Home·The theorem·The proof in one reading·Architecture·Units·Manuscript·Glossary·FAQ·Work items·Data·Method·Provenance·Changes
← III.6 The stationary branch: background field of the effective action (notes)Part III·Unit outlineIII.8 Connected activities and backward dependence cones (notes) →

III.7. Matching the Wilson energy to fourth order — detailed notes

Contents

1. What this unit proves 2

2. What it uses 9

3. Strategy 10

4. Steps 12

4.1 Fourier kernels of local functionals 12

4.2 The lattice Poincaré lemma in momentum space 12

4.3 Classical emission and the Wilson jet 13

4.4 Two-leg kernels: the \(F^2\)-coefficient and irrelevance 14

4.5 The one-loop term in covariant form 15

4.6 Uniform invertibility of the fluctuation operators 15

4.7 The slope of the regulated determinant 16

4.8 From the regulated determinant to the blocked determinant 18

4.9 The step and the iteration 19

5. Where the difficulty lies 21

6. Notation 21

7. What remains unexplained 23

The full account of unit III.7: 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/Next/STATIONARY/, YM/WilsonBulk/Spectral/, YM/UVExtension/Projection/, and a few lines of YM/UVExtension/Reward/. Size: 19,945 lines in the main-theorem closure: 19,391 in Next/STATIONARY, 286 in UVExtension/Projection, 215 in WilsonBulk/Spectral, and about 50 elsewhere.

Conventions. Levels \(m = 0, \ldots, N\) run from the coarsest level (\(m = 0\)) to the Wilson lattice (\(m = N\)). Blocking from fine level \(m + 1\) to coarse level \(m\) uses an odd factor \(L_m\). The stage \(J_m\) is the index of the constants used at level \(m\); it satisfies \(J_{m + 1} \in \{J_m, J_m + 1\}\). In one step the fine torus \(\Lambda_f\) has side \(f = Lc\) and the coarse torus \(\Lambda_c\) has side \(c\); in general \(| \Lambda_n | = n^4\). Fields are Lie-algebra link variables \(x\), and configurations are \(U = e^x\). \(E_W (U) = \sum_p \left( 1 - d_{\text{rep}}^{- 1} \text{Re} \hspace{0.17em} \text{tr} \hspace{0.17em} U_p \right)\) is the Wilson energy in the fixed matrix representation of dimension \(d_{\text{rep}}\), \(C_G\) is the adjoint Casimir constant of I.1, and \(\kappa = 11 C_G / (24 \pi^2)\).

1. What this unit proves

At level \(m\) the effective action has the form \(\beta \hspace{0.17em} \text{main}_{\mathbf{c}, R} (U) + \text{loop}_R (U)\), plus polymer activities (III.10). Here \(R\) is a path datum: an exponentially weighted, explicit sum of gauge-covariant curvature monomials along lattice paths. This unit constructs the path data and the coupling shifts.

Iterated over all levels, (A)–(C) give the kinetic data satisfying the conditions required in III.9. The slope \(\kappa\), a defined constant in I.1 and III.9.1, is identified here with the one-loop coefficient of the blocked action.

Proposition III.7.1 (path data, beds, Fourier kernels).

\(\displaystyle \text{main}_{\mathbf{c}, R} (U) = \frac{1}{384} \sum_s \left[ \sum_p \left( c_{\text{dir}} \left( 1 - d_{\text{rep}}^{- 1} \text{Re} \hspace{0.17em} \text{tr} \hspace{0.17em} U_p \right) + \tfrac{1}{2} | \ell_R (U)_p |^2 \right) + \sum_x \sum_k (\mathcal{A}^{(3)}_k +\mathcal{A}^{(4)}_k) (U, x) \right]_{U \to sU} .\)

This is an average over the 384 signed permutations \(s\) of the axes. \(\text{loop}_R\) is the same average of the sum of all loop atoms. These are the unclipped analytic germs; the small-field clipping of the full action does not enter them.

Status: definitions. Path data, weight clauses, main term, beds and Floor are specifications of III.1; the Fourier notions are defined in this unit. Lean: RG.PathData, RG.Slot, RG.Atom, PathData.Bounds, metric_zero, loop_jets, RG.Clip, Precision.mainTerm, Precision.Bed, Attached, Floor, Firm, Rays, Recorded, alphaPay, tip_fuel.

Proposition III.7.2 (the Wilson jet at the finest level). Let \(G\) be a gauge group, \(\sigma_0 > 0\) and \(A, Z \geqslant 0\). There are \(\sigma \in (0, \sigma_0]\) and \(C \geqslant 0\) with the following property.

Hypotheses: sides \(n_i \to \infty\); a clip with \(c_{\text{dir}} \leqslant Z\); a bed \(B\) such that

Conclusion: there is a path datum \(Q\) such that:

  1. the rows of \(Q\) are the entries of \(\rho\) placed on straight lattice routes, \(Q\) is attached to \(B\), and all loop coefficients vanish;

  2. \(Q \in \text{Bounds} (\mathbf{c}; \tau, k, \Lambda)\) for all \(0 < \tau \leqslant \sigma\), all \(k \geqslant \max (C, 6 Ae^{4 \sigma_0})\) and all \(\Lambda \geqslant 0\);

  3. for every \(i\) and every \(j \leqslant 4\), the functions \(x \mapsto \text{main}_{\mathbf{c}, Q} (e^x)\) and \(x \mapsto E_W (e^x)\) on \(\Lambda_{n_i}\) are real-analytic at \(0\) and have equal \(j\)-th derivatives there.

Status: theorem. Lean: Stationary.launch, rowPath, rowPrice, SameJet.

All stages and all volumes.

Proposition III.7.3 (the one-loop slope). Let \(G\) be simple. For \(\sigma, \varepsilon > 0\), \(C, Z \geqslant 0\), \(\tau > 0\) and \(A \geqslant 0\) there is \(D \geqslant 0\) with the following property.

Hypotheses:

For a link field \(x\) on \(\Lambda_{n_i}\) and \(U = e^x\) put

\(\displaystyle \Pi_{\mu} (x) = \tfrac{1}{2} \log \det (\mathcal{H}_{\mathbf{c}, R} (U) + d_U d_U^{\ast} +\mathcal{M}_{\mu} (U)) - \log \det (d_U^{\ast} d_U +\mathfrak{m}_{\mu} (U)) .\)

Here:

Conclusion: there are a strip-analytic two-leg kernel \(\hat{Y}\) and numbers \(e_i \to 0\) such that:

\(\Pi_{\mu}\) is the one-loop effective action of the fine leading part in the background \(U\), in Feynman background gauge: \(d_U d_U^{\ast}\) is the gauge-fixing term and \(d_U^{\ast} d_U\) the ghost operator. The infrared regulator of mass \(\mu = 1 / L\) stands in for the block-spin constraint. \(D\) does not depend on \(L\). The error \(e_i\) decays exponentially in \(n_i\), at an \(L\)-dependent rate.

Status: theorem, for all odd \(L \geqslant 1\): the minimal block factor produced is \(1\). Lean: pouring (proves Fluent, i.e. HorizCalc for all parameters), PoleTest, fineMag, rankAux, grenade, Gauge.slopeConstant.

Proposition III.7.4 (one renormalization step of the leading part). Fix a simple \(G\), kernel parameters \(K\) (penalty \(a > 0\)), a stage \(J\), and constants \(\varepsilon, w, \delta > 0\) and \(D, Z, C, Z' \geqslant 0\) for the fine and coarse beds and clips. There are \(A_{\ast} \geqslant 0\) and \(s > 0\) (output amplitude and width) such that, for every input width \(\tau > 0\) and amplitude \(A \geqslant 0\), there are:

with the following property.

Hypotheses.

Conclusion. There are a coarse path datum \(Y\) and \(\alpha \in \mathbb{R}\) such that:

  1. the rows of \(Y\) are the straight-route rows of \(T\), and \(Y\) is attached to \(T\);

  2. \(Y \in \text{Bounds} (\mathbf{q}; \eta', A', \Lambda')\) whenever \(0 < \eta' \leqslant s\), \(A' \geqslant A_{\ast}\) and \(\Lambda' \geqslant \max (\Lambda_0 (L), \Lambda / 2)\);

  3. \(| \alpha + \kappa \log L| + 1 \leqslant d\);

  4. classical match: for every \(i\), every choice \(t\) of block-kernel data and every \(j \leqslant 4\), the \(j\)-th jets at \(0\) of

\(\displaystyle x \mapsto S_t (x, g_i (x)) \quad \text{and} \quad x \mapsto \text{main}_{\mathbf{q}, Y} (e^x)\)

coincide. Here \(S_t (x, z) = \text{main}_{\mathbf{p}, R} (U_{x, z}) + a \sum_e \left( 1 - d_{\text{rep}}^{- 1} \text{Re} \hspace{0.17em} \text{tr} (V_e \hspace{0.17em} \Pi_t (U_{x, z})_e^{- 1}) \right)\) is the fine leading part plus the block-spin penalty. \(U_{x, z}\) is the fine configuration lifted from \(V = e^x\) through the forest with free variables \(z\), and \(\Pi_t\) is the block predictor;

  1. finite bound: for every \(i\), every \(\Theta \geqslant \Theta_0 (L)\) and every \(c' \leqslant \gamma\), consider the one-loop function

\(\displaystyle \text{th}_{i, t} (x) = \text{loop}_R (U_{x, g_i (x)}) + \tfrac{1}{2} \log \det \text{Hess}_z S_t (x, g_i (x)) - \log \prod_e \textsf{J} (g_i (x)_e) .\)

It is analytic at \(0\), the Hessian determinant is positive near \(0\), and for all coarse lattice momenta \(p, p'\) and real plane waves \(w_{p, \xi} (x) = \text{Re} (e^{ip \cdot x} \xi)\) and \(w_{p', \zeta}\), with every direction-colour component of \(\xi, \zeta\) of modulus at most \(1\),

\(\displaystyle | D^2 \text{th}_{i, t} (0) [w_{p, \xi}, w_{p', \zeta}] - D^2 \left( \alpha \hspace{0.17em} \text{quad}_{\mathbf{q}, Y} + \text{loop}_Y \right) (0) [w_{p, \xi}, w_{p', \zeta}] | \leqslant | \Lambda_{m_i} | \hspace{0.17em} \Theta \hspace{0.17em} e^{- c' m_i} .\)

Here \(\text{quad}_{\mathbf{q}, Y}\) is the signed-permutation average of the Wilson-plaquette and row-square terms in \(\text{main}_{\mathbf{q}, Y}\), with the cubic and quartic atoms omitted. Its Hessian is the coarse quadratic form. The comparison function is also analytic at \(0\). The estimate holds for every \(t\).

On the constants. \(A_{\ast}\) and \(s\) are fixed before \((\tau, A)\). The drift is \(d = d_{\text{cmp}} + D_{\text{slope}} + 1\), where \(d_{\text{cmp}}\) is the comparison constant of §4.8 and \(D_{\text{slope}}\) is the constant of III.7.3. Both may depend on \((\tau, A)\) but are independent of \(L\). The construction gives \(\Lambda_0 (L) = a_0 L^4 + b_0 \log L + c_0\) and \(\Theta_0 (L) = b_1 L^4\), with nonnegative constants depending on the fixed input parameters; both are \(O (L^4)\).

Iteration. Fix positive stage widths and lower bounds for the free quadratic forms, stage amplitude bounds, kernel parameters \(K\) and clip bounds. The constants are then chosen in the following order:

  1. maximal path widths \(\sigma_j\) and minimal path amplitudes \(A_j\);

  2. for each positive sequence \(\tau_j \leqslant \sigma_j\) and sequence \(A'_j \geqslant A_j\): drifts \(d_j \geqslant 0\), rates \(\gamma_j > 0\) and minimal block factors \(u_j\);

  3. for every compatible choice of stage parameters, with direct clip coefficients below their prescribed bounds, drifts at least \(d_j\) and block-factor lower bounds at least \(u_j\), and then every admissible choice of block factors: volume thresholds and positive Hessian lower bounds \(\eta_j\);

  4. for every admissible protection sequence \(\theta\): entry thresholds \(\Delta_j\) for the reference coupling.

For each such choice, take post-step constants \(\Gamma_j \geqslant 1\) and \(0 < c_j \leqslant \gamma_j\) and the Hessian lower bounds \(\eta_j\) just chosen. The conclusion holds for every history satisfying the compatibility, size and coupling thresholds of III.2; every free-field horizon satisfying the strip, Floor, symmetry, top-identity, interpolation and three-form coercivity conditions of III.10; every second horizon with the same beds and the same interpolation and curl-transport kernels; and every local Haar chart ratio. For such data there are kinetic data \((\alpha_{N, m}, P_{N, m}, g_{N, m})\) satisfying all the following conditions:

Status:

Lean: Transition, Scud, Hull, Bubble, FullPaths, Kin.Planned, ClassicalMatch, tested, thermal, FiniteBound, draft.

Proposition III.7.5 (uniform inverse of the completed free operator). Let \(j\) be a stage, \(K\) kernel parameters, \(\sigma, \varepsilon > 0\) and \(\Omega \geqslant 0\). There is \(\eta > 0\), and for every block factor \(L\) a constant \(A_L \geqslant 0\), with the following property.

For every admissible patch of stage \(j\) and block factor \(L\), and every bed \(B\) with \(H \in \text{strip} (\sigma / L, \Omega)\) and \(\text{Floor} (B, \mathbf{c}, n ; \varepsilon, D)\) for all \(n\):

In the rooted coordinates, the joint variables are a fine link field \(v\) and a coarse site field \(h\). Before gauge completion, the quadratic form is

\(\displaystyle \langle dv, H \hspace{0.17em} dv \rangle + a \|- Qv + d_c h\|^2,\)

where \(Q\) is the linear block average and \(d_c\) the coarse gradient. The completed operator adds \(F_{1 / L}^{\textsf{T}} F_{1 / L}\), where \(F_{\mu} = d_{\hspace{-0.17em} 1}^{\ast} P_f +\mathfrak{m}_{\mu} X\), \(P_f\) projects onto fine links, and \(X\) is minus the derivative of the logarithm of the fine rooted frame. Here \(d_{\hspace{-0.17em} 1}\) is \(d_U\) at \(U = 1\). All matrices use physical counting bases. The decay rate \(\eta\) is uniform in \(L\); only the amplitude depends on \(L\).

Status: theorem; used in III.4. Lean: WilsonBulk.uniform_backing, completedM, jointInverse.

Proposition III.7.6 (Faddeev–Popov identity for the one-loop term). Fix an admissible step, a clip \(\mathbf{p}\) and a datum \(R \in \text{Bounds} (\mathbf{p}; \sigma, A, \Lambda)\) for some \(\sigma > 0\) and \(A, \Lambda \in \mathbb{R}\). Let \(g (0) = 0\) be a real-analytic stationary branch for an axial forest, stationary for every choice of block-kernel data, with second variation \(\geqslant \eta \sum_e |v_e |^2\) (\(\eta > 0\)). Let \(\mu > 0\), let \(t\) be a choice of block-kernel data, and let \(\textsf{J}\) be a local Haar chart ratio. Then for coarse fields \(x\) near \(0\):

\(\mathcal{P}_{\mu} (x)\) acts on all fine-link and coarse-frame directions. It is the Hessian in left exponential coordinates of the joint action

\(\displaystyle \mathscr{S}_t (V ; U, h) = \text{main}_{\mathbf{p}, R} (U) + a \sum_e \left( 1 - d_{\text{rep}}^{- 1} \text{Re} \hspace{0.17em} \text{tr} (V_e [\text{gauge}_h \Pi_t (U)]_e^{- 1}) \right),\)

at \(V = e^x\) and the lifted joint stationary configuration, plus \(F_{\mu}^{\textsf{T}} F_{\mu}\). Here \(F_{\mu} = d_U^{\ast} P_f +\mathfrak{m}_{\mu} (U) X\), with the rooted-frame derivative \(X\) as in III.7.5. The joint stationary configuration combines the lifted fine field \(U_{x, g (x)}\) with its coarse frame. \(\mathcal{G}_{\mu} = d_U^{\ast} d_U +\mathfrak{m}_{\mu} (U)\) is the ghost operator at the lifted configuration. The underlying algebraic identity is

\(\displaystyle T^2 \det \mathcal{P}_{\mu} = \det M \hspace{0.17em} (\det \mathcal{G}_{\mu})^2,\)

with \(M\) the axial Hessian and \(T\) the Haar Jacobian of the chart.

Status: theorem; used in the renormalization and remainder comparisons of III.4, III.2 and VII.2. Lean: tribute, bounty, medals.

2. What it uses

3. Strategy

Classical part. At every level the quadratic part of the leading action is fixed by the free-field bed of that level. The cubic and quartic parts are curvature atoms, determined by Taylor agreement through order four with a given gauge-invariant function \(E\). At the top, \(E\) is the Wilson energy. At each step, \(E\) is the tree-level coarse action \(x \mapsto S_t (x, g (x))\). The atom-emission lemma (§4.3, Lemma 4.4) serves both cases. Its mechanism has four parts:

  1. The Taylor coefficients of a translation sum of local analytic functionals have Fourier kernels on \(\mathbb{Z}^4\) that are strip-analytic. Finite-volume coefficients agree with \(| \Lambda_n |\) times these kernels up to \(o (| \Lambda_n |)\) along any sequence of sides; this is how volume-uniformity enters.

  2. Gauge invariance makes the kernels transversal in each leg (linearized Ward identity).

  3. Transversal kernels with at least three legs factor through the curvature. This is the lattice Poincaré lemma, proved by contracting the Koszul complex of the differences \(e^{ip_j} - 1\), one axis at a time, with Hadamard division. A one-leg solve reduces the width from \(\sigma\) to \(\sigma / 4\) once, regardless of the four coordinate divisions within it. Successive leg solves cause further width losses, depending only on the fixed number of legs.

  4. The real, colour-invariant Laurent coefficients of the factored kernels are the coefficients of atoms, and exponential decay is the strip bound.

One-loop part.

  1. In the axial chart the one-loop term is \(\text{loop} + \tfrac{1}{2} \log \det - \log (\text{Haar Jacobian})\). An exact Faddeev–Popov identity turns it into \(\tfrac{1}{2} \log \det\) of a covariantly completed Hessian minus \(\log \det\) of a ghost operator.

  2. Its second-order Taylor coefficient is a two-leg kernel. Its \(F^2\)-coefficient \(\alpha\) is compared, with an \(L\)-independent error, to that of the regulated Feynman-gauge background-field determinant on the fine lattice.

  3. The latter is computed. The lattice integrand is the continuum bubble integrand \(h (k)\) of the vector loop (weight \(\tfrac{1}{2}\)) and the ghost loop (weight \(- 1\)), up to errors \(O (|k|^{- 3} + \mu^2 |k|^{- 6})\); its angular average is \(- \tfrac{11}{3} C_G |k|^{- 4}\). Integration over the shell \(1 / L < |k| \leqslant 1\) gives \(- \kappa \log L\), and the rest is bounded by absolute values.

  4. After subtracting \(\alpha\) times the bed's quadratic kernel, the remaining kernel vanishes to fourth order at zero momentum. It is therefore represented by loop atoms satisfying the two moment conditions, i.e. by irrelevant terms. Inherited fine loop atoms contribute with a factor \(L^{- 2}\).

Constants. The output width and cubic/quartic amplitude of a step are fixed before the input amplitude. The input enters the output only with factors \(L^{- 1}\) or \(L^{- 2}\), which a large \(L\) makes small. The strip widths and inverse bounds for the completed operators are uniform in \(L\) in block-scaled norms. The newly generated loop-atom amplitude is \(O (L^4)\). It is absorbed by \(\text{loopCap}_m\), which grows with \(\hat{\beta}_m\).

Relation to Balaban. The construction uses background-field renormalization with block-spin constraints. Here:

4. Steps

4.1 Fourier kernels of local functionals

Lemma 4.1 (translation sums of local functionals; limits of kernels). (a) Let \(F_k\) be functions of configurations on \(\mathbb{Z}^4\). Suppose \(F_k\) depends only on links in a box of size \(d_k\), has a holomorphic extension with bound \(w_k\) on a polydisc of radius \(r\) around the identity, uniformly after translation and periodic restriction, and \(\sum_k w_k \exp (4 (t + 1)^2 \sigma d_k) \leqslant A\), with \(r, \sigma > 0\) and \(w_k \geqslant 0\). Then for every side \(n\) the translation sum on \(\Lambda_n\) is analytic at \(1\). Its \((t + 1)\)-th Taylor coefficient is recorded exactly by a kernel that is strip-analytic of width \(\sigma\) and size \(A \hspace{0.17em} c (t, r)\), independent of \(n\). For the Wilson energy, kernels of every order exist with width \(1\). (b) Suppose a strip-analytic kernel records \(F_i\) on \(\Lambda_{n_i}\) with error \(| \Lambda_{n_i} | \epsilon_i\), where \(n_i \to \infty\) and \(\epsilon_i \to 0\). If the \(F_i\) are analytic, invariant near \(0\) under global colour rotations and lattice gauge transformations, and their Taylor coefficients of positive orders \(1, \ldots, t\) vanish, then the kernel is transversal in each leg, real and colour invariant. Sketch: (a) Cauchy estimates on the polydisc bound each local Taylor coefficient; its Fourier support has size at most \(4 (t + 1)^2 d_k\), which gives the stated exponential factor. (b) Each property holds exactly for the finite-volume coefficients; divide by \(| \Lambda_{n_i} |\) and pass to the limit along lattice momenta approaching any prescribed real momenta. Strip analyticity gives the required continuity. Lean: gathered, Wilson_stream, withdrew, wedlock. (proof read at top level; Ward identity input statement only)

4.2 The lattice Poincaré lemma in momentum space

Lemma 4.2 (one-leg solve). Let the momentum space have \(d\) scalar coordinates, including four selected coordinates, and let \(\delta\) contract differential forms with \(q_j (p) = e^{ip_j} - 1\) in those four coordinates. Let \(\sigma > 0\), \(A \geqslant 0\), and let \(f\) be a form of degree \(r \geqslant 0\) with the following properties:

Then \(f = \delta B\), where \(B\) has degree \(r + 1\), is strip-analytic of width \(\sigma / 4\) and size \(4 A \hspace{0.17em} \pi_d (\sigma)\) with \(\pi_d (\sigma) = (4 / \sigma) (2 (1 + \sigma / 4) / (\sigma / 4))^d\), vanishes to order \(\max (s - 1, 0)\), and takes values in the same subspaces on real momenta. Sketch: Comb one axis at a time. \(f (p) - f (p) |_{p_j = 0} = q_j (p) D_j (p)\) by Hadamard division (div_var). \(D_j\) has width \(\sigma / 4\) and size \(A \pi_d (\sigma)\) by Cauchy estimates, while the restriction \(f|_{p_j = 0}\) keeps width \(\sigma\). After all four axes, no positive-degree component remains. In degree zero the remainder is \(f\) with all selected momenta set to zero, which vanishes by hypothesis. The factor \(4\) counts the axes; the width loss is a single one. For positive vanishing order, division by \(q_j \sim ip_j\) lowers it by one; order zero remains order zero. Lean: wall_lift.

Lemma 4.3 (transversal kernels are curvature kernels; atoms). For every \(t \geqslant 2\) and \(\sigma > 0\), there are \(\eta \in (0, \sigma]\) and \(K \geqslant 0\) such that every \((t + 1)\)-leg kernel \(f\), strip-analytic of width \(\sigma\) and size \(A \geqslant 0\) and transversal in each leg, admits the factorization

\(\displaystyle f (p) = \sum_b \prod_i \hat{s} (p_i)_{b_i a_i} \hspace{0.17em} g_b (p),\)

for real momenta, where \(\hat{s}\) is the symbol of the lattice curl and \(g\) is a plaquette kernel, strip-analytic of width \(\eta\) and size \(AK\), with \(\eta, K\) depending on \(t, \sigma\). If \(f\) is also real and colour invariant, the Laurent coefficients of \(g\) can be taken real, colour invariant and exponentially decaying. Placed on straight routes, they define atoms. If \(f\) records an analytic functional invariant under signed permutations, the signed-permutation average of their translation sum reproduces that Taylor coefficient up to the recording error. Sketch: Solve the dependent-momentum leg first, then the independent legs by Lemma 4.2, preserving the subspace constraints from the earlier solves. Symmetrize by conjugate reflection to obtain real Laurent coefficients and project the colour tensors onto the invariant subspace. Lean: allleg, strokes, colors_align. (proof read at top level)

4.3 Classical emission and the Wilson jet

Lemma 4.4 (emission of cubic and quartic atoms). Let \(\sigma, \sigma_0 > 0\) and \(A, Z \geqslant 0\). There are \(\tau \in (0, \sigma]\) and \(C \geqslant 0\) with the following property.

Hypotheses:

Conclusion: there are invariant cubic atoms \(X\) and quartic atoms \(V\) with sums of exponential weights at most \(MC\) at width \(\tau\) such that, for every choice \(f\) of loop atoms and every \(i\), \(x \mapsto E_i (x) - \text{main}_{\mathbf{c}, (\rho \text{-rows}, f, X, V)} (e^x)\) vanishes to order \(5\) along rays. Sketch: At order \(3\), subtract the full bare row action, whose quadratic jet agrees with that of \(E\). The residual kernel is transversal (Lemma 4.1(b)) and is matched by cubic atoms (Lemma 4.3). At order \(4\), subtract the same bare row action and the cubic atoms; the remainder is transversal and is matched by quartic atoms. Lean: assembly.

Lemma 4.5 (proof of III.7.2).

  1. Take \(E_i = E_W\) on \(\Lambda_{n_i}\). The kernels of orders \(3\) and \(4\) come from Lemma 4.1(a).

  2. Under \(D = 1\), Floor and invariance, \(H\) agrees with the identity on exact 2-forms. Hence \(E_W\) minus the quadratic term of the bare datum (rows of \(\rho\), no atoms) vanishes to order \(3\) (top_Wilson).

  3. Lemma 4.4 gives \(X, V\), and \(Q\) is formed from the rows of \(\rho\), zero loop atoms, \(X\) and \(V\).

  4. Order-\(5\) vanishing along every ray of the difference of two analytic functions gives equality of all derivatives of order \(\leqslant 4\) (matched).

  5. The weight clauses come from the row prices \(6 Ae^{4 \sigma_0}\), the zero-momentum normalization (Floor on the one-site torus, symbol_match), and the common atom bound \((1 + C_3 + C_4) C\), where \(C_3, C_4\) bound the Wilson Taylor kernels and \(C\) is the emission constant.

Lean: launch, beckon, budding.

Lemma 4.6 (every admissible leading part is Wilson up to irrelevant terms). Fix \(\sigma, \tau > 0\), \(\varepsilon \in \mathbb{R}\) and \(C, Z, A \geqslant 0\). Let \(n_i \to \infty\), \(c_{\text{dir}} \leqslant Z\), and let \(R \in \text{Bounds} (\mathbf{c}; \tau, A, \Lambda)\) be attached to a bed with \(\text{Floor} (B, \mathbf{c}, n ; \varepsilon, C)\) for every \(n\), invariant gauge form on each \(\Lambda_{n_i}\) and \(H \in \text{strip} (\sigma, C)\). Then the Taylor kernels of orders \(2\), \(3\) and \(4\) of \(\text{main}_{\mathbf{c}, R} - E_W\) vanish at zero momentum to orders \(3\), \(2\) and \(1\) respectively. Their strip bounds depend only on \((\sigma, \varepsilon, C, Z, \tau, A)\). In words: all admissible leading parts share the marginal (dimension-four) part of \(E_W\). Sketch: \(\hat{H} (0) = 1\) makes the quadratic difference vanish to order \(3\). Ward identities then force the cubic and quartic differences to vanish to orders \(2\) and \(1\). Apply the same Ward argument to the invariant cubic and quartic atom sums, whose quadratic jets vanish. These statements concern their full nonlinear Taylor coefficients, including transport and curvature expansions. Lean: greyson, veneer, extra_curves, saunter, tideMark. (proof read at top level)

4.4 Two-leg kernels: the \(F^2\)-coefficient and irrelevance

Lemma 4.7 (a two-leg kernel without \(F^2\)-term vanishes to fourth order). Let \(G\) be simple. Let \(f\) be a two-leg kernel that is strip-analytic, transversal, real, colour invariant, symmetric under exchange of the legs, and invariant under signed permutations. If the \(p_q^2\)-coefficient of \(f_{(b, b), (w, w)} (p)\), for \(p\) along an axis \(q \neq w\), vanishes for one colour \(b\), then \(f\) vanishes to order \(4\) at zero momentum. Consequently, for every bed with positive metric strip width, \(\hat{H} (0) = 1\) and signed-permutation-invariant quadratic form, \(f - \alpha_b (f) \hspace{0.17em} \hat{K}_B\) vanishes to order \(4\), where \(\hat{K}_B\) is the bed's quadratic kernel. Sketch:

Lemma 4.8 (loop atoms are irrelevant). Let \(R \in \text{Bounds} (\mathbf{c}; \sigma, C, \Lambda)\). The two-leg kernel of the loop atoms, symmetrized over legs and lattice symmetries:

Sketch: The kernel is \(\hat{s} (- p)^{\textsf{T}} \hat{K} (p) \hat{s} (p)\) with \(\hat{K} (p) = \sum \text{coeff} \hspace{0.17em} e^{ip \cdot y}\). Here \(\hat{s} = O (p)\), and the two moment conditions give \(\hat{K} (0) = 0\) and \(\nabla \hat{K} (0) = 0\). Lean: filing_ready.

4.5 The one-loop term in covariant form

Lemma 4.9 (proof of III.7.6). Let \(H_{\text{joint}}\) be the Hessian of the joint action before gauge completion, and let \(\textsf{R}\) select its axial directions. Three determinant identities combine:

Together they give \(T^2 \det \mathcal{P}_{\mu} = \det M (\det \mathcal{G}_{\mu})^2\). \(\det M > 0\) by the second-variation lower bound and \(\det \mathcal{G}_{\mu} > 0\) since \(\mathcal{G}_{\mu} \geqslant \mathfrak{m}_{\mu} > 0\); take logarithms.

In addition, the one-loop function is analytic at \(0\) (born), and a finite bound proved for one forest transfers to any other (transfer_bound). Lean: tribute. (proof read; determinant factorization inputs statement only)

4.6 Uniform invertibility of the fluctuation operators

Lemma 4.10 (real stability and complex invertibility). Fix a stage \(j\), a penalty \(a > 0\), a lower bound \(\varepsilon > 0\) and a bound \(A_F \geqslant 0\) for the gauge-completing map \(F\). Let

\(\displaystyle \mathcal{C}_F (p) =\mathcal{Q}_B (p) + F (- p)^{\textsf{T}} F (p),\)

where \(\mathcal{Q}_B\) is the block-Bloch symbol of the joint quadratic form in III.7.5. Suppose \(B\) has metric strip bound \((\sigma / L, \Omega)\) and Floor lower bound \(\varepsilon\) on every torus, and \(F\) is holomorphic and periodic in width \(\sigma > 0\), with block-scaled norm at most \(A_F\). On real momenta require

\(\displaystyle F (p) R_{\text{gauge}} (p) =\mathcal{G}_{1 / L} (p), \qquad F (- p)^{\textsf{T}} = F (p)^{\ast},\)

where \(R_{\text{gauge}}\) inserts a fine gauge direction into the joint variables and \(\mathcal{G}_{1 / L}\) is the free regulated ghost symbol.

(a) There is \(y > 0\), depending only on \((G, j, a, A_F, \varepsilon)\), such that \(\|\mathcal{C}_F (p) v\| \geqslant y \|v\|\) in the block-scaled coordinates for every real \(p\), every \(L\) and every patch of stage \(j\).

(b) If \(0 < \sigma \leqslant 1\), \(\Omega \geqslant 0\), the patch is admissible, and \(F\) represents the rooted gauge-completing map on a sequence of tori with coarse sides tending to infinity, then there are \(\eta \in (0, \sigma]\) and \(C \geqslant 0\), depending only on \((G, j, K, \varepsilon, \sigma, \Omega, A_F)\), such that \(\mathcal{C}_F^{- 1}\) is holomorphic in width \(\eta\) and has block-scaled norm at most \(C\), uniformly in \(L\).

Sketch: The real estimate controls the quotient by gauge directions using the quadratic form and block penalty; the displayed identity controls the remaining gauge directions. Strip bounds give a uniform perturbation estimate from real to nearby complex momenta. Choose the width so that this perturbation is smaller than the real lower bound.

Lean: stadium, post_cert. (proof read at top level; real quotient and strip perturbation estimates not expanded)

Lemma 4.11 (all five operators). Fix the numerical parameters of III.7.4. There are \(0 < \eta \leqslant w_0 \leqslant \min (w, \delta, 1 / 128)\) and \(C \geqslant 0\), depending only on the free-field bounds, \(G, K\) and \(J\), with the following property. Under the geometric, fine-bed, fine-datum and interpolation hypotheses of III.7.4, form the actual fluctuation symbols using the free interpolation \(A_0\) and the block predictor. At zero background, the following five operators are invertible in width \(\eta\), with inverse norm at most \(C\) in their respective block-scaled coordinates, uniformly in \(L\) and in the input path width, cubic/quartic amplitude and loop amplitude:

The unaliased free auxiliary operator and the free regulated ghost operator satisfy corresponding inverse bounds. Sketch: Bound the fine auxiliary and ghost inverses directly in Fourier space and transfer the auxiliary bound to its aliased copy. Apply Lemma 4.10 to the joint completion, then use finite-volume compatibility for the completed and contact-removed symbols. Take minima of widths and maxima of constants. Lean: honored, GrantF, completion_choice. (proof read at top level)

Lemma 4.12 (proof of III.7.5).

  1. Choose \((\eta, C)\) by Lemma 4.10(b) for the regulator family of the completed operator.

  2. For each \(L\), a constant \(W (L)\) converts block-scaled norms into the plain strip norm.

  3. The bound on the inverse gives the strip bound \((\eta / 2, \text{cost} (\eta, CW (L)))\).

  4. The inverse Fourier coefficients are real by finite-volume compatibility. On every torus their Fourier multiplier is the inverse symbol, which gives the matrix identity.

Lean: uniform_backing.

4.7 The slope of the regulated determinant

Lemma 4.13 (the continuum bubble). Assume \(G\) is simple. Take an external field of colour \(b\), polarization \(e_1\) and momentum \(t \hspace{0.17em} e_0\), and loop momentum \(k \neq 0\).

Writing \(c_{\text{gh}} (t)\) for the ghost bubble and \(c_{\text{vec}} (t)\) for the vector bubble,

\(\displaystyle \partial_t^2 c_{\text{gh}} (0) - \tfrac{1}{2} \hspace{0.17em} \partial_t^2 c_{\text{vec}} (0) = 2 h (k), \qquad h (k) = 4 C_G \left( \frac{k_1^2}{|k|^6} - \frac{4 k_1^2 k_0^2}{|k|^8} - \frac{1}{|k|^4} \right) .\)

The terms in \(k_1^2\) come from the orbital coupling: one adjoint scalar's worth, \(\tfrac{1}{2} \cdot 4\) vector components minus one ghost. The term \(- |k|^{- 4}\) comes from the spin coupling of the vector field. Lean: haloRate, poleVec.

Lemma 4.14 (the shell integral). For \(0 < \mu \leqslant 1\),

\(\displaystyle \frac{1}{(2 \pi)^4} \int_{\mu < |k| \leqslant 1} h (k) \hspace{0.17em} d^4 k = - \kappa \log \mu^{- 1} .\)

Sketch: By rotation invariance \(\int k_1^2 |k|^{- 6} = \tfrac{1}{4} \int |k|^{- 4}\), and \(\int k_1^2 k_0^2 |k|^{- 8} = \tfrac{1}{24} \int |k|^{- 4}\), the latter from averaging over \(e_i \pm e_j\). Also \(\int_{\mu < |k| \leqslant 1} |k|^{- 4} = 2 \pi^2 \log \mu^{- 1}\). So \(h\) averages to \(4 C_G (\tfrac{1}{4} - \tfrac{1}{6} - 1) |k|^{- 4} = - \tfrac{11}{3} C_G |k|^{- 4}\): orbital \(+ \tfrac{1}{3} C_G\), spin \(- 4 C_G\). Finally \(\tfrac{11}{3} C_G \cdot 2 \pi^2 / (2 \pi)^4 = \kappa\). Lean: poleSpin, joy_two, joy_heights, polar_m.

Lemma 4.15 (the lattice integrand is close to \(h\)). Under the hypotheses of III.7.3, let \(\Phi_{\mu}\) be the unintegrated second-derivative kernel of \(\Pi_{\mu}\), built from the vertex kernels of Lemma 4.6:

\(\displaystyle \Phi_{\mu} = \tfrac{1}{2} \hspace{0.17em} (\text{vector film}) - (\text{ghost film}) .\)

Let \(k_b (k) = \tfrac{1}{2} \partial_s^2 \Phi_{\mu, (1, b), (1, b)} (s \hspace{0.17em} e_0 ; k) |_{s = 0}\) be its integrand at loop momentum \(k\). The recorded quadratic Taylor kernel is

\(\displaystyle \hat{Y}_{\mu} = \frac{1}{2 (2 \pi)^4} \int_{[- \pi, \pi]^4} \Phi_{\mu} (\hspace{0.17em} \cdot \hspace{0.17em} ; k) \hspace{0.17em} d^4 k,\)

and \(\alpha_b (\hat{Y}_{\mu}) = (2 \pi)^{- 4} \int_{[- \pi, \pi]^4} \text{Re} \hspace{0.17em} k_b (k) \hspace{0.17em} d^4 k\). There are \(a, b' \geqslant 0\), uniform over all data of III.7.3 and all \(\mu \in (0, 1]\), such that:

\(\displaystyle |k_b (k) | \leqslant \frac{a}{(|k| + \mu)^4} \text{on the Brillouin cube}, \qquad |k_b (k) - h (k) | \leqslant b' (|k|^{- 3} + \mu^2 |k|^{- 6}) \text{for } |k| \geqslant \mu .\)

Sketch: Each film is (tadpole) \(-\) (bubble). The tadpoles are \(O (|k|^{- 2} + \mu^2 |k|^{- 6})\). The lattice bubbles differ from the continuum ones by \(O (|k|^{- 3} + \mu^2 |k|^{- 6})\), using the softness of Lemma 4.6. Lemma 4.13 identifies the continuum combination. Lean: astronomy, Rally.coin, Rally.shell, polar_clock, amena. (proof read at top level; analytic tadpole and bubble estimates not expanded)

Lemma 4.16 (from the bounds to the slope). For \(0 < \mu \leqslant 1\), under the strip regularity and two bounds of Lemma 4.15 with \(a, b' \geqslant 0\),

\(\displaystyle | \alpha_b + \kappa \log \mu^{- 1} | \leqslant (2 \pi)^{- 4} \left( a \tfrac{\pi^2}{2} + a (2 \pi)^4 + 3 \pi^2 b' \right) .\)

Sketch: Split the cube into three regions.

Lean: absoluteGeode.

Lemma 4.17 (proof of III.7.3).

  1. Lemma 4.6 and Lemma 4.1 give the vertex kernels of \(\text{main}_{\mathbf{c}, R} = E_W + (\text{main} - E_W)\).

  2. The film \(\Phi_{1 / L}\) represents \(\Pi_{1 / L}\) (gemF). For fixed \(L\) it is strip-analytic, since the vector and ghost symbols are invertible at real momenta, by Floor and the regulator, and hence in a strip (HouseWord.done).

  3. One half of the normalized loop-momentum integral of the second-derivative film is the Taylor kernel \(\hat{Y}\). Comparing the torus Riemann sum with the integral gives an error \(| \Lambda_{n_i} | K_L e^{- z_L n_i / 4}\) (profit).

  4. Lemmas 4.15–4.16 with \(\mu = 1 / L\) give (b), with \(D = (2 \pi)^{- 4} (\ldots)\) independent of \(L\).

Lean: pouring.

4.8 From the regulated determinant to the blocked determinant

Lemma 4.18 (comparison of \(F^2\)-coefficients). Assume \(G\) is simple and the hypotheses of III.7.4, together with the interpolation, vertex and inverse bounds of Lemma 4.11. Consider the two-leg films of:

Then:

Sketch:

Lean: quoting, medalCue, coreCue, rankCue. (proof read at top level; contact identities not expanded)

Lemma 4.19 (aliasing preserves \(\alpha\)). Under the hypotheses of Lemma 4.18, the \(F^2\)-coefficient of the auxiliary film on the coarse lattice equals \(\alpha_b (\hat{Y})\), where \(\hat{Y}\) is the kernel of III.7.3 for \(\Pi_{1 / L}\) on the fine tori \(\Lambda_{Lm_i}\). Sketch: The coarse symbol at momentum \(p\) is the sum over the \(L^4\) aliases of the fine symbol. The nonzero aliases vanish to order \(4\). The zero alias agrees through order two with \(L^4 L^{- 2}\) times the fine kernel at \(p / L\). The additional \(L^{- 2}\) from momentum rescaling gives \(L^4 L^{- 2} L^{- 2} = 1\), so the \(F^2\) coefficient is unchanged. Lean: pole_reduced, carry_plan, fine_alpha. (proof read at top level; high-alias and zero-alias estimates statement only)

4.9 The step and the iteration

Lemma 4.20 (construction of the coarse datum). Assume the geometric, bed, Haar-chart and stationary-branch hypotheses of III.7.4. Fix a common Fourier width \(\sigma > 0\), coarse-bed width \(w > 0\), coarse metric and root bound \(M \geqslant 0\), and a bound on the coarse direct clip coefficient. Require \(H_T, \rho_T \in \text{strip} (w, M)\). There are \(s \in (0, \min (\sigma, w)]\) and \(K, D \geqslant 0\), independent of the following kernel amplitudes, with this property.

Suppose the tree-level action has exact Taylor kernels of orders \(3\) and \(4\), of width \(\sigma\) and size \(X \geqslant 1\). Suppose a two-leg kernel \(\hat{A}\) of width \(\sigma\) and size \(\Phi\) records the determinant part with errors \(| \Lambda_{m_i} | e_i\), where \(e_i \geqslant 0\) and \(e_i \to 0\). Suppose a two-leg kernel \(\hat{L}\) of the same width and size \(R_0\) records the inherited loop functional exactly and vanishes to order \(4\) at zero momentum. Set \(\alpha = \alpha_b (\hat{A})\) for one fixed colour \(b\).

There is a coarse datum \(Y\), with the straight-route rows of \(T\) and attached to \(T\), whose classical jets agree through order four. It satisfies the weight clauses at every \(0 < s' \leqslant s\), every row/cubic/quartic bound

\(\displaystyle A' \geqslant \max \{6 Me^{4 w}, XK\},\)

and every loop bound

\(\displaystyle \Lambda' \geqslant (\Phi + | \alpha |C_T + R_0) D,\)

where \(C_T\) is the uniform strip bound for the quadratic kernel of \(T\) at width \(\min (\sigma, w)\).

For every \(i, \Theta, c', \Omega\), the finite bound of III.7.4 holds with right-hand side \(| \Lambda_{m_i} | \Theta \Omega e^{- c' m_i}\) whenever \(B_G e_i \leqslant \Theta \Omega e^{- c' m_i}\), where \(B_G = 2 (4 \dim \mathfrak{g})^2\).

Sketch:

Lean: color_pipe. (proof read at top level)

Lemma 4.21 (proof of III.7.4).

\(\displaystyle M = 1 + \tfrac{a + 1}{6} + \tfrac{(c + 1) + 3 (a + 1)^2 E}{24} .\)

Here \(E\) bounds the completed free inverse. The term \(3 (a + 1)^2 E\) bounds the three pairings of two cubic vertices joined by one propagator.

Lean: transition_cap. (proof read at top level)

Lemma 4.22 (loop caps and the iteration).

Lean: clocks, geared.

5. Where the difficulty lies

6. Notation

A determinant film records the second derivative before loop-momentum integration. Its quadratic Taylor kernel is one half of its normalized loop-momentum integral.

Written proof Lean Meaning
\(R = (\ell, \mathcal{A}^{(2)}, \mathcal{A}^{(3)}, \mathcal{A}^{(4)})\) RG.PathData path datum (rows, loop/cubic/quartic atoms)
leg, atom RG.Slot, RG.Atom transported curvature; curvature monomial
\(\text{Bounds} (\mathbf{c}; \sigma, C, \Lambda)\) PathData.Bounds exponential summability, invariance and moment clauses
\(\text{main}_{\mathbf{c}, R}\), \(\text{loop}_R\) mainTerm, loop leading part; loop functional
clip \(\mathbf{c}\), \(c_{\text{dir}}\) RG.Clip clipping parameters; direct
bed \(B\), Floor, attached Bed, Floor, Attached free quadratic-form data of a level
straight-route rows of \(\rho\) rowPath, emitBase rows realizing the bed roots on straight lattice routes
strip-analytic \((\sigma, K)\) Firm bounded holomorphic periodic kernel
vanishing to order \(s\) along rays Rays raywise \(O (t^s)\) at \(t = 0\)
\((t + 1)\)-leg kernel; records Picture, Recorded Fourier kernel of a Taylor coefficient
\(\alpha_b (\hat{K})\) alphaPay \(F^2\)-coefficient, normalized by the bed
\(\Pi_{\mu}\) PoleTest, fineMag regulated background-field one-loop action
\(\mathcal{G}_{\mu}\), \(\mathcal{P}_{\mu}\) grenade, medals ghost operator; completed Hessian
\(\mu = 1 / L\) inch regulator mass
\(h (k)\); \(k_b (k)\) poleVec; keelRay continuum bubble integrand; lattice integrand
bounds of Lemma 4.15 Waves1, Waves2 global infrared bound and shell comparison bound
\((2 \pi)^{- 4}\) cartWeight normalization of the momentum integral
\(S_t (x, z)\); \(\text{th}\) tested; thermal fine action plus penalty; one-loop function
classical match; finite bound ClassicalMatch; FiniteBound items 4 and 5 of III.7.4
slope statement HorizCalc, Fluent III.7.3 for given / all parameters
step property; step record Transition; Bubble III.7.4
top datum BranchRow III.7.2 at level \(N\)
kinetic data, planned conditions Kin, Kin.Planned, FullPaths output to III.9
invertibility bounds GrantF, Freezer strip invertibility with inverse bound
medal, core, auxiliary films medalCue, coreCue, rankCue §4.8
\(\hat{\beta}_m\), \(\text{loopCap}_m\) hatBeta, loopCap reference coupling; loop-amplitude cap
\(\kappa\) Gauge.slopeConstant \(11 C_G / (24 \pi^2)\)
\(D^r F (0) / r!\) noteAt complexified Taylor coefficient
analytic gauge and lattice symmetry Pointwise hypotheses of atom emission
\(\pi_d (\sigma)\) priceDivision one-coordinate division constant in \(d\) momentum coordinates
joint free form; gauge completion frontR, postroom, cannonade fine-link and coarse-frame operators

7. What remains unexplained

  1. Analytic estimates below the stated lemmas. The real quotient estimate and complex-strip perturbation estimate used in Lemma 4.10, and the analytic tadpole and bubble bounds used in Lemma 4.15, have not been checked through every subsidiary proof. Their statements support the estimates given here. The determinant-factorization and Haar-density lemmas of Lemma 4.9 remain inputs at statement level. The contact identities used in Lemma 4.18 and the high-alias and zero-alias estimates used in Lemma 4.19 likewise remain inputs at statement level.

  2. Choice of regulator power. The ninth power is part of the explicit definition, and the cited estimates use it. A minimal power, or a conceptual reason for choosing exactly nine, has not been established.

  3. Supply of the global hypotheses. III.4 supplies the free-field horizons and history conditions. The free/stationary connection explicitly constructs the two horizons with identical beds and interpolation kernels before applying the iteration. The discharge of these conditions in every downstream application has not been checked.

  4. Limits of path data. The projection results support the compactness argument in III.13: exponential weights give Hilbert-space parameters for curvature coefficients (fitting); coefficientwise convergence with uniform row bounds gives convergence of row-action jets (insideLine); compatible source jets and a common positive Hessian lower bound give continuity of stationary, tree-level and one-loop jets (interview). The complete compactness construction and passage through all its quantifiers belong to III.13 and remain outside this unit.

← III.6 The stationary branch: background field of the effective action (notes)Part III·Unit outlineIII.8 Connected activities and backward dependence cones (notes) →