). The tableau and sphere data are constructed in the assembly.
Choose the stage records and floors first, then the block factors \(\ell_j\), post-\(L\) costs and smallness exponents, delays, the barrier \(\bar{\beta}\), and finally the wait. Increasing the appropriate threshold meets each requirement: the block factors dominate the stage constants, entry couplings pay later costs, and the wait provides comparison windows and large tori. Persistent requirements combine by taking maxima of floors and chaining delay and wait increases. This gives common constants for III.2.4–III.2.7.
The stage index \(J_m\) advances by at most one per level. It enters a stage only after its entry thresholds are met and stays there long enough for one prescribed comparison window. At each fixed stage the volume bill is bounded by a polynomial in \(\log s\), hence eventually by \(e^s\). Since \(s_m \geqslant m \log 3\), this pays every later volume demand. One period offset pays the finitely many steps of the first plateau.
For fixed \(N, p\), start with the Wilson family over \([\bar{\beta} + (\kappa / 2) s_N, \bar{\beta} + 2 \kappa s_N] \subset \mathfrak{B}_N\). At each step the drift bound sends the endpoint couplings strictly beyond the endpoints of the next window. A refined intermediate value argument selects a smaller parameter interval on which the new family stays in that window, so the next step applies. At level \(0\) the window is \(\{ \bar{\beta} \}\). Summing the drifts upward from there and using \(\omega_{J_{m + 1}} \leqslant \log L_m / J_{m + 1}^2\) gives the running law uniformly in depth and volume.
In sections 4.1–4.2, \(G\) is simple and the data are those of III.2.1.
Lemma 4.1 (large block factors). For each stage \(i\) there is \(M_i\) such that every odd \(n \geqslant M_i\) satisfies the conditions of III.2.1(2) at stage \(i\). Hence the stage block factors \(\ell_i = 2 \max_{j \leqslant i} M_j + 1\) exist. Sketch: below the first stage only oddness and the size floors are required. At later stages, the remaining conditions have the form \(c / \log n \leqslant \delta\) with \(\delta \in \{\kappa / 4, i^{- 2} \}\) and \(\delta > 0\). Each holds for large \(n\) because \(\log n \to \infty\). The running maximum makes \(\ell\) nondecreasing. Lean: Rolls.eventually_large, Rolls.exists_rolls.
Lemma 4.2 (smallness exponents). For every positive sequence \(\alpha\), the running minimum \(\theta_j = \min_{i \leqslant j} \min ((i + 1)^{- 1}, \alpha_i)\) is positive, nonincreasing, and satisfies \(\theta_j \leqslant \min (\alpha_j, 1 / (j + 1))\). Lean: Protection.exists_margin.
Lemma 4.3 (barrier). For every \(b\), take \(\bar{\beta} = \max (2, x_{\ast}, b, \beta^{\text{ent}}_{j_0})\), where \(x^{- 1 / 2} e^{(\log \log x)^2} \leqslant 1\) for all \(x \geqslant x_{\ast}\). This is a barrier. Sketch: \((\log y)^2 \leqslant \frac{1}{2} y\) for large \(y\), applied with \(y = \log x\). Lean: Bar.exists_above.
Lemma 4.4 (the volume bill). Define
where \(c = \lceil \log^2 (\bar{\beta} + \kappa s) \rceil\) and \(c' = \lceil \log^2 (\bar{\beta} + \kappa (s + \log \ell_j)) \rceil\). For each fixed \(j\), \(\text{bill}_j (s) \leqslant e^s\) for all sufficiently large \(s\). At \(s = s_m\) with \(j = J_{m + 1}\) we have \(c = B_m\), \(c' = B_{m + 1}\) and \((j + 3) \ell_j = R_m\). So the bill is the sum of the three volume demands of III.2.2(c),(e). Sketch: \(c, c' = O ((\log s)^2)\), so the bill is bounded by a polynomial in \(\log s\). Lean: Bar.bill, Bar.allocated.
Lemma 4.5 (entering a stage). For each \(j > j_0\) there is \(n_j\) with the following property. For every level \(m \geqslant n_j\) and every \(s \geqslant m \log 3\):
\(\beta^{\text{ent}}_j \leqslant \bar{\beta} + \kappa s\);
\(\sigma^{\text{ent}}_j (\bar{\beta}) \leqslant s\);
\(s > 0\);
\(\log \ell_j \leqslant s / j\);
\(\text{bill}_j (s) \leqslant e^s\).
Sketch: each condition holds for all large \(s\). Take the largest of the thresholds, and \(n_j\) with \(n_j \log 3\) above it. Lean: Bar.enters.
Lemma 4.6 (windows). For each \(j\) there is \(c_j\) such that, for all \(k \geqslant c_j\) and all \(s\) with \(k \log 3 \leqslant s \leqslant k \log \ell_j\), we have \(\sigma^{\text{rdy}}_j \leqslant s\) and \(h_j (s) \leqslant k / 2\). Sketch: for large \(s\), \(h_j (s) \leqslant s / (2 \log \ell_j)\) by sublinearity, and \(s \leqslant k \log \ell_j\). Lean: Wait.lead.
Lemma 4.7 (exhaustion). For any \(b, c : \mathbb{N} \to \mathbb{N}\) there is \(J' : \mathbb{N} \to \mathbb{N}\) with the following properties:
\(J'_0 = J'_1 = 0\), \(J'\) is nondecreasing, \(J'_{i + 1} \leqslant J'_i + 1\) and \(J' \to \infty\);
\(b (J'_{i + 1}) \leqslant i\) whenever \(J'_{i + 1} > 0\);
for every \(i\) there is \(k \in \mathbb{N}\), \(k \geqslant c (i)\) with \(J' \equiv i\) on \([k - \lfloor k / 2 \rfloor, \hspace{0.17em} k + \lfloor k / 2 \rfloor]\).
Sketch: set \(a_0 = 0\) and \(a_{i + 1} = a_i + b (i + 1) + 2 k_i + 1\), where \(k_i = c (i) + 2 a_i + 1\). Put \(J' = i\) on \([a_i, a_{i + 1})\). Stage \(i + 1\) starts at level \(a_{i + 1} > b (i + 1)\). The window around \(k_i\) lies in \([a_i, a_{i + 1})\) because \(k_i / 2 \geqslant a_i\) and \(\tfrac{3}{2} k_i < a_{i + 1}\). Lean: Program.exhaust.
Lemma 4.8 (the schedule meets all entrance conditions; proof of III.2.2).
Construction.
Apply Lemma 4.7 with \(b (0) = 0\) and \(b (i) = n_{j_0 + i}\) for \(i > 0\) (Lemma 4.5) and \(c (i) = c_{j_0 + i}\) (Lemma 4.6), and set \(J_m = j_0 + J'_m\), \(L_m = \ell_{J_{m + 1}}\) and \(s_m = \sum_{i < m} \log L_i\). Then \(m \log 3 \leqslant s_m \leqslant m \log \ell_{J_m}\), using that \(\ell\) and \(J\) are monotone.
Let \(m_{\ast}\) be such that \(J'_m \geqslant 1\) for \(m \geqslant m_{\ast}\). Put \(M = n_{\min} + \max_{i \leqslant m_{\ast}} \text{bill}_{j_0} (s_i)\) and \(n_p = 4 (M + p + 1)\).
Verification.
Entry. If \(J_{m + 1} = j_0\), then \(\beta^{\text{ent}}_{j_0} \leqslant \bar{\beta} \leqslant \hat{\beta}_m\). Otherwise \(m \geqslant n_{J_{m + 1}}\) and \(s_m \geqslant m \log 3\), so Lemma 4.5 applies.
Volume off the first plateau. There \(\text{bill}_{J_{m + 1}} (s_m) \leqslant e^{s_m} \leqslant n_p e^{s_m}\). This gives (c), the cover condition, and the patch condition, the latter because sides increase with \(m\).
Volume on the first plateau. There \(\text{bill}_{j_0} (s_m) \leqslant M < n_p\).
Mesh. For \(m \geqslant m_{\ast}\), Lemma 4.5 gives \(\log L_m / s_m \leqslant 1 / J_{m + 1} \to 0\).
Windows. At the level \(k\) given by Lemma 4.7 for stage \(j = j_0 + i\), \(s_k\) lies in the range of Lemma 4.6. So \(h_j (s_k) \leqslant k / 2\), and \(J \equiv j\) on \([k - h_j (s_k), k + h_j (s_k)]\).
Remaining conditions. Monotonicity of \(k_{J_m}, q_{J_m}, a_{J_m}\) comes from the sheet. The drift and log-cost bounds come from Lemma 4.1, and the bounds on \(\theta\) from Lemma 4.2. Block control comes from the kernel's branches.
Divisibility variant. Given \(c \geqslant 1\) and \(z\), the index \(p = c (M + z + 1) - (M + 1) \geqslant z\) gives \(n_p = 4 c (M + z + 1)\).
Lean: Wait.ready, Program.book, scaled_lower, scaled_upper, Wait.engine.
Lemma 4.9 (running law from the transfer inequality). Consider a history on a level system of III.2.2 with \(\beta_{N, p, 0} = \bar{\beta}\) and \(| \beta_{N, p, i + 1} - \beta_{N, p, i} - \kappa \log L_i | \leqslant \omega_{J_{i + 1}}\) for all \(i < N\). Then it obeys the running law of III.2.3. Sketch:
Since \(\hat{\beta}_{i + 1} - \hat{\beta}_i = \kappa \log L_i\) and \(\beta_{N, p, 0} = \hat{\beta}_0\), induction upward from level 0 gives \(| \beta_{N, p, i} - \hat{\beta}_i | \leqslant D_i := \sum_{k = 1}^i \omega_{J_k}\).
Given \(\varepsilon > 0\), choose \(i_0\) with \(J_{i + 1}^{- 2} \leqslant \varepsilon / 2\) for \(i \geqslant i_0\). Then \(\omega_{J_{i + 1}} \leqslant \frac{\varepsilon}{2} \log L_i\), so \(D_i \leqslant D_{i_0} + \frac{\varepsilon}{2} (s_i - s_{i_0})\).
Finally \(D_{i_0} \leqslant \frac{\varepsilon}{2} s_i\) for large \(i\), because \(s_i \to \infty\).
The bound is uniform in \(N \geqslant i\) and \(p\). Lean: trajectory.
Lemma 4.10 (gates). Put \(\beta^-_m = \bar{\beta} + \frac{\kappa}{2} s_m\) and \(\beta^+_m = \bar{\beta} + 2 \kappa s_m\). Then \([\beta^-_m, \beta^+_m] \subset \mathfrak{B}_m\), because \(\bar{\beta} > 0\). Suppose a level-\(m\) representation \(\rho\) relays a level-\((m + 1)\) representation \(\rho'\), and \(\omega_{J_{m + 1}} \leqslant \frac{\kappa}{4} \log L_m\). Then:
if \(\beta (\rho') = \beta^-_{m + 1}\), then \(\beta (\rho) \leqslant \beta^-_m - \frac{\kappa}{4} \log L_m < \beta^-_m\);
if \(\beta (\rho') = \beta^+_{m + 1}\), then \(\beta (\rho) \geqslant \beta^+_m + \frac{3 \kappa}{4} \log L_m > \beta^+_m\).
The drift hypothesis follows from the entrance conditions and III.2.1(2), since \(L_m = \ell_{J_{m + 1}}\). Lean: floorBeta, ceilBeta, limits, gates, Drifting.pay.
Lemma 4.11 (narrowing). Let \(x \leqslant y\) and let \(f\) be continuous on \([x, y]\) with \(f (x) \leqslant d \leqslant u \leqslant f (y)\). Then there are \(x \leqslant z \leqslant w \leqslant y\) with \(f (z) = d\), \(f (w) = u\) and \(f ([z, w]) \subset [d, u]\). Sketch: let \(w\) be the least point of \(f^{- 1} (u) \cap [x, y]\) and \(z\) the greatest point of \(f^{- 1} (d) \cap [x, w]\); both sets are nonempty by the intermediate value theorem and compact by continuity. Suppose some \(c \in [z, w]\) has \(f (c) < d\). Then the intermediate value theorem on \([c, w]\) gives a point of \(f^{- 1} (d)\) after \(z\), a contradiction. Suppose instead \(f (c) > u\). Then it gives a point of \(f^{- 1} (u)\) before \(w\), again a contradiction. Lean: narrow.
Lemma 4.12 (nested intervals). Fix \(N, p\) and assume (start), (advance) and the drift bound. One can choose intervals \(I_m = [l_m, u_m]\), \(0 \leqslant m \leqslant N\), with \(I_N = [\beta^-_N, \beta^+_N]\) and \(I_m \subset I_{m + 1}\) for \(m < N\), and partial trajectories \((\rho^x_i)_{m \leqslant i \leqslant N}\) for \(x \in [l_m, u_m]\), such that:
\(\rho^x_N\) is the start representation at bare coupling \(x\);
every \(\rho^x_i\) is admissible with \(\beta (\rho^x_i) \in \mathfrak{B}_i\);
consecutive levels relay;
\(x \mapsto \rho^x_m\) is continuous;
\(\beta (\rho^{l_m}_m) = \beta^-_m\) and \(\beta (\rho^{u_m}_m) = \beta^+_m\).
Sketch: downward induction.
For \(m = N\), restrict the start family to \([\beta^-_N, \beta^+_N]\).
For \(m + 1 \to m\), the family at level \(m + 1\) has all couplings in \(\mathfrak{B}_{m + 1}\), so (advance) yields a continuous family \(\rho^x_m\). By Lemma 4.10, \(f (x) = \beta (\rho^x_m)\) satisfies \(f (l_{m + 1}) < \beta^-_m\) and \(f (u_{m + 1}) > \beta^+_m\).
Lemma 4.11 gives \([l_m, u_m]\) on which \(f\) maps onto \([\beta^-_m, \beta^+_m] \subset \mathfrak{B}_m\) with the right endpoint values.
The restriction is needed: (advance) applies only to families whose couplings lie in the band at every point. Only the continuity of \(x \mapsto \beta (\rho^x_m)\) enters the intermediate value argument. The other continuity clauses are carried because (advance) requires them at the next step. Lean: steps.
Lemma 4.13 (endpoint and assembly; proof of III.2.3). Since \(s_0 = 0\), we have \(\beta^-_0 = \beta^+_0 = \bar{\beta}\). So \(x = l_0\) gives a trajectory from level \(N\) to level 0 with \(\beta_{N, p, 0} = \bar{\beta}\). Choose one such trajectory for every \((N, p)\). Then:
the path data are \(P^{\mathcal{R}}_{N, m}\), independent of \(p\);
admissibility gives the budget, clip, polymer bounds and band membership;
(start) gives the Wilson weight with \(v = 0\);
the relays give the transfer identity and the drift inequality;
Lemma 4.9 gives the running law.
Hence the history is selected. Lean: successful, Competes, Recipe.races.
Lemma 4.14 (composition rules). Persistent numerical requirements can be combined in the following forms:
(i) a floor \(f\) such that the requirement holds for every later block factor;
(ii) "for every delay \(V\) there is \(V' \geqslant V\) such that every later delay, every barrier and wait, and every level system meeting those entrance conditions satisfy \(P\)";
(iii) the same with waits instead of delays.
Floors combine by pointwise maximum. Suppliers of forms (ii) and (iii) combine by chaining, because the orders on delays and waits are transitive and the conclusions persist under raising. Lean: postnumeric_quote_diagonal_pair_requests, and the chains in joinExpensesShipWithPrediction.
Lemma 4.15 (diagonal payments; proof of III.2.8).
For (b), raise \(\beta^{\text{ent}}_j\) to \(\max \left( \beta^{\text{ent}}_j, \exp \exp (1 + \sum_{n, i \leqslant j} |F (n, i) |) \right)\). The entrance condition \(\beta^{\text{ent}}_{J_{m + 1}} \leqslant \hat{\beta}_m\) then bounds the partial sums at stage \(J_{m + 1}\) by \(\log \log \hat{\beta}_m\). Since \(J \to \infty\), each fixed \(n\) satisfies \(n \leqslant J_{m + 1}\) from some level on.
For the single-array form, raise \(\beta^{\text{ent}}_j\) to \(\max (\beta^{\text{ent}}_j, F (j))\).
For (c), encode the requests. Take \(\text{pre} (j) = \max_{\text{code} (\alpha) \leqslant j} \text{pre}_{\alpha} (j)\). Raise \(\beta^{\text{ent}}_j\) until the finitely many costs with \(\text{code} (\alpha) \leqslant j\), evaluated on the stage-\(j\) geometry, are below \(1 + \log (2 + \log x)\) for \(x \geqslant \beta^{\text{ent}}_j\).
For (d), fix the plan before the block factors, apply (c) to its requests, and raise the delay once more for the inverse charge, whose block floor is zero.
Lean: eventuallyPaid, minDelay, buy.
Lemma 4.16 (the sheet). Let \(G\) be simple, let \(K\) have branches at every patch, let the order data \(v\) satisfy the order conditions of III.2.4, and fix a sequence \(C_j \geqslant 1\). There are a sheet \(s\) with kernel \(K\) and the prescribed first stage, orders, powers and support exponents; two price maps \(O, T\); real sequences \(\lambda, \gamma\) with \(\gamma_j > 0\); and a nondecreasing sequence \(K'_j \geqslant 1\). For every tariff and scope there is a floor \(f \geqslant f^s\) with the following properties.
The sheet satisfies the stationary-path suitability conditions at stages \(j \geqslant j_0\) and the budget plans for \(C, K'\). For every floor \(n \geqslant f\) and every admissible block-factor sequence \(\ell\), there is a sequence \(\eta\) satisfying the stationary budget-gap condition. For every post-\(L\) cost record \(F\) with \(\eta_F = \eta\), \(\Gamma_F \geqslant 1\) and \(c_F \leqslant \gamma\), every protection, additional tariff, nonnegative lower bound \(z\) and delay \(A\), there are a delay \(D \geqslant A\), a transition-bound array bounded below by \(z\), and two further transition arrays. Every later delay \(E \geqslant D\) satisfies the pricing, ringing and comparison conditions. For every barrier, wait and level system meeting those entrance conditions, the budget and multisource conditions hold, and there are a Gaussian horizon and kinetic data with planned stationary paths, the Wilson top path, the cadre and amplification conditions, and the retained free and stationary data specified by the perceived-estimate package.
Sketch: the free-kernel and stationary suppliers first fix widths and preliminary bounds. The sheet construction fixes clips and the prescribed orders; the startup suppliers enlarge path bounds and floors. The final profile construction raises drift allowances to dominate the stationary supplier's bounds and raises column bounds without changing widths, paths or the prescribed orders. After choosing \(\ell\), the stationary and multisource suppliers raise the delay and construct the kinetic data on each admitted level system. The retained data identify these kinetic paths with the revised stationary paths. Lean: sailsW, Perceived. (proof read at the level of the composition)
Lemma 4.17 (the native purchase and prediction). Let \(G\) be simple, fix the perceived-estimate package of Lemma 4.16 at a floor \(v\), and let \(\tau > 0\). There are a floor \(j \geqslant v\), programs, initial wages, targets, expenses and a baton, chosen before the block factors. The targets have the canonical chart and halo \(500\); the expenses have the prescribed earnings, deficit \(- 7 / 16\) and degree charge \(0\).
For every floor \(l \geqslant j\), admissible block factors exist. For every such choice there is a wage threshold. For every wage above it there are duration bounds, a register and auxiliary size bounds. For every postage cap there are postage rates at most that cap and scouts satisfying the register condition. For every exchange datum and tariff there are reset prices, reset bounds and post-\(L\) costs. For every positive sequence \(\theta^0\) there are protection \(\theta \leqslant \theta^0\) and step records with \(\varepsilon \leqslant 1 / 4\) on transitions whose output stage is at least the first stage. For every menu and initial delay there are booths; for every auxiliary record there is an ordered delay above the initial delay.
Every later delay, barrier, wait and admitted level system satisfies the budget and multisource conditions and carries a directed datum on that same system. Its shelter conditions hold at the two pairs of tolerances \((1 / 16, 1 / 16)\) and \((\tau, \tau)\), with the third tolerance \(1 / 8\) in both cases. It also has the top kinetic data, cadre, amplification, step conditions, origins and rank-covariance data. The offering condition additionally assumes divisible periods.
Sketch: the shelter supplier gives the initial floor and data. A further floor pays the register's length requirement; a delay supplier supplies the step conditions before the booths are chosen. The shelter output then supplies the remaining data. For III.2.6, combine the prediction's floor \(u\) with \(j\) by maximum. Insert its delay before the booth delay; persistence then gives the prediction on the same emitted factor whenever the divisibility condition holds. Lean: companyK (proof read at the level of the composition); company.
Lemma 4.18 (proof of III.2.4 and III.2.5). The proofs proceed in four steps.
joinRadarsShipWithPrediction takes the sheet from Lemma 4.16 and calls joinExpensesShipWithPrediction. That theorem starts from Lemma 4.17 and threads about thirty further suppliers through it:
floor suppliers, joined into one floor;
delay suppliers \(V \leqslant V_1 \leqslant \ldots \leqslant V_7 \leqslant \ldots \leqslant w_6 \leqslant \ldots\), chained.
richFaresWithPrediction adds:
additional costs and floors before choosing the block factors;
a cross-term floor, so the final floor is \((t \vee L \vee \mu) \vee\) cross-floor;
shells raised to \(\max (S_j, \cdot)\), the mesh and installment delays, and a wait enlargement.
plentifulWithPrediction adds:
trace floors and the fixed trace-control floor;
trace delays, including a raise of the entry couplings;
two wait enlargements.
It concludes with Trace.Over.gang.
Floors are combined by maximum and delay and wait bounds by transitivity, as in Lemma 4.14. The final family conclusion uses the hypotheses of III.2.5.
(proof read at the level of the composition; the suppliers were not read)
Lemma 4.19 (the final assembly; proof of III.2.7). allFactor… makes the choices in order:
Pre-\(L\) data.
Order data: \(j_0 = 4\), \(k_j = 6 + j\), \(q_j = 2 + j\), \(a_j = 100\); and \(Q_i = i + 2\).
A kernel with penalty \(b\).
A tableau at a parameter \(\xi \geqslant 2\) and sphere data with \(\omega_v = 1 / 1024\) and \(0 < \omega_t \leqslant 1\). The contraction supplier chooses \(\omega_t\); its auxiliary constant \(u \geqslant 0\) satisfies \(6 u \omega_t \leqslant \omega_v\).
The plentiful construction. Apply III.2.5. Pick a choice \(c\) of the demand.
Block floor and block factors. The block floor is the maximum of five floors: the plentiful floor, a density floor, the treasury floor, the period-plan floor and \(e_c\). Fix block factors, wages and post-\(L\) costs.
Smallness exponent and delays. Take \(\theta^0 \equiv 1\) and initial data \(\beta^{\text{ent}} = \sigma^{\text{ent}} = 0\), \(S_j = g_j = 1\). Raise delays in this order: halo, step, thermal requests, \(\Lambda\) and both demand arrays together with the all-stage payment of the zeroth post-\(L\) row, treasury, period plan, auxiliary budget, silk, custody, the delay of III.2.5, the quote bounds, and the header bounds.
Barrier. Take \(\bar{\beta} \geqslant 2\) (Lemma 4.3).
Wait. First apply the wait supplier of III.2.5 to a zero-window wait. Separately enlarge a slim wait for density bounds. Choose a common enlargement satisfying both waits and the required coverage bounds.
Level system. Take the level system of the divisibility variant of III.2.2, and the factor from III.2.5.
Finally engageWithCustodyFactor builds the run from the factor. (proof read at the level of the composition)
Lemma 4.20 (mechanisms behind III.2.9). Under the hypotheses of III.2.9, the estimates follow as follows.
Division: analytic division first gives complex coefficient packets on a smaller strip. Taking their real parts and using conjugation symmetry preserves the identity at real momenta and gives the stated real kernels.
Pullback gain: divide \(R\) into terms with \(d\) forward differences and move those differences onto \(B\). The two cross bounds contribute \(L^{- 2}\), the sum over fine sites contributes \(L^4\), and the differences contribute \(L^{- d}\), giving \(L^{2 - d}\).
Contraction: apply the gain with \(d = 3\) and choose \(N > \max (\tau / \eta, K^2)\). Then \(K / L \leqslant L^{- 1 / 2}\) for \(L \geqslant N\).
Multilinear transfer: divide in the independent momenta and distribute the \(d\) differences among the \(r + 1\) crosses. Their bounds contribute \(L^{- (r + 1 + d)}\); summing the free translation gives \(L^4\). The width reserve pays for symmetrization and recentering.
Odd-part cancellation: reflection cancels the first odd contribution. The rate calculus controls the remaining second derivative with every exponent strictly below \(2\). The second derivative of a function composed with \(y \mapsto - y\) equals its original second derivative at \(0\).
Local read-out: subtract the presentations. Equality of densities expresses the coefficient difference as the difference of the regular remainders on the healthy domain. The local projection kills nonwrapping supports. A wrapping support has an exponentially small contribution, bounded using the side condition; finitely many test functionals then control the coefficient norm.
(proofs read at the level of the displayed declarations; the analytic division, transfer and local-projection inputs are not independently audited)
Order and persistence. Each floor must be fixed before the block factors that pay it. Costs depending on the chosen factors can be paid by later entry thresholds. Delay and wait suppliers must remain valid under all subsequent increases; otherwise chaining them does not give a common set of constants.
Uniformity in depth and volume. The constants and schedule are common to every \(N, p\). The finite first plateau is handled by one period offset \(M\), giving \(n_p = 4 (M + p + 1)\). Later volume conditions follow from the exponential growth of the side length.
Continuity during shooting. Each new interval must keep every member of the family inside the next coupling band. The nested-interval argument preserves this condition while fixing the endpoint couplings. It requires continuity of the full representation family at each new step.
Two drift bounds. The bound \(\omega_j \leqslant (\kappa / 4) \log \ell_j\) gives the shooting gates. The bound \(\omega_j \leqslant \log \ell_j / j^2\) gives the running law. Both are paid after the stage constants are fixed.
Eventual payment. The entry thresholds can absorb any fixed real array, including one evaluated after the block factors are chosen. Each fixed row is paid eventually; a single level need not pay every row. The stronger all-stage condition for the zeroth post-\(L\) row is purchased separately.
| Written proof | Lean | Meaning |
| sheet; stage record \(X_j\); first stage \(j_0\); floors \(f_j\) | Sheet, Stage | data fixed before the block factors |
| \(k_j, q_j, a_j, w_j, \varpi_j, \omega_j, \gamma_j, r_j\) | order, power, supportExponent, width, paths, drift, columns, clip.relaxation | fields of the stage record |
| \(\mathcal{C}_j\) | (in Large) | \(1 + \varpi_j + w_j^{- 1} + \gamma_j + r_j + a_j\) |
| \(\ell_j\); \(L_m = \ell_{J_{m + 1}}\) | Rolls; Schedule.block | stage and step block factors |
| \(\theta_j\) | Protection | smallness exponent of the loop cap |
| \(\beta^{\text{ent}}_j, \sigma^{\text{ent}}_j (\bar{\beta}), S_j, g_j\) | Delays | entry thresholds, shells, growth |
| \(\bar{\beta}\) | Bar | barrier: coupling at level 0 |
| \(\sigma^{\text{rdy}}_j, h_j, n_{\min}\) | Wait | readiness, windows, minimal period |
| meets the entrance conditions | Paid | III.2.2(a)–(d) |
| \(\text{bill}_j (s)\) | Bar.bill | required sites per unit length |
| \(J'\), \(J_m = j_0 + J'_m\) | Program | stage index |
| \(n_p = 4 (M + p + 1)\) | Program.page | level-0 side, with \(M \in \mathbb{N}\) |
| recipe \(\mathcal{R}\) | Recipe | path data and three tests |
| admissible; relays; start | Lives, Relays, Starter | |
| continuous family | Curve, deviation | |
| (start)+(advance) | Recipe.Issues | shooting hypotheses |
| \(\mathfrak{B}_m\); \([\beta^-_m, \beta^+_m]\) | band; floorBeta, ceilBeta | band; shooting window |
| eventually paid | EventuallyPaid | III.2.8(a) |
| plan; treasury funded | Plan, TreasuryFunded | III.2.8(d) |
| emitted factor; prediction | Factor, Prediction | record of all choices; predicate on it |
| sheet with floors \(n\) | raise, bountyV | |
| order data \(v\) | Orders, Draft, WithStage |
Analytic content of the purchased conditions. The common-choice statements III.2.4–III.2.7 and their quantifier order are verified against the displayed Lean interfaces. The predicates in the requirement table, the conditional family package, and the campaign's pillars, funding, news, custody and sky conditions are not fully expanded into mathematical estimates here. Their equivalence to the requirements stated in the owning units remains to be checked.
Underlying suppliers. The proofs composing the sheet, native, joint, plentiful and final-run constructions have been checked at their call sites. The proofs of the underlying analytic suppliers, including those used by the roughly thirty additions in the joint construction, are not independently verified here. The same limitation applies to the plan's dictionaries and inverse-charge supplier, and to the analytic division and projection results used in III.2.9.
Auxiliary parameters. \(K'\) is a nondecreasing sequence bounded below by \(1\); the tableau is a finite interpolation rule of degree \(300\); the sphere condition is a contraction estimate. The detailed analytic roles of \(Y\), \(Q\), wages, postage, exchanges, menus and the remaining auxiliary records still require identification in the step estimates.
Drift normalization. The sheet construction raises the drift allowance to dominate the stationary supplier's sequence. Its identification with the bound \(1 + D\) used in III.9, including the relation between the constants attributed there to III.7 and III.13, remains unverified.
Analytic interfaces. The weighted summability and vanishing-order definitions, the first-site support condition in multilinear transfer, and the rate interpretation of odd-part cancellation have been checked. The full step-scene hypotheses of odd-part cancellation and the coefficient-format and readability conditions of local read-out still need an independent mathematical account. Only the statements of the supplementary Hessian, cubic/quartic extraction and trace-insertion estimates linked after III.2.9 were checked; their proofs remain outside this review.
Scope of the source review. The arithmetic construction, the divisibility argument, and the shooting proof are verified against the source. Other uses of the non-divisible period construction, the allocation of the supplementary analytic estimates to outline units, and the line-count breakdown given at the outset have not been independently checked.